20VC with Harry Stebbings

⁠Why OpenAI and Anthropic Won't Win the App Layer | Glean Founder

4051 summary words 18 min summary Watch video

Start with the signal

18 min read

Summary

At-a-Glance

  • Verdict: Watch fully
  • Core thesis: Enterprise AI will be won by best-of-breed context platforms and open-source cost arbitrage, not by bundling or frontier labs' shallow vertical moves—land grab is now, but sustainable cost structure and discipline still matter long-term.
  • Why it matters: Arvin Jain (Glean founder, Rubrik IPO) offers a rare enterprise view: frontiers are commoditizing, customers fear dependency, open models are viable for 90%+ of workloads, and composable roles/consumption pricing break bundling—but warns AI ROI is still unclear and overabundant capital creates unsustainable startup structures.
  • Best use: Watch to understand enterprise AI buying behavior, open vs. closed model economics, bundling vs. consumption dynamics, and the real barriers to AI productivity (not coding speed but shipping speed, context assembly, and cost control).

Executive Summary

Arvin Jain, founder of Glean (enterprise AI platform) and co-founder of Rubrik (successful IPO), argues that while frontier model providers (OpenAI, Anthropic) and bundlers (Microsoft Copilot) compete aggressively, they won't dominate the application layer. Enterprises fear operational dependency on labs—not just data privacy but loss of institutional learning accumulated in agents. They're actively seeking open-source models (now viable for 90%+ of workloads post-Llama 3.2/DeepSeek) to control cost and destiny. Glean positions itself as a best-of-breed alternative: a 'superset of ChatGPT/Claude/Gemini' connected to enterprise context, delivering search + agents for knowledge work.

Jain believes Microsoft's bundling advantage erodes under consumption pricing (pay-per-task, not per-seat), allowing six tools to coexist if users choose freely. He downplays frontier labs' vertical packs (legal, design, health) as 'shallow' net-new expansion, not displacement. However, he admits Anthropic/Claude Desktop competes directly in Q&A (Glean's largest use case). First-mover advantage helps brand but isn't decisive. Open-source commoditization is real—Glean routes workloads to cheaper models when quality suffices, and predicts majority of enterprise AI runs on open models in three years.

On ROI: pockets of clear value exist (customer support, coding speed), but shipping speed hasn't increased despite 100% AI-generated code at Glean. Bottleneck shifted to code review; AI's true productivity gain requires investing 'around' it—pre-assembling context so agents don't brute-force search and burn tokens. Jain warns overabundant VC capital creates unsustainable structures (e.g., paying $500k/engineer at seed stage). He's paranoid/disciplined but feels pressure to shift toward land-grab mode. Sees composite generalist roles (eng+PM+design; AE+SE) replacing specialists, analyst/sourcer roles disappearing, but rejects the idea of shrinking teams—argues 10x better products require more people if competitors also have AI.

On sovereignty: demand for sovereign models has fallen (enterprises comfortable with US labs under contract), but Jain acknowledges rising geopolitical tension post-Trump export bans. US open-source lags China; he expects Nvidia/others to fund US efforts but doubts regulatory capture will succeed. He's skeptical that Sam Altman's 5% Trump stake buys protectionism—believes US system favors innovation over restriction. Warns founders not to fear frontier cannibalization: labs are 'huge asset, not competition' for app-layer companies. Recruiting easier than SaaS peak (tech layoffs freed talent) except for ML roles (pay scales exploded). Says being CEO is 'not sexy,' requires perpetual mission-driven unhappiness, and believes founders are better when already rich (rational vs. economically impatient).

Key Takeaways

  • Claim: Enterprises fear operational dependency on frontier labs—not just data leakage but loss of institutional learning that accumulates inside agents doing their work. | Evidence: Jain: 'If you don't own the learning [the agent] gains over the years, you're fully dependent on AI companies to get your work done. It's more than technology dependence—it's real operational dependence.' | Caveat: This assumes agents will be long-lived and domain-specific enough to accrue proprietary learning; today most enterprise AI is still shallow Q&A. | Implication: App-layer vendors (Glean, Palantir) can win by offering enterprises control/ownership of agent training and context graphs, not just API access. | Timestamp: 03:20
  • Claim: Open-source models can handle 90%+ of enterprise workloads; Glean's team feels comfortable routing 'majority' to Llama 3.2 as of one month ago. | Evidence: Jain: 'Llama 3.2 is the first time our own team feels comfortable running majority of workloads on that model… We tell customers we pick the right model; if you're okay with open source, we'll use it when it generates high-quality answers.' | Caveat: Adoption is <3 months old; enterprises haven't decided if they'll accept Chinese models (DeepSeek) due to paranoia about 'magic backdoors' despite local inference. | Implication: If Chinese open models dominate (6 of top 7 on OpenRouter), US labs face commoditization unless they lock in via bundling/regulation or open models prove insufficient for frontier tasks. | Timestamp: 07:45
  • Claim: Consumption-based pricing breaks bundling advantage: enterprises can approve six tools and pay only where work happens, unlike seat-based SaaS. | Evidence: Jain: 'Once you move towards consumption, there's no inherent bundling advantage—a business can get six tools, let users choose, and pay per unit of work.' | Caveat: Vendor management overhead still exists (approvals, compliance); Microsoft can still use 'free' bundling psychology even if metered. Jain admits Microsoft is Glean's most significant competitor and bundling 'works.' | Implication: Consumption pricing favors best-of-breed if enterprises are willing to manage multiple vendors; but Microsoft's installed base + compliance inertia remains formidable. | Timestamp: 18:30
  • Claim: AI coding is 100% at Glean but shipping speed hasn't increased—'bottleneck shifted from writer to reviewer'; companies eliminating code review risk unmaintainable codebases. | Evidence: Jain: 'Almost all code is written with AI. Nobody's writing initial code by hand. But we enforce human reviews… Many companies let AI code submit directly to repos. If you have stringent review, it removes the point of fast code development.' | Caveat: Glean is 'more conservative than most'; some companies may accept tech debt to ship faster. Jain admits it's hard to measure productivity gains amid team growth. | Implication: AI's ROI in engineering is fuzzy; speed gains in coding don't translate to shipping unless review/refactoring/context assembly are also AI-assisted. Operators should measure shipping velocity, not lines of code. | Timestamp: 23:10
  • Claim: Glean spent $1M/month on a triage agent that replaced 95% of a 15-person on-call team's work—but cost was 'questionable' vs. human salaries. | Evidence: Jain: 'We built an agent for production triage—handles 95% of issues automatically. But spending a million a month was more than the cost of humans.' | Caveat: This was an extreme case; Jain later says Glean runs 1,000+ people and expects 5,000 in five years, suggesting net headcount growth despite agents. Cost may fall with open models. | Implication: Token costs at frontier pricing can exceed human salaries for high-throughput workloads; open-source arbitrage is critical for sustainable AI ops. CFOs should model token spend vs. headcount explicitly. | Timestamp: 30:50
  • Claim: Frontier labs' vertical packs (legal, design, health) are 'shallow' and net-new market expansion, not displacement of incumbents like Figma or Harvey. | Evidence: Jain: 'I don't know people moving their workload entirely from Figma to Anthropic. It's net new—non-designers using Claude Design, designers still use Figma. AI makes things simpler for non-experts.' | Caveat: Anthropic/Claude Desktop directly competes with Glean in Q&A (largest AI use case globally). Jain admits 'we face that competition every day.' | Implication: Labs may not kill horizontal SaaS but will commoditize low-complexity workflows; app-layer winners must offer deep context integration, workflow orchestration, and cost control beyond model access. | Timestamp: 11:20
  • Claim: Alex Karp (Palantir) is right that AI ROI is unclear for most enterprises; 2026 H2/2027 is when CFOs demand proof. | Evidence: Jain: 'Pockets of value exist—customer support is measurable (10 → 12 cases/day). But most AI spend is on coding, and shipping speed hasn't increased… Engineering productivity is the fuzziest job; hard to tease apart gains from team growth.' | Caveat: Some verticals (support, coding assist) show clear metrics; Jain says companies are 'still learning' effective use and willing to 'pay the cost of being slow' (e.g., mandatory code review). | Implication: Operators should isolate measurable workflows (support tickets, sales calls, knowledge search) and avoid assuming productivity gains in fuzzy domains (product shipping, strategy) without rigorous A/B testing. | Timestamp: 25:40
  • Claim: Composite generalist roles (eng+PM+designer, AE+SE+SA) will replace specialists; analyst and sourcer roles will disappear. | Evidence: Jain: 'We'll see more generalization—away from specialization. Business owners will directly get answers; data analysts who just build dashboards will go away. In recruiting, sourcer role gets consumed into full-cycle.' | Caveat: Composite roles reduce headcount only if workload stays constant; Jain argues 10x better products require 10x more work, so teams grow. This is contradictory unless you believe demand is infinite. | Implication: Hiring should shift toward T-shaped generalists who can leverage AI across functions; specialist roles (BI analyst, technical sourcer) are at risk unless they evolve into strategy/judgment roles. | Timestamp: 49:30
  • Claim: Overabundant VC capital creates unsustainable startup structures—e.g., paying $500k/engineer at seed stage when Google doesn't. | Evidence: Jain: 'Startups raise seed and pay half a million to an engineer. Founder okay, investors okay—but surely not sustainable to win. Google knows they don't need to buy talent like that.' | Caveat: Top ML talent is scarce; startups may need to overpay to compete with labs (OpenAI, Anthropic hiring aggressively). Jain admits Glean has to pay competitively. | Implication: Seed-stage founders should resist VC pressure to overpay for talent; burn discipline matters even in land-grab mode. Investors should model sustainable unit economics, not just growth. | Timestamp: 52:10
  • Claim: Demand for sovereign models has fallen—enterprises comfortable with US labs under contract; only geopolitical concern is Chinese models. | Evidence: Jain: 'Desire was stronger a year back. Nations figured out they can't build one. People believe model companies won't train on enterprise data if you've signed the right contract. Question is: are they okay with Chinese models?' | Caveat: Trump admin banned Anthropic's latest models; Jain acknowledges this shifts European sentiment ('we cannot rely on US individual who could ban access'). US open-source lags China (6 of top 7 models). | Implication: Sovereignty is less about building national models, more about ensuring access amid export controls. Europe/others may accept Chinese models if US restricts access; regulatory capture (Sam/Trump 5%) is a tail risk. | Timestamp: 55:20

Detailed Brief

Enterprise AI Landscape: Bundling vs. Best-of-Breed vs. Labs

  • Claims: Microsoft Copilot is Glean's most significant competitor; bundling 'works' but is weakened by consumption pricing.; Frontier labs (OpenAI, Anthropic) are 'huge asset, not competition' for app-layer companies; their vertical packs are shallow.; Enterprises fear operational dependency on labs (losing institutional learning in agents), not just data privacy.; First-mover advantage helps brand but isn't decisive; Glean benefits from being 'first enterprise AI company.'
  • Evidence: Jain: 'Microsoft is formidable. We hear: we're a Microsoft customer, we have Copilot, doesn't make sense to consider [Glean].'; Jain: 'Anthropic's vertical packs are shallow. People not moving from Figma to Claude—it's net new (non-designers using Claude Design).'; Jain: 'If you don't own the learning [agents] gain, you're fully dependent on AI companies. This is more than tech dependence—it's operational.'; Jain: 'We get credit for being first to bring RAG, semantic search to enterprise. Gives us brand and right to compete vs. giants.'
  • Caveats: Bundling psychology + vendor management overhead still favor Microsoft; consumption doesn't eliminate compliance/approval friction.; Claude Desktop competes directly in Q&A (Glean's largest use case); Jain admits 'we face that competition every day.'; First-mover isn't requirement or savior; execution and context depth matter more.
  • Implications: App-layer vendors must differentiate on context integration, workflow orchestration, and cost control—not just model access.; Consumption pricing creates opportunity for multi-vendor environments if enterprises accept management overhead.; Labs may expand TAM (non-experts using AI) rather than displace incumbents; watch for usage overlap vs. cannibalization.

