AI Engineer

Agents for Everything Else — swyx

935 summary words 4 min summary Watch video

Start with the signal

4 min read

Summary

30-second take

Swyx (co-founder of AI Engineer conferences) shares how a 9-person team runs multi-million dollar conferences using coding agents (primarily Devin) not for lines of code but for "everything else"—design implementation, data management, research, even buying a lobster. His core thesis: agents remove yak-shaving blockers, enable non-technical teammates to ship faster (because it's fun), and unlock parallelism that traditional productivity models miss. The talk is less about agent capabilities and more about organizational workflow transformation: employees enjoy work more when feedback loops disappear, and "agent experience" (APIs/CLIs over dashboards) is becoming primary. He's openly arguing with his team about replacing SaaS tools entirely. This is early-adopter conviction meeting real operational friction.

Key takes

  • Non-technical employees are more productive and engaged when agents remove waiting/blocking on engineers. Swyx's designer in Indonesia started annotating Figma screenshots with red lines to prompt Devin, shipping animations and polish work that never happened before—because the feedback loop is instant and "fun."
  • Agents excel at parallelizing yak-shaving (dependency hell, data syncing, research) more than raw code output. Traditional productivity metrics undervalue the cognitive load saved when agents crawl dependency trees or handle boring ETL from external vendor APIs.
  • Coding agents are "breaking containment" beyond code into knowledge work. Swyx used Devin to research/buy a lobster for an event stunt, sync speaker schedules from email screenshots, and replace CMS tools with code-as-source-of-truth managed by agents. The unlock: serverless on-demand knowledge work that would've required EAs or junior employees.
  • Cost curve deflation (100x per 12-18 months) makes "replace SaaS with bespoke agent workflows" economically viable. Swyx is actively debating with his team about kicking out SaaS tools they can now build/maintain themselves with agents—though he acknowledges employee concerns are valid because they deal with the consequences when it breaks.
  • The primary user of software is shifting from humans to agents. Vercel's dashboards now serve 60% bots; MCP/CLI/API matter more than custom UIs. This is the "agent experience" design trend Swyx flags as top-3 for 2026.

Useful details

  • AI Engineer conferences: 9 full-time people, $9M+ revenue (meets "tiny team" definition: more millions than employees).
  • Stack: Figma, React, Supabase, Tidal, Google Sheets, Sessionize—all non-AI initially.
  • Workflow examples:
  • Designer shows Figma → Swyx adds Devin to thread → pixel-perfect website shipped (AI.engineer live site today).
  • 207-reply Slack thread from Devin iterating on a throwaway Easter egg feature (ultra-wide screen hover animation) because designer found it fun.
  • Speaker schedule managed entirely by Devin: forward email screenshot, agent syncs changes to code-based CMS.
  • Tweet about viral design aesthetic → 127-reply iteration → shipped feature.
  • Bought conference lobster prop by asking Devin to research London suppliers, returned phone numbers/emails.
  • Tools mentioned: Devin (primary), Town (knowledge management agent for wikis), Cowork (used to hook Devin to Figma API).
  • Predicted 6,000-person SF conference with same 9-person team size.
  • "AGI pills" gag: physical pills handed out at conference to "prescribe" to insufficiently AGI-pilled coworkers.
  • Key pattern: ditching Sanity CMS → code as source of truth → agents manage the code.

Caveats / counterpoints

  • Swyx explicitly says "we do get it wrong" when pushing to replace SaaS tools, and advises not to ignore employee concerns—they're the ones who handle the fallout.
  • No hard data on error rates, rollback frequency, or quality control mechanisms when agents manage critical conference infrastructure (speaker schedules, vendor data).
  • The talk is heavily skewed toward operational/non-coding tasks; little discussion of agent limitations in complex reasoning, ambiguous requirements, or edge cases.
  • All examples are from a small, technical, high-conviction team. Unclear how this generalizes to larger orgs with compliance/security/review processes.
  • Vercel's "60% bots" stat is mentioned without context on what those bots are doing (CI/CD deploys? actual autonomous agents? scrapers?).
  • No mention of cost at scale—Devin credits, API costs, or whether the SaaS replacement math actually pencils out beyond the initial hype phase.