Open-Source Commoditization & Model Economics

  • Claims: 90%+ of enterprise workloads can run on open models (Llama 3.2, DeepSeek); Glean routes to cheapest viable model.; Inferencing costs are 'absurdly expensive'—Glean spent $1M/month on triage agent, more than human salaries for 15-person team.; Model pricing paradoxically increased in last 6–9 months (vs. historical decline); Jain expects order-of-magnitude cost drops ahead.; Enterprises driven to open source by cost, not data sovereignty (comfortable with labs under contract); question is Chinese models.
  • Evidence: Jain: 'Llama 3.2 is first time our team feels comfortable running majority of workloads on it… majority of enterprise AI on open models in 3 years.'; Jain on triage agent: '$1M/month token spend—actually more than cost of humans. It's questionable if it's more efficient.'; Jain: 'We saw something bizarre: every model increased per-token price [recently]. Everybody thought it'd keep falling.'; Jain: 'Fear of labs training on data is no longer there if you sign right contract. Cost drive is real now.'
  • Caveats: Open-source viability is <3 months old; enterprises haven't decided on Chinese models (paranoia about 'magic backdoors').; Pricing increase may reverse as labs prove unit economics pre-IPO; trend unclear.; High token costs at frontier pricing don't negate value if quality/speed justifies—Jain says dev tools 'dramatically underpriced.'
  • Implications: CFOs should model token spend vs. headcount explicitly; open-source arbitrage critical for sustainable AI ops.; If Chinese models dominate open leaderboard and US restricts exports, geopolitical risk to enterprise AI supply chain increases.; Model layer commoditizing fast—value accrues to context/orchestration layers (Glean, Palantir) and vertical workflow integrations.; Startups betting on frontier-only pricing risk margin compression; multi-model routing + cost control is competitive moat.

AI Productivity & ROI Reality Check

  • Claims: 100% of Glean code is AI-generated, but shipping speed hasn't increased—bottleneck is code review, not writing.; Clear ROI exists in narrow domains (customer support: 10 → 12 cases/day); fuzzy in engineering/product shipping.; Most enterprise AI spend is on coding; most usage is Q&A/summarization (everyone uses it); advanced use cases are 5% of employees.; Power-law adoption: some users spend $10–15k/month in tokens, others $20; no exec mandate needed at Glean (native AI company).
  • Evidence: Jain: 'Almost all code written with AI. Nobody writes initial code by hand… But we enforce human reviews. Many companies submit AI code directly to repos.'; Jain: 'Shipping speed hard to measure—result of larger, more tenured team. Customer support is easy to measure: agents resolve 10 → 12 cases.'; Jain: '#1 application of AI globally is information seeking, Q&A. Everybody's doing that. Advanced use cases limited to 5% of employee base.'; Jain: 'Power law—some spend $15k/month, others $20. We didn't budget tokens; let people figure out what they can do.'
  • Caveats: Glean is 'more conservative than most'—enforces review; some companies accept tech debt to ship faster.; Engineering productivity is 'fuzziest job'; hard to tease apart AI gains from team growth/maturity.; Early adopters (5%) drive disproportionate spend; unclear if broader workforce will reach advanced use cases.
  • Implications: Operators should measure shipping velocity (features/releases) not lines of code; AI's coding speed gain doesn't auto-translate to product velocity.; AI ROI is clearest in repetitive, measurable workflows (support, triage, transcription); avoid assuming gains in creative/strategic work without A/B tests.; Token budgeting is immature—most companies didn't plan; 2026/2027 will force CFOs to justify spend vs. outcomes.; Power-law usage suggests targeted enablement (identify high-value use cases, train 5% champions) > blanket rollout.

Team Structure, Talent, & Capital Discipline

  • Claims: Glean has 1,000+ people today, expects 5,000 in five years—net growth despite AI (composite roles + 10x more work to compete).; Composite generalist roles (eng+PM+design, AE+SE) will replace specialists; analyst/sourcer roles disappear.; Recruiting easier than SaaS peak (tech layoffs freed talent) except ML roles (pay exploded); startups paying $500k/engineer unsustainable.; Jain is 'too disciplined'—feels pressure to shift toward land-grab spending but believes discipline builds sustainable business.
  • Evidence: Jain: 'Over 1,000 people now. Hopefully 5,000 in five years… To make same revenue, you have to do 10x the work—produce 10x better product.'; Jain: 'Generalization away from specialization. Data analysts who just build dashboards go away. Sourcer role consumed into full-cycle recruiting.'; Jain: 'Recruiting was getting easier vs. SaaS peak—Meta, others laying off. But top ML talent sought after; pay scales completely changed.'; Jain on seed startups paying $500k: 'Surely not sustainable path to win. Google knows they don't need to buy talent like that.'
  • Caveats: Glean's growth is atypical—Jain admits 'every CEO I sit with is shrinking teams.' His logic assumes infinite demand (10x better product needed).; Composite roles reduce headcount only if workload constant; Jain's argument contradicts unless you believe competition forces perpetual scope expansion.; Top ML talent is scarce; startups may have no choice but to overpay to compete with labs.
  • Implications: Enterprise SaaS may grow headcount (against AI shrinkage trend) if they expand TAM/product scope faster than productivity gains; consumer/low-margin businesses will shrink.; Hiring should prioritize T-shaped generalists who can leverage AI across functions; specialist roles at risk unless they evolve to strategy/judgment.; Seed-stage founders: resist VC pressure to overpay talent; model sustainable burn even in land-grab. Investors: scrutinize unit economics, not just growth.; Glean's 5x growth target signals land-grab mode—watch for burn rate, dilution, and whether they maintain gross margins amid open-source pricing pressure.

Geopolitics, Sovereignty, & Regulatory Capture Risk

  • Claims: Demand for sovereign models has fallen—enterprises comfortable with US labs under contract; question is Chinese models.; Trump export ban on Anthropic's latest models shifts European sentiment ('can't rely on US individual who could ban access').; US open-source lags China (6 of top 7 OpenRouter models); Nvidia/others investing in US open models but results TBD.; Jain doubts Sam Altman's 5% Trump stake buys regulatory capture—believes US system favors innovation over restriction.
  • Evidence: Jain: 'Desire [for sovereign models] was stronger a year back. Nations figured out they can't build one… comfortable with US labs if right contract.'; Jain: 'Trump banning Anthropic models a month ago—Europeans say we cannot rely on US individual who could ban our access to intelligence.'; Jain: 'Only country producing models outside US is China. A little bit in France (Mistral). US open-source weak—requires upfront investment, not open-friendly.'; Jain on Sam/Trump 5%: 'I doubt that's gonna happen. I believe more in US system… right now you don't need to curb open source—it's too far behind.'
  • Caveats: Export bans are recent (1 month); too early to see sovereign model response. Mistral exists but not dominant.; Jain may underestimate political risk—Sam's 5% could be quid pro quo for regulatory moat (taxes on Chinese models, export controls).; US system historically favors innovation, but AI is strategic (cf. chips, TikTok); national security overrides market logic.
  • Implications: Europe/others may accept Chinese models if US restricts access—regulatory capture could backfire by ceding market to DeepSeek/China.; Investors in US labs: watch for export control risks (customer lock-out) and political favoritism (regulatory moat vs. anti-trust).; Open-source is geopolitical—US needs domestic Llama alternative or risk China becoming default enterprise inference provider.; Operators: build multi-model routing (US labs + open) to hedge geopolitical supply chain risk.

Notable Concepts & Terms

  • RAG (Retrieval-Augmented Generation): Glean pioneered enterprise RAG—retrieving company context from 100s of systems, augmenting LLM prompts so agents don't 'brute force' search. Jain says this is 90% of AI ROI work: pre-assembling context to reduce token burn and latency.
  • MCP (Model Context Protocol): Anthropic's protocol for connecting Claude to enterprise systems. Jain says customers ask 'what's different from MCP?'—implies MCP is rudimentary vs. Glean's deep semantic context graphs. Competitive threat but Glean differentiates on depth.
  • Consumption-based pricing: Pay-per-task vs. seat licenses. Jain argues this breaks bundling (Microsoft can't lock customers in if they pay only where work happens). Key shift in SaaS → AI transition; favors multi-vendor, best-of-breed stacks.
  • Composite roles: Jain's term for generalists who collapse 3–4 specialties (e.g., eng+PM+designer, AE+SE+SA). Predicts AI enables this; reduces headcount if workload constant, but Jain believes 10x better products require more work → net team growth.
  • Llama 3.2 / DeepSeek viability threshold: Jain says Llama 3.2 (released ~1 month ago) is first time Glean's team feels comfortable routing 'majority' of workloads to open model. Marks inflection from closed-only to open-viable for 90%+ enterprise tasks.
  • Token spend vs. headcount trade-off: Jain's triage agent cost $1M/month in tokens—more than 15-person team's salaries. First time tech cost is compared to labor in same sentence. Open models flip this; frontier pricing unsustainable at scale.
  • Sovereign model demand (falling): Jain says enterprise desire for national models peaked a year ago, now declining—comfortable with US labs under contract. But Trump export bans may reverse trend. Europe/China may diverge.
  • Shipping speed vs. coding speed: Jain's distinction: AI writes 100% of Glean's code but shipping velocity unchanged. Bottleneck shifted to review. Coding speed ≠ product velocity. Key metric for engineering ROI.

Operator Notes / Why Ken Should Care

  • For agent systems & AI ops: Glean's $1M/month triage agent shows token costs can exceed salaries—open-source arbitrage is critical for sustainable agentic workflows. Invest in multi-model routing, cost dashboards, and context pre-assembly (RAG) to avoid brute-force token burn. Measure shipping velocity (features/releases) not lines of code.
  • For content/business & investing: Consumption pricing breaks bundling—enterprises can approve multiple tools and pay where work happens. Best-of-breed vendors (Glean, Harvey, etc.) have opening vs. Microsoft if they can prove ROI. But bundling psychology + compliance inertia still favor incumbents. Watch for CFO scrutiny in 2026/2027 ('where's my AI ROI?').
  • For GTM & workflow: Composite roles (AE+SE, eng+PM) are coming—hire T-shaped generalists who can leverage AI across functions. Analyst/sourcer roles at risk unless they evolve to strategy. Power-law adoption (5% of users drive advanced use cases)—target champions, don't blanket-roll-out. Show/tell AI wins in leadership meetings (Nikesh/Palo Alto model).
  • For recruiting & talent: Easier to hire vs. SaaS peak (tech layoffs freed talent) except ML roles (pay exploded). Seed-stage $500k/engineer salaries unsustainable—resist VC pressure to overpay. Model burn discipline even in land-grab. Glean expects 5x headcount growth (1k → 5k) despite AI—atypical; most CEOs shrinking teams.
  • For model landscape & geopolitics: 90%+ of enterprise workloads viable on open models (Llama 3.2, DeepSeek) as of ~1 month ago. Enterprises driven by cost, not sovereignty—comfortable with US labs under contract. But Trump export bans shift sentiment; Chinese models dominate open leaderboard (6 of top 7). US needs domestic open alternative or risks ceding inference layer to China. Regulatory capture risk (Sam/Trump 5%) but Jain skeptical it succeeds.
  • For capital allocation & startup strategy: Overabundant VC capital creates unsustainable structures (overpaying talent, ignoring unit economics). Jain is 'too disciplined'—feels pressure to land-grab but believes sustainable business needs charging for product, ROI on marketing, capital efficiency. Seed/Series A founders: model token spend vs. headcount, multi-model cost arbitrage, and measurable ROI before scaling burn. Investors: scrutinize if startup can win on fundamentals or just VC subsidy.

Watch Map

  • 00:00: Intro: Arvin's paranoia mindset ('only paranoid survive'), Glean overview (superset of ChatGPT/Claude, connected to enterprise context)
  • 03:20: Enterprise fear of frontier labs: operational dependency, losing institutional learning accumulated in agents
  • 07:45: Open-source viability: 90%+ workloads on Llama 3.2, Glean routing to cheapest model, Chinese model paranoia
  • 11:20: Frontier labs' vertical packs (Figma, legal) are 'shallow,' net-new expansion not displacement; Claude Desktop competes in Q&A
  • 18:30: Consumption pricing breaks bundling; Microsoft still formidable via 'free' psychology + vendor management overhead
  • 23:10: AI coding: 100% AI-generated at Glean but shipping speed unchanged—bottleneck is review; some companies skip review, risk tech debt
  • 25:40: AI ROI unclear: pockets of value (support, coding speed) but shipping/productivity gains fuzzy; Alex Karp right, 2026/2027 CFO scrutiny
  • 30:50: Glean's $1M/month triage agent cost more than 15-person team salaries—token costs 'absurdly expensive' at frontier pricing
  • 35:00: Debate: AI enables smaller teams vs. 10x more work requires more people. Jain expects Glean 1k → 5k headcount (atypical vs. CEO trend)
  • 42:00: Token budgeting: power-law usage ($15k/month vs. $20), no exec mandate at Glean (native AI co), Q&A universal but advanced use cases 5%
  • 49:30: Composite roles (eng+PM+design, AE+SE) replace specialists; data analyst/sourcer roles disappear
  • 52:10: Recruiting easier vs. SaaS peak (layoffs) except ML talent; seed startups paying $500k/engineer unsustainable, overabundant capital risk
  • 55:20: Sovereign models: demand falling (comfortable with US labs), but Trump export bans shift European sentiment; US open-source lags China
  • 58:00: Quickfire: advice (study CS, don't worry), legacy AI leader (Google), CEO realities (not sexy, perpetual unhappiness), founder style (discipline under pressure)

Source/Metadata

  • Title: Why OpenAI and Anthropic Won't Win the App Layer | Glean Founder
  • Transcript words: 15675
  • Duration seconds: 3608
  • Timestamp note: Timestamps estimated from transcript structure; YouTube chapters not provided in raw transcript but inferred from topic shifts.
Full transcript 10083 words · 69 min read
0:00

SPEAKER_00

90% or greater of use cases cannot be fully handled by many different models including

0:05

SPEAKER_01

open source models. Arvin Jain, the incredible founder of Glean, is one of the technology luminaries of the last decade. He founded Rubrik, which obviously IPO'd very successfully and is a brilliant public company now. He's gone on to found Glean, an incredible business today

0:21

SPEAKER_00

that's raised money from Kleiner Perkins and many other great investors. For almost all other AI companies that are not doing frontier model training, they should see the model companies as a huge asset. Once you move towards consumption, there's no inherent bundling advantage. You have to do 10 times the work to get the same amount of revenue from your customers.

0:51

SPEAKER_01

Arvind, I'm so excited for this. We have a mutual friend in Mamoun who says many wonderful things about you, and I think he's one of the greatest ambassadors of our time. So I'm really excited for this. So thank you for joining me. Thank you for having me. Now I think with entrepreneurs, you're either thrilled by winning and it's that chase to win,

1:19

SPEAKER_00

[SPEAKER_01] or you're terrified of losing and it's that fear of losing that inspires you. Which one are you? That's a good question. I think I would probably say the latter.

1:28

SPEAKER_01

[SPEAKER_00] I'm always worried about what can go wrong and that keeps me up at night. I love that. "It's only the paranoid survive." Has it always been that way? Yes, mostly. Even with the success you've had? Yes. It's so interesting. You know, Rubric

1:47

SPEAKER_00

[SPEAKER_01] was a phenomenal success, public company today, and you're one of the co-founders. [SPEAKER_01] It doesn't change with time? No. Because I think number one, every time you produce a new company or build a new project, it's like starting from scratch in my opinion. You have some good lessons from before, but it's a new world, it's a new environment. Think about Glean. It's fundamentally different from Rubric, and it's always possible, especially in the world of AI. You have to think that way because there is a disruption every single day. If you start to focus more on building on what you've already built,

2:21

SPEAKER_00

that's the winning mindset. You're winning with something, you want to double down on it. I think that's not sufficient in

2:25

SPEAKER_01

this new AI world. Can I ask, for those that don't know, can you provide a 60 second summary on what

2:31

SPEAKER_00

Glean is and how you work? Glean is an enterprise AI company. We started as a search company for businesses to help an employee quickly find information that they need that's buried across one hundred or one thousand different systems inside their company. That was how we started—it's Google for your work life. But then over time as AI models got better, it evolved into an AI platform. So today, the way to think about Glean is that first, it's a superset of ChatGPT, Claude, Gemini, all of those combined into one product experience. It's a co-worker

3:12

SPEAKER_01

[SPEAKER_00] for your employees, and it's connected to all of your company's context and how work happens inside your company. Mr. Alex Karp from Palantir went on CNBC last week and said that the large enterprises

3:31

SPEAKER_00

[SPEAKER_01] in the world were more skeptical than ever of frontier model providers. You work with some of the largest and have [SPEAKER_01] incredible customers. Do you agree with him? Are they more skeptical than ever? Two things. One, they're terrified of them. In the sense that, just like every software company is worried about whether they'll be in business, will the models eat it all. Similarly, enterprise leaders are also worried that their core IP, their data, their information, as well as their way of learning, their way of doing things—will it all be subject to too much

4:09

SPEAKER_00

technology dependence on these model providers? That feeling is there for sure. But I think what he said was that AI is not working

4:15

SPEAKER_01

[SPEAKER_00] in enterprises. Then everybody's afraid to actually say so because it's not supposed to be a cool thing before. Before we get to AI not working, because I think it's probably one of the most important questions, but it's a whole separate segment. Do you think they're right to be afraid of

4:28

SPEAKER_00

[SPEAKER_01] the frontier model providers eating their lunch or not? Well, yes. Depending on the enterprise. I think if we are talking about fundamentally changing how people work, and if we are saying that the majority of the work that we do today is going to be done by an agent which is fully powered by one of these frontier model companies, then in some sense, you have now transferred a lot of your operations to these technology providers. This is more than technology dependence. This is actually a real operational dependence on the companies that are actually running those agents for you. It's interesting. If you think about how work happens

5:16

SPEAKER_00

over time, when you initially do a task for the first time, maybe you'll document the process—what are the 10 steps you need to take to actually complete some piece of work. Then over time, people start to optimize and tweak that process. A lot of it never gets documented, and you just, based on doing this work over and over again, you've now built all these learnings that you apply in real time to do this work in the future. All of that institutional learning is actually going to accumulate in that agent that is doing that work. If you don't have any control on running that agent yourself, if you don't own the learning

5:55

SPEAKER_00

that it gains over the years, then you're basically fully dependent on these AI companies to get your work done. So it's absolutely a fundamental question in front of enterprises today: how do they actually use these AI technologies but still retain control and all the compounding learnings that happen

6:19

SPEAKER_01

with AI belong to the enterprises?

6:25

SPEAKER_00

[SPEAKER_01] Are you seeing enterprise customers root away from frontier model providers towards open source? That's something that's happening now. I think we are at a real inflection point with open source. Part of it, you're waiting on the open source models to actually get better. There's the desire there for many years. No enterprise that we talk to is okay with saying that, look, I can get my work done with OpenAI or with Anthropic and I'm good. Everybody wants to make sure that they are in control of their destiny, that they get to use many of these models. And now given that AI has become so expensive, if you look at the stories all the time about companies

7:09

SPEAKER_00

coming up with an annual budget for AI and they run past that within a month or two, poor CFOs. That's sort of really accelerated that desire for open source because now we have really good models in open source.

7:26

SPEAKER_01

What do they care about? Do they care about cost? Do they care about ownership in terms of their data staying on? [SPEAKER_00] make sure that they are in control of their destiny that they get to use many of these models

7:38

SPEAKER_00

and now given that AI has become so expensive, right, if you look at, you probably hear stories all the time about companies coming up with an annual budget for AI and they run past that within a month or two so in this poor CFOs, yeah, so that has really accelerated that desire for open source because we now have really good models in open source. What do they care about? [SPEAKER_01] Do they care about cost? Do they care about ownership in terms of their data staying on prem in models that they actually can have visibility on? What is it?

7:49

SPEAKER_00

I think right now the open source drive is coming from the cost point of view. I mean, there are certain businesses, of course, that have the requirements to actually keep all the inferencing workload within their own private data centers. When AI just came, there were companies that were a lot more afraid of getting their data outside of their own control and model companies training with their data, but that fear is no longer there. People believe that the model companies are going to be responsible and not train their models on enterprise data, so long as you've signed up for the right kind of contract. So right now, the drive is coming from cost.

7:56

SPEAKER_00

[SPEAKER_01] In terms of where you sit in the landscape, every one of your ambassadors that I spoke to asked this question, which is the obvious question: do you worry that Anthropic will do what they did to Figma, say, or what they've done with legal or what they're doing with health and move into your space and cannibalize your business?

8:02

SPEAKER_00

First of all, I think we should be careful in terms of what they've actually done for Figma or legal space or finance space. They are launching these vertical packs, but I think they're quite shallow in my opinion. And I don't actually know of people who are moving their workload entirely from Figma or from any other tool to Anthropic. It's actually net new always. I think it's expanding the market. For example, now in design, the designers still use Figma, but the non-designers are using Cloud Design, right? So that's what we're seeing: AI is making things simpler. If people are not experts, the primary users of that particular tool, they can start to do some of that work with Cloud.

8:07

SPEAKER_00

[SPEAKER_01] So you don't worry that they all put emphasis on moving into enterprise and being that context, whether they're doing it or not?

8:12

SPEAKER_00

Well, they're already doing it. We actually face that competition every day with enterprise customers. People often ask us, well, Cloud can also connect with enterprise systems through MCP. So what's different? What can Clean do which Cloud cannot? So we had to go and explain what context really is and why it's actually complicated to build it. So we are competing. In fact, I would say that they probably started to compete with us before others because if you think about Cloud co-work as an application or Cloud desktop, the primary use case for that has always been question answering, right? That's the largest application or use case for AI in the world today is in fact information seeking and question answering.

8:18

SPEAKER_01

How important do you think being first to market is? [SPEAKER_00] It's actually very advantageous, but it's only a thing that helps you. It's not going to carry you. For us, we actually get a lot of credit for being the first enterprise AI company in the world, the first ones to bring RAG into the enterprise, the first ones to build conceptual semantic search, and that actually gives us the brand and the right to compete in this market even though now we're much smaller compared to the giants that OpenAI and Anthropic have become. So it's a huge asset, but neither is it a requirement nor is it a savior.

8:33

SPEAKER_01

How do you advise founders who are losing sleep at night worried that the frontier model providers will come into that space?