Ken relevance

High relevance. This maps directly to Ken's thesis on AI ops and agent-first workflows:

  • GTM/operations angle: Swyx's "replace SaaS with agents" playbook is exactly the wedge for Ken's Agentic OS vision—small teams can now compete with larger orgs by automating knowledge work that previously required headcount.
  • Content/community: The "AGI pill" gag and conference storytelling are examples of high-leverage brand-building that Ken could replicate in Backdrop or future ventures.
  • Investing: Swyx's point about "agent experience" (APIs/CLIs over dashboards) is a design shift Ken should watch in portfolio companies—tools that don't expose MCP/API will lose to those that do.
  • Personal workflow: Ken's assistant agents could adopt the "code-as-CMS" pattern for managing data pipelines, research queues, or deal flow. The "forward email screenshot → agent handles it" interaction is replicable.
  • Trend alert: "Coding agents breaking containment" into knowledge work and the 60% bot users stat both signal that agent-first infrastructure (not just LLM wrappers) is the next competitive moat.

Caution: Swyx's team is extreme early adopters with high technical fluency. Ken should discount some of the ease-of-adoption claims when thinking about enterprise customers or less technical teams.

Watch verdict

Watch fully. Swyx is a signal-rich operator with skin in the game. His concrete workflow examples (email-to-schedule-sync, Figma-to-code, lobster research) are more valuable than abstract agent theory, and his framing of "agents for everything else" is the exact mental model Ken needs for operationalizing AI in non-engineering contexts. The employee engagement angle (work is more fun → more output) is underexplored in most agent discourse and relevant for Ken's talent/team-building strategy. The caveats he does share (employee pushback, things breaking) are honest and useful.

Search this video Resume transcript at 0%
Full transcript 2396 words · 11 min read
0:15

SPEAKER_01

The future of work has many paths. Our next presenter will discuss the path that he walked with Devin as he organized this very conference. Please join me in welcoming to the stage the co-founder of AI Engineer Conferences, SWIX.

0:35

SPEAKER_00

Hi everyone. I am not the Chief AI Officer of the UK. Unfortunately, he had to leave for a personal reason. But you get me. Thanks for staying so long. I hope everyone is having a good time. Thank you.

0:53

SPEAKER_00

It's so endearing and heartwarming to hear from you guys. I'll take you a little bit into how we build AI Engineer with AI. And it's probably the biggest revelation that I've had. So we've had a lot of really warm reception from you guys, and I think it's really great. And I think this is something that we really try to engineer. And hopefully, this is our first event in London. Hopefully, you have us back next year. One thing I wanted to, for those who are newer to us, I do one of these keynotes every single AI. The very first one, three years ago, I talked about the productivity gain that you get from the increased usage of AI.

1:35

SPEAKER_00

And the second one, we talked about how you should just use more AI because the cost curve of AI is going down roughly 100 times per every 12 to 18 months. And I think it's still continuing to trend that way. The third year, we started to talk about tiny teams, which was this definition that I had that teams with more millions in revenue than number of employees. And I even curated an entire track at the World's Fair about this, where we summarized it as the tiny teams' playbook if you're interested in building that. The reason I like this emphasis is because I think people are maybe too egotistical about looking for the one-person billionaire or unicorn founder.

2:18

SPEAKER_00

Every company can have a tiny team, whether you're small or large. And I think when I look at how we run AI Engineer, me being the leadership of Ben Lee and myself, we're also a tiny team. This is us. It's just nine full-time people. And we're running a business that is more than $9 million, so we are a tiny team. And I wanted to show you the most significant changes in our workflow since we started this three years ago. By the way, this is our taking the AGI pill moment. Did you guys get the AGI pills? Yes. Very proud of this. This is my brainchild. If one of your coworkers is not sufficiently AGI-pilled, you should prescribe one of these.

3:03

SPEAKER_01

You're all AGI doctors now.

3:07

SPEAKER_00

Okay. Our stack was very stable and completely non-AI, which is very ironic for an AI conference. We do Figma, React, Superbase, Tidal, Google Sheets, Sessionize. And then I had this funny, weird moment where I joined Cognition and started using coding agents seriously at work, mostly because they were free. And I started adding it to the company Slack. And then I started doing things with it and showing people, hey, here's how you use it to do coding on the company website. All well and good. And something strange starts happening. I start introducing... This is a workflow of our contract designer now full-time, showing me a Figma page and asking me to go through it,

3:54

SPEAKER_00

and expecting that we would take a week, two weeks, four weeks to turn it into reality. I just added Devin to it. And ultimately, before I had to add Devin to it, I had to hook up Devin to Figma. And I'm not going to be doing that. So Cowork is doing it for me. You should use Cowork for doing this. And which, by the way, leads me to my first lesson, which is anytime there's random yak shaving, I think one underappreciated benefit of agents is that they save you the yak shaves. All the dependency tree crawling of oh, no, I have to do that first. Oh, no, I have to do that first. Particularly when it comes to installing dependencies or fixing Python dependencies.