8:38

SPEAKER_01

[SPEAKER_00] I would say absolutely don't worry about that. As a founder, you have to solve problems, not worry. Number one, right? Yes, you have to always anticipate what they're going to do. You have to see their current capabilities, but I think for almost all other AI companies that are not doing frontier model training, they should see the model companies as a huge asset, not competition, in my opinion. We actually believe that everything that Anthropic is doing, everything that OpenAI and Google is doing, as well as all the innovation that's happening in open source, that's great news for us. We don't worry about that. I don't think of that as competition. In fact, they've allowed us to deliver a product that we could never do without that help.

8:43

SPEAKER_00

[SPEAKER_01] Do you not think we're seeing the ultimate commoditization of the model layer? When you speak about Anthropic, OpenAI, and the rise of the model layer and the speed with which new models are coming out, especially from open source and Chinese providers, are we not seeing the ultimate commoditization of the model layer?

8:49

SPEAKER_00

One thing is clear: let's talk about enterprise use cases. Ninety percent or greater of use cases cannot be fully handled by many different models, including open source models. So there is definitely commoditization from that perspective. In fact, we at Lean, that's one of our core value adds to our customers, which is cost control. We will actually tell them that, hey, look, as people actually complete their tasks on our platform, we actually pick the right model for you, and if you're okay with using open source models, we'll use them when we think it's appropriate, when it's going to generate a high quality answer.

8:56

SPEAKER_00

[SPEAKER_01] What percent of customers are not okay with open source models?

9:04

SPEAKER_00

This is actually so new. We actually only have, I would say, open source truly coming to that level within three months of frontier capabilities. That just happened literally a month back or not even a month, right? What I would say is Llama 3.2 is the very first time where our own team, for example, feels comfortable that now we can run majority of our workloads on that model. So we are yet to find what people are going to tell us, and I don't think from a point of view of open source and using the model, everybody's going to be fine. The question is going to be: are they okay with the Chinese model or not? That's the only question here. It's not open source versus closed source.

9:10

SPEAKER_00

[SPEAKER_01] Why would they not be okay with the Chinese model when you look at the ownership that you have, the ability to have it on prem, you're not sharing anything back to China? our own team for example feels comfortable that now we can run majority of our workloads on that model. So we are yet to find what people are going to tell us, and I don't like, you know, from a point of view of open source and using the model, everybody's going to be fine. The question is going to be: are they okay with the Chinese model or not? That's the only question.

9:28

SPEAKER_00

[SPEAKER_01] Here, it's not open source versus closed source. Why would they not be okay with the Chinese model when you look at the ownership, that you have the ability to have it on-prem, you're not sharing anything back to China? Why would you not be?

9:34

SPEAKER_01

[SPEAKER_00] I think it's just comfort. It's just, what if something goes wrong? There's always paranoia and fear. What if there's a back door, some magic back door that we don't even understand? Then that could be a back door. So there are some concerns. There's also, if you use these models and if it becomes a known thing, it could be used against you in some ways by your competitors and things like that. So variety of factors, but ultimately it boils down to who's willing to be bold because this is a new thing. In large enterprises, I have to make this move, and the early movers will make the move first and then it'll become a more normal thing.

9:38

SPEAKER_00

[SPEAKER_01] I'm doing this show to learn: if 90% of enterprise workflows can be done with open models, have we completely mispriced the frontier model landscape? That's a very different time.

9:43

SPEAKER_00

I do feel the model business, regardless of open source, the there's plenty of competition even within the labs and then more companies are coming into that space. So in that fierce competition, even in a three-way race, I think you can actually get good amount of pricing pressure. And now with open source, it actually is an order of magnitude cheaper prices. So I actually heard rumors that OpenAI was going to drastically reduce their model prices in response to these developments and competition from open source. So the model business on its own is probably not as lucrative as everybody believes. But these companies now have a lot more things. They're not model companies only.

9:49

SPEAKER_00

[SPEAKER_01] If they're doing shallow things in those adjacencies, they're not exactly going to generate a trillion dollars of revenue, like Dario said. If you think about it, these two labs are very fundamentally different businesses. OpenAI of course has an amazing consumer product, and Anthropic—the interesting thing that is happening is people are actually building on top of their platform. So when you think about Anthropic right now, there are a lot of folks who are actually developing automations and skills, creating these MCP servers to their internal systems, connecting it all to Claude. So there's an ecosystem that's being developed. So they should very much be considered an application-level company, not just a model company.

9:55

SPEAKER_00

[SPEAKER_01] If you were to make a guess, in three years time, what percent of your workflows do you think are through open source? Well, we've been telling customers I believe that majority of enterprise workloads will actually be on open source models in three years for sure. [SPEAKER_01] Yeah, another competitive element that you face from the model providers is Microsoft. Microsoft have made a phenomenal business on the back of creating a 70% as good product but bundling it into a bundle for enterprises and then selling it with a nice sticker on it. Yeah, how do you think about the bundling pressure from Microsoft Copilot as a competitive threat?

10:13

SPEAKER_00

Well, for us, they are one of our most significant competitors. And the bundling strategy actually works. And to fight against that, there's always been room for best-of-breed software. People and all customers think of us exactly like that. If you are trying to bring a great search product, if you're trying to build a horizontal comprehensive AI platform, they know that we do it better. So companies are willing to invest on top of that as part of the bundled product suite from Microsoft. But the other thing is, what's making bundling not as effective of a strategy anymore is the fact that AI is moving towards consumption-based models. Once you move towards consumption, there's no inherent bundling advantage because a business can get six tools and let the users choose where they want to do their work. Wherever they do their work, I have to pay for it for that particular unit of work. So consumption can ultimately break that bundling strategy.

10:19

SPEAKER_00

[SPEAKER_01] Respectfully, I don't know if it does in enterprise because they will make you compliant as an enterprise bundle. You'll go through approvals processes, sign-off processes internally. For the largest enterprises in the world—your VWs or your Fords or your GEs or Tyson Chickens, I always use them as examples—they'll approve Microsoft as one vendor. If they're suddenly having to approve 15 vendors, forgetting the pricing and the transactions, it creates a vendor management problem that they didn't have before.

10:24

SPEAKER_00

That is true. But I think if you go and talk to companies that have been on the other side of Microsoft, most of them will talk about pricing as the main killer.

10:30

SPEAKER_01

Because I think it's hard to compete with free. Who's a fiercer competitor—Microsoft or the frontier models?

10:35

SPEAKER_00

Good question. I think it's early to tell, but Microsoft is formidable. If we look at our experience as we go and prospect, I think we would hear this answer more often: well, we are a Microsoft customer and we already have Copilot, so therefore it doesn't make sense for them to consider us. We do hear that, and we hear that more often than we hear from somebody who has embraced one of the lab products and therefore there's nothing else that they're going to do.

10:43

SPEAKER_00

[SPEAKER_01] We mentioned Alex Carp at the beginning and I interrupted you. Let's before we dive—we're not getting value. Yeah, I think 2026 H2 and 2027 is the year where everyone goes, "Hang on a minute. Is this spend generating output or return?" [SPEAKER_01] How do we think about the return on investment that enterprises are getting, and is Alex Carp right in saying everyone's going to go, "Where's my return?"

10:55

SPEAKER_00

I would say there are pockets of value realization today. For example, take customer support as a vertical. I think there it was easy to measure productivity. You could actually say that in your company, your support agent resolves 10 cases a day, and now they're able to do 12 because of AI. So you could see that being measured.

11:02

SPEAKER_00

[SPEAKER_01] Let's before we dive in, are we not getting value? Yeah, I think 2026 H2 and 2027 is the year where everyone goes. So hang on a minute, yeah, is this spend generating output or return? Yeah, ROI. How do we think about the return on investment that enterprises are getting and is Alex Carp right in saying everyone's going, what, where's my return?

11:07

SPEAKER_00

I would say that there are pockets of value realization today. For example, take customer support as a vertical. I think there it was easy to measure productivity. You could actually say that in your company your support agent resolves 10 cases a day and now they're able to do 12 because of AI. So you could see that being a very concrete measure of productivity increase. That's a use case where AI is actually pretty good because a lot of the time that is spent by the support teams is about reading knowledge and then summarizing it to your customers. So there are definitely areas where there is clear value realization and enterprises are feeling good. Some other ones are more complex. For example, I think the majority of the AI spend right now is on coding. And you know, coding as a practice has changed. Most developers now actually use AI to write code. They're not writing it by hand anymore. So that part, you can in some ways say that yes, AI made a big impact. But are they shipping the products faster or not? That's where we hear most of the companies saying that no, the actual shipping speed of products has not increased even though coding speed increased significantly. Because I know that's only a small part of overall shipping a product. Has your shipping speed increased? I would say it's hard to actually measure. That's the challenge because engineering productivity is one of the most difficult things to measure. It's the fuzziest of the jobs out there. If you look at some of the metrics like lines of code written, of course we're writing way more lines of code now. But if you look at whether we're shipping features at a greater pace, yes we are. But it's also a result of a larger team. We have a team that is more tenured than it was before. So sometimes it's hard to tease it apart. But with that, what do we do as a company? We are right now saying that we're just going to keep investing. What percent of Glean code do you think is written by AI now? It's probably almost 100 percent. Nobody's actually writing the initial code by hand anymore. Maybe sometimes you've got an artisan in the corner. So almost all the code is being written with AI. But we actually have to enforce human reviews. You cannot actually generate tons of AI code and then just check it into the repos. We're probably more conservative than most other companies. There was in fact a discussion inside the company that well, now I can write so much code and the real bottleneck has shifted from the person who writes the code to the person who reviews. So there was a proposal to actually eliminate code reviews and then just submit directly. Many companies are doing that. They let AI write the code and it gets submitted directly into the repos.

11:14

SPEAKER_01

If you have to have a stringent code review process, it almost removes the point of having a fast code development process. [SPEAKER_00] That is true. So that's why I think what it's doing right now is we're still in the learning phase of using AI properly and effectively, thinking about long-term ramifications of it. Because when you write code with AI, you can write a million lines of code, but it becomes incredibly hard to actually maintain it, understand it, and manage it over time.

11:26

SPEAKER_00

[SPEAKER_01] So what AI does, you have AI that does refactoring and AI that does security? Yeah, the only thing is that it's not that perfect right now. So you have to. I think we would rather right now be willing to pay the cost of being slow, of reviewing the code. So we're still faster than before because the writing part is actually much faster now and the person who writes is the one who actually does the first review. So when you say AI ROI is really a throughput problem, what does that mean?

11:37

SPEAKER_00

The first thing that we have to do is make sure that you are able to bring the right context to these AI agents. The way I think most enterprises today are rolling out AI is they actually just throw it into the system and connect AI with all of enterprise systems in a rudimentary manner using MCP servers. Now you're letting any piece of work that you are trying to do with AI brute force its way into trying to figure out and assemble the right raw materials that it needs to complete the task. In this mode, AI is super slow. It takes a lot of time to actually just assemble the basic information it needs to do the work. It also becomes very costly because most of the tokens are being burnt just trying to assemble the right context for that given task. And you're trying to use AI for things where it's not even good at or needed. So instead, what we talk about is to make AI really perform and deliver, you have to invest around it. You have to make sure that you provided the right context so that it can actually work faster at lower cost.

11:42

SPEAKER_00

[SPEAKER_01] What does it mean invest around it? And are we wrong as CEOs to be urging all of our team members to be trying to replace themselves with AI even if it means that we're wasting tokens? I think it's a wrong goal in my opinion to say that hey, replace yourself with AI. First of all, I think you're giving too much credit to AI. When you say that, it's just not ready right now. Give me a name of one job that you can replace with AI. For example, do you think it can replace your EA? [SPEAKER_01] Mine? No, but I'm a diva.

12:01

SPEAKER_00

For most people, I think it can do the majority. And that's the thing. It can actually take care of a lot of things for any given role, but it cannot replace the final, you know, the last intangible. [SPEAKER_01] But that can be a tipping point where actually for a lot of people, if it does 90 percent, fine, you know what, you'll do that birthday present for your wife because it's once a year, it's not very often. And Claude isn't quite personal enough to know your wife's preferences of perfume. So it's very. But that role will get cannibalized. I'm not sure. And I'll tell you why. I think you want to be performing the best in whatever you do, and I don't think you're

12:15

SPEAKER_01

[SPEAKER_00] take care of a lot of things for any given role but it cannot replace that the final you know the last intangible no but that can be a tipping point where actually for a lot of people if it does 90 fine, you know what you'll do that birthday present for your wife because it's it's

12:28

SPEAKER_00

[SPEAKER_01] once a year it's not very often and claude isn't quite personal enough to know your wife's preferences [SPEAKER_01] of perfume and so it's very but they'll but that role will get cannibalized i'm not sure and i'll tell you why i think the you want to be performing the best in whatever you do and i don't think you're going to take a 90 solution but i'm not cost constrained being a dick well i mean look the they're kind of like you know it's not about you not being cost constrained it's about you have to be competitive in your in your work with others i remember they also have all the ai

13:09

SPEAKER_00

tools that you have but if they also have a human on top you know how are you going to compete with

13:13

SPEAKER_01

them so do you not think how many people would you have now in the in our company yeah we're

13:18

SPEAKER_00

[SPEAKER_01] over a thousand people now over a thousand people how many do you think you'll have in five years time well hopefully you know five thousand wow so yeah we're going to grow but that is very atypical you i [SPEAKER_01] sit with the biggest ceos in the world and every single one of them is shrinking teams and every [SPEAKER_01] single one of them is saying i absolutely don't believe in it why why well i think first thing logically the uh take two companies take cola and pepsi or two companies that compete with each other um one company desires to shrink and the other one is still has a

13:59

SPEAKER_00

lot more people both of them have full access to the same ai tools and technology and so now the question is

14:03

SPEAKER_01

[SPEAKER_00] is the if you were trying to do the same amount of work and you believe you can do it with fewer [SPEAKER_00] people and therefore you shrink um your competition can also do the same but they chose actually not

14:14

SPEAKER_00

to do the same amount of work they chose to actually elevate and build a 10x better product or build 10 times more produce 10 times more goods because they have more people they're going to be larger they're going to beat you but i don't think more [SPEAKER_01] people makes for better products that's a different thing if i can cut head count yeah and [SPEAKER_01] then afford the best frontier models the best technology for my 100x engineers because i've [SPEAKER_01] reduced head count yeah the best engineers will want to come to my company and actually i'll be [SPEAKER_01] creating better products faster with my smaller team that's a good point but i think that

14:56

SPEAKER_00

[SPEAKER_01] you know that argument and i think more people slow down everything but

15:01

SPEAKER_01

[SPEAKER_00] that's not an ai argument that argument has always been true sure you combined with the ai element of you're able to ship more if you're able to ship more i promise you when you have and you know this

15:15

SPEAKER_00

[SPEAKER_01] when you have more people they'll just put up the barriers to get in the way of that product going out well look the even in the we have these ai discussions right now but before that i think that just post covid many companies felt they were bloated they cut down 15 percent 20 of their staff and every ceo came out and said they're actually as a result of that they're actually moving 20 faster right so a lot of companies came and talked about that so that's that's the argument is always there at some point you know teams get large they start to slow each other down

15:56

SPEAKER_00

you know humans do that i also believe in that but the ultimately people are also your asset and you have to be able to deploy them correctly in the right set of projects and i don't think the world's greatest companies are going to be companies with 100 people or and look at the model companies you know same for them why are they hiring so aggressively do you not think that the best people

16:22

SPEAKER_01

will want to work with the best technology and we'll see an increase in technology spend by the

16:27

SPEAKER_00

[SPEAKER_01] biggest companies in the world from 8 to 12 where it is today to maybe 16 to 20 yeah and then actually [SPEAKER_01] you'll see a reduction in headcount but an increase in technology spend and the best people want to go where they have the best tools and equipment i'm not sure about that either because i think technology is actually not supposed to increase in cost first of all i think do you admit that the currently this technology is priced in absurdly for what it delivers [SPEAKER_01] i think it totally depends on what it's doing so no i don't at all for cursor or for any of the dev

17:03

SPEAKER_00

[SPEAKER_01] tools i think it's still dramatically underpriced when you look at mark benioff spending 300 million on [SPEAKER_01] anthropic it's 3.7 percent of developer salaries i think that's relatively small i would say it's

17:13

SPEAKER_01

[SPEAKER_00] absurdly expensive i'll give an example we had this we have a really cool triage agent for

17:18

SPEAKER_00

engineering and we have 15 people team on call team that was supposed to that their work was actually triage every single production issue that happens any system alerts you know things

17:30

SPEAKER_01

[SPEAKER_00] that are going bad and we built this agent that actually now is taking care of 95 of those [SPEAKER_00] issues automatically for them but even there is actually doing it a cost which is actually [SPEAKER_00] questionable is it actually more effective that is more efficient than humans [SPEAKER_00] we were spending a million dollars a month on that particular agent

17:53

SPEAKER_00

and that was actually more than the cost of a million a month yeah are you buying cristiano ronaldo what are you doing well i mean costs are like that i mean there's it is quite [SPEAKER_01] expensive so the sorry you said you said because i discussed this a lot on the show so you're [SPEAKER_01] making me much smarter you think that spending 3.8 percent of developer salaries on these tools is a lot [SPEAKER_01] if you think that's a lot then these model providers are absolutely screwed well i think the point that i'm making is the 3.8 percent number is actually doesn't seem high at all

18:32

SPEAKER_00

when you look at it that way yeah but i also know that already you see with open source that you can do the same amount of work for a tenth of the cost right that's number one and number two historically for as far as i can remember you know we've not put technology cost and labor cost in the same sort of sentence ever before this is the first time we're actually hearing that that hey i would rather have fewer humans and more tokens the first time and i just feel this is not how technology works the models are supposed to get cheaper and cheaper the tech is

19:13

SPEAKER_01

going to be more and more affordable but it's i'm so sorry this is so funny for me because you're the co-founder of clean and rubric and so who am i but a podcaster but this is exactly what technology is for this is agents being proactive having an opinion making a decision they should absolutely be included or put in the same sentence as labor because they are replacing the labor that we [SPEAKER_00] used to spend money on i think good technologies figure out how to make technology really cheap

19:43

SPEAKER_01

[SPEAKER_00] that hey i would rather have fewer humans and more tokens the first time and i just feel like this is not how technology works. The models are supposed to get cheaper and cheaper. The tech is going to be more and more affordable but it's—i'm so sorry this is so funny for me because you're the co-founder of Clean and Rubric and so who the fuck am i but a podcaster, but this is exactly what technology is for. This is agents being proactive, having an opinion, making a decision. They should absolutely be included or put in the same sentence as labor because they are replacing the labor that we used to spend money on.

19:47

SPEAKER_00

I think good technologies figure out how to make technology really cheap and it's going to happen here too. That's my belief. You're going to see inferencing costs come down by orders of magnitude. I think we saw something bizarre actually in the last six to nine months—every model actually increased their per token price. If you go back 15 months, everybody thought that the per token price is going to just keep falling like it was before. So we don't know what happened here. This is also sort of unique. They needed to prove that they were good businesses before they went public. That's what happened. I can say things that you can't.

19:53

SPEAKER_00

[SPEAKER_01] Yeah, yeah, but my bet is on AI getting much much cheaper than what it is today. If AI gets much much cheaper than it is today, these already loss-making businesses which prop up our entire global economy pretty much at this point are very threatened.

20:00

SPEAKER_01

Yeah, i mean my take remains the same. So okay, so it's really interesting. So you don't expect an engineering team to get smaller in the future?

20:05

SPEAKER_00

I think per person productivity is going to shoot up but so will the demands. To make the same amount of revenue, you have to produce a 10x better product in the future. Unfortunately, when you think about token spend internally, how did you sit down and think about it as a team sitting with your CFO? How did you go through the decision of how to think about token budgeting?

20:13

SPEAKER_00

Well, I think we did probably what most companies did, which is we didn't do anything. I think because we were in this phase of let people figure out what they can do with this tech. What did you see? People went crazy. People didn't adopt it. What happened? There's a power law. In our company and also at all of our customers, you will see some people who spend ten thousand dollars or fifteen thousand dollars in tokens every month and then you have others who are spending twenty dollars. One thing is interesting though—that everybody has embraced AI to some degree. Everybody is using the basic—as I mentioned before, the number one application of our use case for AI today in the world is information seeking, question answering, and everybody's doing that. So you see the entire—everybody on the team and our team as well as our customers, they're all doing that. Everybody's asking questions. Everybody's getting some basic summarization, information synthesis going, but the advanced use cases are limited to like five percent of the employee base.

20:23

SPEAKER_00

[SPEAKER_01] Is there anything you do as a leader to try and infuse AI as aggressively as possible? We had Nikesh from Palo Alto on every week. He has a leadership meeting where he's like show and tell and everyone needs to stand up and show something that they've done with AI that week—that it replaces what they do, improves their job, whatever it is.

20:30

SPEAKER_00

Is there anything that you can do? Yeah, that's actually a really good idea. I've thought about doing that. We limited the token maxing dashboards and I always thought that was not the right idea—to just reward people who are consuming more tokens. I felt like we didn't need to do that. We are a native AI company ourselves and people are already kind of educated enough and they will use AI when they need to. But executives sharing a success story—we haven't demanded it from every single exec, every single week, but we have the showcase. In our town hall, for example, we'll always ask people to share those wins like every town hall. There's a section dedicated to these are the new AI agents that teams are using to do work, to work differently.

20:36

SPEAKER_00

[SPEAKER_01] Can I ask in terms of the execs and the people that you have? I think recruiting has never been harder. How hard is recruiting today with some of the largest model providers as we said paying just enormous salaries that we haven't seen before? Yeah, I would actually say maybe the last two or three months—let's put that aside for a minute. I would say that recruiting was actually getting easier compared to the SaaS peak.

20:46

SPEAKER_01

Wow, why?

20:51

SPEAKER_01

[SPEAKER_00] Because I think companies have been more—they haven't been growing their head count. If you look at the largest employers of tech talent, many of them actually haven't been growing. Many of them have been laying off continuously. I think about Meta, for example. In every year there's significant layoffs and I don't know the overall head count, but my guess is that it's probably down from the peak of 2021 or 2022. So actually there was more talent available in the market than before. But if you now start to talk about okay AI talent, ML talent, top people are sought after way more than ever before. And the pay scales have completely changed. Not just from the model companies but even from startups because startups are also giving them too much money to compete. So even startups these days pay a lot.

20:57

SPEAKER_01

We have to. Yeah, yeah, we have to because their alternatives are so large too. And actually if it costs three, four, five hundred grand for a great dev, well the two million dollar seed round just doesn't go anywhere. Yeah, for the founder building the team, even if they don't take a salary, if I'm gonna hire four people, yeah, I need six million bucks. [SPEAKER_00] That's right. Do you think founders should raise large seed rounds?

21:20

SPEAKER_00

I think it's better. I always prefer to raise as much for a round as you can from the get-go. What round felt the most highly priced?

21:28

SPEAKER_00

First, actually we never actually went out to raise except for our first round of the company. We always had somebody come in and it was a relationship that got built over some time and kind of became the de facto—they are going to be the ones putting money in. I would say our Series C probably felt the most, i guess you could say the most expensive because we barely had any business. Definitely sub two or three million dollars, maybe five million dollars. I don't remember exactly, but then the valuation was north of a billion. So that was extreme. I think we—you know, we take what we get.

21:36

SPEAKER_00

[SPEAKER_01] I mean, that's incredible. Do you worry about scaling into that when you're doing it or would you just head down and think this is great—a low dilution for a high price? The way we thought about it more was that was a statement to be made to the prospective employees, more than anything else, if we wanted to make the market understand that we're building something special. because we barely had any business, maybe sub two or three million dollars, maybe five million dollars, I don't remember exactly, but then the valuation was north of a billion. That was extreme. I think we take what we get.

21:56

SPEAKER_00

[SPEAKER_01] That's incredible. Do you worry about scaling into that when you're doing it or would you just head down and think this is great, a low dilution for a high price? The way we thought about it more was that was a statement to be made to the prospective employees, more than anything else. If we wanted to make the market understand that we're building something special, that kind of gives us that validation. [SPEAKER_01] Do employees care who your investors are?

22:18

SPEAKER_00

[SPEAKER_01] Absolutely yeah. A lot of founders are like, what, the best people don't care, they're there for the mission. I'm always like, I promise you, if you have Kleiner or you have DST or you have Sequoia, great candidates suddenly want to talk to you a lot more.

22:24

SPEAKER_00

Yeah, investor reputation directly impacts your reputation. When you look today, what have you changed your mind on most in the last 12 months? I've always personally, my style has been a little bit too disciplined to be the right strategy anymore. I get that feedback from my team that we are trying to be conservative, trying to make sure that our capital goes a long way. In that mindset, we may lose the land grab. I'm feeling the pressure to change it myself, change how I think about how we should be spending, how we should be investing. But at the same time, I have this fundamental belief that a business is always built on discipline. You have to charge for the product. It has to generate value for the customers. For every dollar that you invest in marketing, there has to be some good return back from it. You're not assuming that you just keep raising the money to make up for all those things that were not there.

22:31

SPEAKER_00

[SPEAKER_01] Do you agree with that when you have examples like Uber, which proves that a bad business model can turn good with scale? Yeah, that's what I'm saying. That's where I feel that pressure, that perhaps my way of thinking is incorrect. Do you think it is a land grab? We're absolutely in a land grab. No question. Every single company in the world wants a product like ours today. Either we get in today or it's going to be 10 times harder to actually get in the future. [SPEAKER_01] We spoke about job displacement. We had an interesting conversation around that. What job does not exist today that you think will be incredibly common in three to five years time?

22:54

SPEAKER_00

Composite roles will be very common. For example, somebody who can build a product, I don't know what to call them, but they act like engineers, product managers, designers. Similarly, in go to market, somebody who can sell the product and they are capable of not only doing the business negotiations, but they can actually demo the product, they can actually talk about use cases. Instead of having that segregation between account executives and solution engineers and then sales solution architects, I think we will see more and more generalization of roles, away from specialization. In fact, I was trying to drive that very hard even in our own company.

22:59

SPEAKER_00

[SPEAKER_01] In our company, but I'm sorry, I mean this in a nice way, the composite roles goes exactly against the idea of maintaining team size because if you have composite roles where you bring in four different specialties into one, that is smaller teams. It is, yes. But you have to do 10 times the work to get the same amount of revenue from your customers. In the future, you have a much smaller team to deliver the same amount of work that you used to deliver before. You just are forced to do more. [SPEAKER_01] Got you. Okay, and then what role do we have today will we not have? What do we look at and go, oh my gosh, I can't believe we used to do that?

23:15

SPEAKER_00

A lot of analyst roles. The data analyst roles which are not business thinkers. They were given a task like, I need to see this data, and then they produced specific dashboards, configure back-end systems. I think that kind of work definitely goes away. I think business intelligence is going to be very different. Business owners will directly be able to get answers to their questions. So business analysts and data analysts, that's one. Many HR roles, sources, for example. In recruiting, that's the role that I think is going to definitely get consumed into a full cycle recruiting role.

23:23

SPEAKER_00

[SPEAKER_01] I do have to ask one final one. We're sitting here in Europe and it brings about a question of sovereignty. The US and Europe bluntly have not come up to muster on open source models. Do you think we will have a world of sovereign models? And do you think, given what we've seen in the last month or so, that we need to have sovereignty over our models?

23:31

SPEAKER_00

The desire for sovereign models is strong. It was probably stronger a year back compared to now. I feel like I'm hearing less of it. There was a period where every nation thought that they could build one. AI was still in its early stages, but then a lot of those nations actually figured out that's not going to be the way. They're okay with letting their enterprises within their own countries use OpenAI or Anthropic or all the other models. I'm not an expert. I don't know whether this trend is on the rise or on the fall a little bit.

23:38

SPEAKER_00

[SPEAKER_01] It's unequivocally on the rise given what we saw with the Trump administration banning Anthropic's latest models. It's understood from a lot of Europeans that we cannot rely on a US individual who could ban our access to intelligence. But where are the results from it? That was a month ago, so to expect a sovereign model within three weeks would be tough. [SPEAKER_01] Yeah, yeah, but even before that, I think it just hasn't happened, right? The only country in the world that has produced models outside of the US is China. And then of course, maybe a little bit in France with Mistral.

23:59

SPEAKER_00

[SPEAKER_01] Is that simply an incentive problem? The lack of open source community in the US? I think there is a good open source community in the US in many other areas. [SPEAKER_01] But what open model from the US? No, you're right. We don't have open models. But it's not because open source as a movement, as a concept, is weak in the US. It's actually quite strong. It's probably if you think about models, they require a lot of upfront investment, which is not open source friendly in many ways. China, and then of course maybe a little bit in France with—

24:31

SPEAKER_00

[SPEAKER_01] Is that simply an incentive problem? The lack of open source community in the US and why we don't have any US open source to any real degree in substantiveness?

24:37

SPEAKER_00

No, I think there is good open source community in the US in many other areas. I mean, what open model from the US? No, there's no—yeah, you're right that we don't have open models, but it's not because open source as a movement, as a concept, is weak in the US. It's actually quite strong. It's probably—if you think about models, they require a lot of upfront investment, which is not open source friendly in many ways. A lot of open source software has been skunk works—like developers getting no funding associated with them, and they still get something built. They couldn't build models that way. So that's why naturally this thing didn't work out. And you need these techniques where super high investment is not needed.

24:42

SPEAKER_01

Do you worry then when you look at the state? I spend a lot of time on OpenRouter and I see the model usage and traffic. Anthropic today was the first US model and seventh—the first six were Chinese. Do we just not care that the CCP are funding the top six models? The fact that you can actually run inferencing on those in that contained environment makes people feel comfortable. [SPEAKER_00] I don't think the US will feel okay with that trend. There's good work that's happening now to actually promote open source and model development in the US. There's some models coming out.

24:55

SPEAKER_00

[SPEAKER_01] The alternative is that Sam and OpenAI give five percent to Trump, and then he puts regulatory capture on Anthropic and OpenAI and puts attacks on open source. I hope not. I doubt that's going to happen. [SPEAKER_01] Why would I—I'm sorry—why else would Sam give them five percent? It's a quid pro quo. I need you, you need me. I mean, I just believe more in the US system and ultimately curbing—I don't think right now you need to curb open source. It's too far behind in the US.

25:25

SPEAKER_01

Don't you think Sam and Dario are sitting thinking, "Oh wow, we underestimated this and this is a core threat to our business"?

25:32

SPEAKER_00

They probably are thinking that, but I don't think they can fix that through regulation. [SPEAKER_01] Don't you think Sam will be calling up Trump, who he has a direct line to, saying, "Hey, the CCP are funding our biggest competitors and we cannot promise that there isn't a backdoor to Xi Jinping. You need to stop this and I'll give you five percent for your troubles"? Isn't the argument the other way around? Right now there are all these open source models which are very good and they're all built in China. The US needs to build its own. We can't be seen as a country that doesn't innovate on technology. So it's actually paramount for the US to build.

25:51

SPEAKER_00

[SPEAKER_01] I think Sam will be saying it takes billions of dollars and years of time. Trump should defend America and support OpenAI and Anthropic and put barriers up to prevent Chinese open source models from getting adoption.

25:55

SPEAKER_01

[SPEAKER_00] Taxes, bans—those maybe, yes. But US open source models are going to have a lot of tailwinds. This is a known accepted issue that every technologist in Bay Area talks about. There's a lot of motivated parties that actually want to promote development of great open source models in the US, including Nvidia, for example. They're putting a lot of investment in promoting development of great open source models in the US.

26:02

SPEAKER_00

[SPEAKER_01] I hope they succeed. A multi-model world is important for all of us. Listen, I'm doing a quick fire round with you. I say a short statement, you give me your immediate thoughts. Does that sound okay? Okay. [SPEAKER_01] What's your biggest advice to someone studying computer science today? It's fine to study it. Don't get too worried because of what other people are telling you. [SPEAKER_01] Which legacy company has adopted AI the best, do you think? Well, are you willing to call Google a legacy company? [SPEAKER_01] Yeah.

26:54

SPEAKER_00

Yeah, so Google probably rates higher than anybody else in terms of not only embracing AI internally but also launching products. But I guess they are AI companies, so it's kind of hard—it's unfair to put them in that category. [SPEAKER_01] You start a new company and you can only take one ambassador who do you take with you? I think I'll take one of our existing ones. We create relationships with all of them. [SPEAKER_01] Which one would you take? I don't know. I won't answer that question. I just don't have the answer. I need to think about it. I think it's probably circumstantial, depending on what I'm doing. Different people bring different strengths.

27:25

SPEAKER_00

[SPEAKER_01] What would you most like to change about the startup ecosystem that we see today?

27:32

SPEAKER_01

[SPEAKER_00] I actually do think that there is too much capital available today for startups, and it's actually sometimes creating failure paths for people. I think they're not getting what it takes to build a great company. I'll give you an example. A startup that has raised a seed round decides to pay half a million dollars to an engineer. And this is happening today. The startup founder is okay with it, the investors are okay with it, but it's surely not a sustainable path to actually win. They're paying it while Google is not, because Google knows that they don't need to actually buy talent like that. So I think that's one thing—this overabundance of capital is getting startups to create structures which are not going to be sustainable for them.

27:38

SPEAKER_01

Do you worry about the lack of exit options that are now becoming more and more real? What I mean is, honestly, if you don't have a billion in revenue today it's hard to go public. Tech acquirers—your big companies are very specific about what they want to buy. Meta licking its wounds from having a portfolio that's full of failures—it's a tough landscape. Startups have never been easy.

27:43

SPEAKER_00

I think in fact, for all I would say in the last 25 years that I've seen, it's been easier to build a startup and get a good exit from it these days than it used to be in the past. Startups are brutal—it's a brutal game. [SPEAKER_01] What does no one know about being a founder and CEO from the outside that they should know? It's not a sexy job. It's actually one of the most stressful things, and you really have to be crazy. [SPEAKER_01] I think they know that now. One for me is that you have to consistently be unhappy. You should never be happy as a CEO because there's always something that needs doing or could be done better.

28:08

SPEAKER_01

[SPEAKER_00] Telling someone you will never be happy is something that I—I'm jarred by. Yes, that's a good one. This is a tough job all around. I think oftentimes people who have not done it feel that there's a lot of glamour. They feel that—

28:16

SPEAKER_00

Being a founder and CEO from the outside, they should know that it's not a sexy job. It's actually one of the most stressful things and you really have to be crazy I think.

28:24

SPEAKER_01

I think they know that now. I think one for me is that you have to consistently be unhappy. You should never be happy I think as a CEO because there's always something that needs doing, could be done better. Telling someone you will never be happy is something that I jarred by. [SPEAKER_00] Yes, that's a good one. This is a tough job all around and I think oftentimes people who have not done it feel that there's a lot of glamour. They feel that you're going to make a lot of money and your life will be fantastic, you're going to have a lot of respect. I think almost all of those things are irrelevant. This is—you have to be truly mission oriented to survive as a founder.

28:37

SPEAKER_01

Did your style change with money? You've been successful before I think better. I think founders are better and investors are better when they are already rich if I'm being blunt. I think you make more rational, sound decisions that are not made with economic impatience. I think for me maybe not.

28:43

SPEAKER_01

[SPEAKER_00] But at the same time, I've been a man with minimal needs and my needs are already met a long time back. So I guess I've definitely built these startups without that worry of can I feed my family. So yeah, maybe that has helped me. But as I've seen more success it doesn't change me fundamentally. I still have to have that drive, you have to work continuously, work more than every other person in your company, lead by example and keep pushing. You have to have this rational need to make something big happen.

28:50

SPEAKER_00

[SPEAKER_01] I so appreciate your time. I apologize for being robust in my discussion. Yeah, I think it was a different interview to a lot of interviews that you do where it was more discursive, but I so appreciate the time and you've been fantastic dude. [SPEAKER_01] Yeah, thank you. you have to be competitive in your in your work with others i remember they also have all the ai tools that you have but if they also have a human on top you know how are you going to compete with

29:22

SPEAKER_01

them so do you not think how many people would you have now in the in our company yeah we're over a thousand people now over a thousand people how many do you think you'll have in five years time

29:32

SPEAKER_00

well hopefully you know five thousand wow so yeah we're gonna grow but that is very atypical you i

29:40

SPEAKER_01

sit with the biggest ceos in the world and every single one of them is shrinking teams and every single one of them is saying i absolutely don't believe in it why why well i mean i i think like

29:51

SPEAKER_00

first thing logically the uh take two companies take cola and pepsi or like you know two two companies that compete with each other um one one company desires to shrink and the other one is still has a lot more people both of them have full access to the same ai tools and technology and so now the question is is the the if you were trying to do the same amount of work and you believe you can do it with fewer people and therefore you shrink um your competition can also do the same but they chose actually not to do the same amount of work they chose to actually you know elevate and build a 10x better product or

30:36

SPEAKER_00

build 10 times more um produce 10 times more you know uh goods because they have they have more people they're going to be larger they're going to be they're going to beat you but i don't think more

30:47

SPEAKER_01

people makes for better products that's that's a different thing if i can cut head count yeah and then afford the best frontier models the best technology for my 100x engineers because i've reduced head count yeah the best engineers will want to come to my company and actually i'll be be be creating better products faster with my smaller team that's a good point but i think that you know that argument and i think more people slow down everything but

31:13

SPEAKER_00

that's not an ai argument that argument has always been true sure you combined with the ai element

31:19

SPEAKER_01

of you're able to ship more if you're able to ship more i promise you when you have and you know this when you have more people they'll just put up the barriers to get in the way of that product going out

31:29

SPEAKER_00

well look the like even in the we have the we have these ai discussions right now but before that i think that um just post covid um like like many companies felt they were bloated they cut down 15 percent 20 of their staff and every ceo came out and said they're actually as a result of that they're actually moving 20 faster right so a lot of companies came and talked about like you know talked about that so so that's it that's it that's the argument is always there like you know at some point you know teams get large they start to slow each other down you know humans do that i also believe in that but

32:09

SPEAKER_00

but the ultimately um people are also your asset and you have to be able to deploy them correctly um in the right set of projects and and the i don't i don't think the the world's greatest companies are going to be companies with 100 people or uh and and look at the model companies you know like same same for them like why are they hiring so aggressively do you not think that the best people

32:34

SPEAKER_01

will want to work with the best technology and we'll see an increase in technology spend by the biggest companies in the world from 8 to 12 where it is today to maybe 16 to 20 yeah and then actually you'll see a reduction in headcount but an increase in technology spend and the best people want to go

32:49

SPEAKER_00

where they have the best tools and equipment i'm not sure about that either because i think technology technology technologies is actually not supposed to increase in cost first of all i think do you admit that the currently this technology is priced in it you know absurdly for what it delivers

33:08

SPEAKER_01

i think it totally depends on what it's doing for so no i don't at all for cursor or for any of the dev tools i think it's still dramatically underpriced when you look at mark benioff spending 300 million on anthropic it's 3.7 percent of developer salaries i think that's relatively small i i would say it's

33:27

SPEAKER_00

absurdly expensive i'll give an example we we had this we have a really cool um uh triage agent for engineering um you know and we have 15 people team on call team that was supposed to that their work was actually triage every single production issue that happens like any system alerts you know things that are going bad and and and we built this agent that actually now is taking care of like 95 of those issues automatically for them but even there is actually doing it a cost which is actually questionable like you know is it actually more effective that is more efficient than humans uh we were we were spending a million dollars a month on on that particular agent

34:14

SPEAKER_00

um the and that was actually more than the cost of a million a month yeah are you buying cristiano ronaldo what are you doing well i mean yeah costs are like that i mean there's it is it is quite

34:26

SPEAKER_01

expensive so the sorry you said you said because i i discussed this a lot on the show so you're making me much smarter you think that spending 3.8 percent of developer salaries on these tools is a lot if you think that's a lot then these model providers are absolutely screwed well i mean i think i think

34:49

SPEAKER_00

the point that i'm making is well the 3.8 percent number is actually doesn't seem high at all like when you look at it that way yeah but i also know that already you see with open source that you can do the same amount of same work for a tenth of the cost right that's number one um number two like historically uh for as far as i can remember you know we've not put technology cost and labor cost you know in the same sort of sentence ever before this is the first time we're actually hearing that that hey i would rather have fewer humans and more tokens the first time and i just feel like you know

35:29

SPEAKER_00

this is not how technology works like the models are supposed to get cheaper and cheaper the tech is

35:33

SPEAKER_01

going to be more and more affordable but but it's i'm so sorry this is so funny for me because you're you know the co-founder of clean and rubric and so who the fuck am i but a podcaster uh but this is exactly what technology is for this is agents being proactive having an opinion making a decision they should absolutely be included or put in the same sentence as labor because they are replacing the labor that we

35:57

SPEAKER_00

used to spend money on i think good technologies figure out how to make technology really cheap and it's gonna it's gonna happen here too that's that's my belief like you know you're gonna see it you you're gonna see inferencing costs um come down by orders of magnitude i think we saw something bizarre actually like the in the last six to nine months like every model actually increased their per token price and like if you if you go back you know 15 months everybody thought that the token like per token price is gonna just keep falling like it was before so so we don't know like you

36:35

SPEAKER_00

know what happened here like you know this is also sort of unique they needed to prove that they were good

36:40

SPEAKER_01

businesses before they went public that's what happened i can say things that you can't yeah yeah but my

36:48

SPEAKER_00

my bet is on ai getting much much cheaper than what it is today if ai gets much much cheaper than it is

36:54

SPEAKER_01

today these already loss making businesses which prop up our entire global economy pretty much at this point are very threatened yeah i mean like you know my take remains the same so okay so it's really interesting so you don't expect like an engineering team to get smaller in the future i i think per

37:18

SPEAKER_00

person productivity is going to shoot up but so will the demands to make the same amount of revenue you have to produce a 10x better product in the future unfortunately when you think about token

37:31

SPEAKER_01

spend internally how did you sit down and think about it as a team sitting with your cfo how did you go through the decision of how to think about token budgeting well i think we did probably what most

37:44

SPEAKER_00

companies did which is we didn't do anything so well i think like you know because we were in this phase of let people figure out what they can do with this tech and what did you see people went crazy people didn't adopt it what happened there's a power law like you know in our company and also at all of our customers you will see some people who spend ten thousand dollars or fifteen thousand dollars tokens every month and then you have others who are spending twenty dollars um one thing is interesting though that everybody has embraced ai to some degree like everybody is using the the basic you

38:19

SPEAKER_00

know as i mentioned before the number one application of our use case for ai today in the world is information seeking question answering and everybody's doing that so you see the entire like everybody on the team and our team as well as our customers they're all doing that everybody's asking questions everybody's getting some basic summarization information synthesis going but the advanced advanced use cases are limited to like five percent of the employee base is there anything you do as a

38:46

SPEAKER_01

leader to try and infuse ai as aggressively as possible we had nikesh from palo alto on every week he has a leadership meeting where he's like show and tell and everyone needs to stand up and show something that they've done with ai that week that it replaces what they do improves their job whatever

39:03

SPEAKER_00

it is is there anything that you can do yeah that's that's actually a really good idea like you know i've i've thought about doing that uh we limited the token maxing dashboards and i always thought that was not the right idea um to to just sort of you know reward people who are consuming more tokens um i feel i felt like you know we didn't need to do that you know we are a native ai company ourselves and people are already kind of um educated enough and they will use ai when they need to um but executives like you know sharing a success story uh we haven't sort of demanded it from every single exec every

39:42

SPEAKER_00

single week um but we have uh we have the showcase like in our town hall for example we'll always ask people to share those wins like every every town hall like you know there's a section dedicated to these are the new ai agents that teams are using to do work to work differently can i ask in terms of

39:59

SPEAKER_01

the execs and the people that you have i think recruiting has never been harder how hard is recruiting today with some of the largest model providers as we said paying just enormous salaries

40:13

SPEAKER_00

that we haven't seen before yeah um i i would actually say i maybe maybe like last two or three months let's let's let's put that aside for a minute i would say that uh recruiting was actually getting easier for um compared to the the sas peak wow why because i think like companies have been more um they haven't been growing their head count like if you look at the big the largest employers of tech talent uh many of them actually haven't been growing many of them have been laying off continuously i think about meta for example right like in every year there's significant layoffs and i don't know the overall

40:56

SPEAKER_00

head count my guess is that it's probably down from the peak of like 2021 or 2022 right um so actually there was more talent available in the market as such um than before but but if you now start to talk about okay ai talent ml talent um top people top people are sought after you know way more than ever before and and and and also the the pay scales have completely changed and not just from the model companies but even from startups because uh because you know startups are also like you know you are giving them too much money to to compete um you know for talent so like even even startups actually these days

41:36

SPEAKER_01

pay a lot we have to yeah yeah we have to because their alternatives are so large too and actually if it costs three four five hundred grand for a great dev yeah well the two million dollar seed round just doesn't go anywhere yeah for the founder building the team even if they don't take a salary if i'm gonna hire four people yeah i need six million bucks that's right do you think founders should

41:59

SPEAKER_00

raise large seed rounds i think it's better like i always prefer to raise as much you know for round as you can uh on on you know from the get-go what round felt the most highly priced so first actually we never actually went out to raise um um you know except for our first round of the company you know we always had somebody come in and it was a relationship that got built over some time and kind of became the de facto like you know that you know they are going to be the ones putting money in um i would say like our our city c probably felt the most um i guess you could say the most expensive

42:40

SPEAKER_00

because we barely had any business like definitely like you know sub two or three million dollars maybe five million dollars i don't remember exactly but then the valuation was north of a billion and so that was that was extreme um like and i think we but i guess we you know we you know we take what we get

42:59

SPEAKER_01

i mean that's incredible do you worry about scaling into that when you're doing it or would you just head down and think this is great a low dilution for a high price the way we thought about it more was that

43:10

SPEAKER_00

that was a that was a you know that was a statement to be made um to the prospective employees um more than anything else if we wanted to make the market understand that we're building something special

43:26

SPEAKER_01

and that kind of gives us that validation do employees give a who your investors are absolutely yeah i mean like you know a lot of founders are like you know what the best people don't care they're there for the mission they're there for and i'm always like i promise you if you have klyner or you have dst or you have sequoia ah great candidates suddenly want to talk to you a lot more

43:46

SPEAKER_00

yeah i mean like investor reputation directly impacts your reputation when you look today what have you changed your mind on most in the last 12 months i've always personally like my style has been a little bit too disciplined to be the right strategy anymore i get that feedback from my team that you know we are trying to be conservative we're trying to make sure that our capital goes a long way and and in that sort of in that mindset we may lose the land grab and so i i'm sort of feeling the pressure to change it uh myself like you know just change change how i think about like how we should

44:34

SPEAKER_00

be spending how we should be investing but at the same time like you know i have this fundamental belief that a business is always built on discipline and and and the real like you you have to charge for the product it has to generate value for the customers you know for every dollar that you invest in marketing there has to be some good return back from it um you're not assume that you just keep raising the money um to make up for all those things you know that were not that were not there

45:03

SPEAKER_01

do you agree with that when you have examples like uber which proves that a bad business model

45:08

SPEAKER_00

can turn good with scale yeah i mean that's that's what i'm saying that like you know that's that's the one where i feel that pressure that like you know perhaps my way of thinking is incorrect do you think it is a land grab we're absolutely in a land grab like you know no question like every single company in the world wants a product like ours today and either we get in today or it's going to be like

45:31

SPEAKER_01

10 times harder to actually get in the future we spoke about kind of job displacement we had an interesting uh conversation around that what job does not exist today that you think will be incredibly

45:42

SPEAKER_00

common in three to five years time well the comp comp comp comp comp roles will be um will be very common so like as for example you know somebody who can build a product um i don't know what to call them but they they can they act like engineers product managers designers uh similarly um in go to market um you know somebody who can sell the product and and they are capable of not only doing the business negotiations but they can actually demo the product they can actually talk about use cases and instead of having that segregation between account executives and you know solution engineers

46:28

SPEAKER_00

and then both sales solution architects i think we will see more and more generalization of roles like away from specialization and in fact i was trying to drive that very very hard even in our own

46:40

SPEAKER_01

company in our company but i'm sorry i mean this in a nice way the composite roles goes exactly against the idea of maintaining team size because if you have composite roles where you bring in four different

46:53

SPEAKER_00

specialities into one that is smaller teams it is yes but as i said like you know you have to do 10 times the work to get the same amount of revenue from your customers in the future you have a much smaller team to deliver the same amount of work that you used to deliver before you just are forced to do

47:10

SPEAKER_01

more got you okay and then what role do we have today will we not have what do we look at and go oh

47:18

SPEAKER_00

my gosh i can't believe we used to do that a lot of analyst roles like or like the the data analyst roles which are not business thinkers you know they were given it asked a day like i need to see this data and then they produced like they sort of go and build those specific dashboards configure back-end systems i think like that that kind of work definitely goes away um the uh i think business intelligence is going to be very different um business owners will directly be able to get answers to their questions so so business and business analysts like you know data analysts you know that's

47:56

SPEAKER_00

sort of one um many hr roles sources for examples so sort of like in recruiting that's the role that i think uh is going to definitely get consumed in like you know into like a full cycle recruiting role i do have

48:11

SPEAKER_01

to ask one final one which is we're sitting here in europe and it brings about a question of sovereignty the us and europe bluntly have not come up to muster so to speak on open source yeah do you think we will have a world of sovereign models and do you think given what we've seen in the last month or so that we

48:31

SPEAKER_00

need to have sovereignty over our models the desire for sovereign models like is strong and it's actually i would say like it was probably stronger uh a year back compared to now at least like you know i feel like i'm hearing less of it to someday you know like the there was a period where every every nation thought that they could build one um they were then ai was still in its early stages but then like a lot of those uh nations actually figured out that you know this you know that's not going to be the way and so they're okay with uh letting you know their enterprises within their own countries you know use

49:14

SPEAKER_00

openai or entropic or all the other models so so i don't i'm not an expert i don't know like you know whether this trend is on the on the rise or or sort of like you know on the fall a little bit i think

49:29

SPEAKER_01

it's unequivocally on the rise given what we saw with the trump administration banning you know anthropics latest models yeah and it's understanding from a lot of especially europeans that we cannot rely on a us individual yeah who could ban our access to intelligence yeah but but where are the results from it i mean that was a month ago so i think to expect a stand-up model within three weeks would

49:53

SPEAKER_00

be tough yeah yeah yeah but but like you know even before that i think the it just hasn't happened right like you know the only country in the world you know that has produced models outside of us is china yeah and then of course maybe a little bit in like you know france with mr is that simply an

50:09

SPEAKER_01

incentive problem the lack of open source community in the us and why we don't have any us open source to

50:16

SPEAKER_00

a real degree in substantiveness no i think the like there is good good open source community in the us in many other areas well i mean what open model from the us no there's no yeah you're right that we don't have open models but it's not because open source as a as it as a movement as it all you know as a concept is weak in us it actually work quite strong it's probably if you think about models the um they require a lot of upfront investment which is not open source friendly in many ways a lot of open source software has been skunk works like developers is getting no funding associated with

50:59

SPEAKER_00

them and they still get something built they couldn't build models that way and so that's why like you know naturally this thing didn't work out um and and you you know you need these techniques you know where like super high investment is not needed um um and the do you worry then when you look at the state you

51:22

SPEAKER_01

know i i i spend a lot of time on open router and i see the model usage and traffic and like you know anthropic today was first us model and seventh the first six were chinese do we just like you have bucket who cares that the ccp are funding the top six models the fact that you know you can actually run

51:42

SPEAKER_00

inferencing on those like you know in in that contained environment makes people feel comfortable and but i don't think you know that's like as a um as us you know like you know us won't feel absolutely won't feel okay with you know that trend there's there's good work that's happening now to actually promote open source and model development in the us there's some models coming out the alternative

52:06

SPEAKER_01

is that sam and open ai give five percent to trump and then he puts regulatory regulatory capture on anthropic and open ai and puts attacks on open source well i hope not you know that i that's uh

52:21

SPEAKER_00

i i i i doubt that that's that's gonna happen yeah why would i'm so sorry i'm only why else would

52:27

SPEAKER_01

sam give them five percent it's a quid pro quo i need you you need me well i mean i guess i just

52:33

SPEAKER_00

believe more in the u.s system and the ultimately curbing uh i don't think right now by the way you need to curb open source like you know it's too far behind in the us you don't think sam and dario

52:49

SPEAKER_01

are sitting going oh wow we underestimated this and this is a core threat to our business i mean like

52:54

SPEAKER_00

they probably are thinking that but i don't think uh they can fix that by by through regulation you

53:03

SPEAKER_01

don't think that sam will be calling up trump but who he has a direct line to saying hey the ccp are funding your biggest our biggest competitors and we cannot promise that there isn't a backdoor to xi jinping you need to stop this and i'll give you five percent for your troubles well isn't the

53:20

SPEAKER_00

argument the other way around like you know that right now there are all these open source models which are very good and they're all built in china and us needs to build its own like you know we use can't be seen as a as a country that doesn't innovate on technology so it's actually paramount for

53:35

SPEAKER_01

the us to build well i think sam will be saying it takes billions of dollars and years of time trump defend america and support open ai and anthropic and put barriers up to prevent chinese

53:48

SPEAKER_00

which is open models from getting adoption taxes bans those those maybe yes but the um but us open source models that they are going to have a lot of tailwinds and they have to like you know this is a known accepted uh like issue that every technologist in bay area you know talks about there's a lot of lot of motivated parties that actually want to promote like including nvidia for example you know they're putting a lot of investment in promoting like you know development of great open source models in

54:23

SPEAKER_01

the us and i hope they succeed absolutely uh a multi-model world is important for all of us uh listen i'm doing a quick fire round with you so i say a short statement you give me your immediate thoughts does that sound okay okay what's your biggest advice to someone studying computer science today it's fine

54:40

SPEAKER_00

to study it don't don't don't be uh don't get too worried because what other people are telling you

54:45

SPEAKER_01

which legacy company has adopted ai the best do you think well are you willing to call google a legacy

54:52

SPEAKER_00

company yeah yeah so google probably rates higher than anybody else in terms of not only embracing ai internally but also uh in their like you know launching products but there i guess they are ai companies so it's kind of hard it's unfair to uh to put them in that category you start a new company

55:13

SPEAKER_01

and you can only take one ambassador who do you take with you well i i think i'll take one of one of our

55:19

SPEAKER_00

existing ones we create relationships with all of them which one would you take i don't know i won't answer that question i just don't have the answer really i get to think about i think it's probably circumstantial depending on like you know what what i'm doing you know different people bring different

55:33

SPEAKER_01

strengths what would you most like to change about the startup ecosystem that we see today i actually do

55:41

SPEAKER_00

think that you know there is um too much capital available today in the for startups and it's actually sometimes creating failure paths for people i think they're not getting what it takes to build a great company i like i'll give you an example like a startup that has raised a seed around decides to pay half a million dollars to an engineer like you were saying before um and this is happening today and the startup founder is okay with it the investors are okay with it but it's just surely not a sustainable path to actually win and they're paying it while google is not because google knows that they don't need to

56:24

SPEAKER_00

actually um uh buy talent like that so so i think that is one thing that i feel um this overabundance of capital is getting startups to sort of create structures which are not going to be sustainable for

56:39

SPEAKER_01

them do you worry about the lack of exit options that are now becoming more and more real what i mean by that is like honestly if you don't have a billion in revenue today it's hard to go public tech acquirers your big companies are very specific about what they want to buy p licking its wounds from having a portfolio that's full of medallias it's a tough landscape like startups have never been easy

57:06

SPEAKER_00

like i think in fact you know for for all for all i would say in in the last 25 years that i've seen i would say it's been it's easier to build a startup and get a good exit from it these days than it used to be in the past like start is a brutal it's brutal game what does no one know about being a founder and ceo from the outside that they should know that it's not a uh a sexy job like it's actually one of the most stressful um things and you really have to be crazy i think

57:46

SPEAKER_01

i think they know that now i think one for me is that you have to consistently be unhappy you should never be happy i think as a ceo because there's always something that needs doing could be done better telling someone you will never be happy is something that i like jarred by yes that's that's

58:06

SPEAKER_00

that's a good one this is this is a this is a tough job all around and i think like oftentimes you know people like who have not done it um they feel that there's a lot of glamour they feel that you know this is going to make a lot of money and and their life will be fantastic they're going to have a lot of respect and i think like almost all of those things you know are irrelevant you know this is a you have to be truly mission oriented for to be to to survive you know as a founder did your style

58:35

SPEAKER_01

change with money you've been successful before i think better i think founders are better and investors are better when they are already rich if i'm being blunt i think you make more rational sound decisions that are not made with economic impatience i think for me maybe not like the

58:57

SPEAKER_00

but but at the same time you know like i've i've i'm a man with minimal needs and my needs are already met like a long time back so so i guess i've i've definitely built these startups without like you know that worry of you know can i feed my family um so yeah like maybe maybe that has helped me but like as i've seen more success it doesn't change me fundamentally in terms of like i still you still have you know you have to have that drive you have to work uh continuously to work you know more than every other person in your company uh lead by example and keep pushing and you have to have

59:34

SPEAKER_00

this rational you know need to to make something big happen i so appreciate your time i apologize for

59:42

SPEAKER_01

being robust in my discussion back yeah i think it was a different interview to a lot of interviews that you do where uh it was more discursive but i so appreciate the time and you've been fantastic dude

59:56

SPEAKER_00

yeah thank you yeah

Reading tools

Type to find a passage

Appearance
Ask this transcript

Add a note