4:33

SPEAKER_00

Fantastic for that. And I think a model of productivity that doesn't sufficiently appreciate parallelism and not just autonomy, and depth of the yak shaving is not fully capturing the benefit of agents. So, anyway, back to the agent story. Hooked up Devin to Figma. And we, in very short order, we have a perfectly functioning website that is pixel perfect to the Figma. And to me, that was a surprise because I'd never done it before. You know, you always mistrust marketing until you see it for yourself. And more importantly, our designer is very happy about it. And that's basically the website that you see live today when you go to AI.engineer.

5:13

SPEAKER_00

The other interesting thing that happened was then we started using it more, right? After one initial success, you start using it more. Something that you can't see because it's very small text, but I'm going to highlight for you, is that that is 207 replies just exploding in usage. What the hell? And when you dig into it, it's very interesting, right? So, first of all, I start kicking off some work and then I go to bed. And then my designer who is in Indonesia wakes up and starts messing with Devin. He starts prompting Devin with red lines on annotations, which is something that Steve Ruiz, one of our speakers from yesterday, does with tldraw.

5:49

SPEAKER_00

And I never taught him to do this. And there's no instruction manual. It was just how would you communicate with another human being. And I work mostly with a non-technical team and I think that's very important that they need to be comfortable with agents. And I think they're finally at the point that they are. We start working on things that we would never normally have worked on. Nobody has reported this, so I assume none of you have discovered it, but there's an Easter egg on the website. Why? Because I put it there. Why? Because it was fun. Because I could, right? So if you're on an ultra-wide, you scan your mouse over the highlights, you'll see an Easter egg.

6:23

SPEAKER_00

I saw a tweet that was viral about a design aesthetic that I liked. I threw it into Devin. Out pops. And then, you know, 127 replies later. Literally, I popped it in. I was thinking, let's just see what the client group will do for me. I don't want to waste my designer's time. I just want to see what Devin does for me. Designer jumps in. Why? Because it was fun. Because I could, right? So if you're on an ultra-wide, you scan your mouse over the highlights, you'll see an Easter egg. I saw a tweet that was viral about a design aesthetic that I liked. I threw it into DevIn. Out pops. And then 127 replies later. Literally, I popped it in.

7:16

SPEAKER_00

I was like, let's just see what the client group will do for me. I don't want to waste my designer's time. I just want to see what Clanker Doof does for me. Designer jumps in. And does, and does, and actually starts working on this thing which I thought was throwaway and fun. And the most interesting thing is so small I can't even read it. I'm so sorry for this. So the reason he starts working on it, even though it's a throwaway project, is because it's fun. And I think that's something that was a big aha moment for me. I am getting more work out of my employees because they enjoy doing it.

7:51

SPEAKER_00

Because the feedback cycle for them from waiting blocking on me or a contract developer that we have is gone. They just literally have the idea, they go do it. Right? And they're doing more things. They're doing animations, they're doing polish. Things that we've just—I'm getting work that I've never gotten out of my employees before. I think that's something that you should appreciate too. If you haven't noticed, I'm no longer talking about agents for coding or how many lines of code I'm producing. I'm getting more productivity out of my humans.

8:22

SPEAKER_00

And I think this is something that is a major theme for this year that I'm really trying to investigate: agents for everything else. Then obviously, I had the success with Figma to website. I had the success with Tweet to website. What else? Right? You start to think about other use cases. This whole conference is a giant data management problem. I have to sync with 130 speakers and a couple dozen sponsors and all the attendees that come in with various needs. And really, it's just the CMS, right? We've messed with Sanity. I'm not the biggest fan of Sanity in the world because I want to keep some sanity to myself.

9:10

SPEAKER_00

But I can throw in spreadsheets and Devin can manage that for me. And I think the unlock happened when I threw away the CMS and just committed that to code, but use that code as my source of truth and let Devin, whatever coding agent you use, start to manage it. And so this entire schedule is managed by Devin. What does that mean? It means that whenever someone comes in with a speaker change, for example, Marder, one of the speakers from today, sends in an email, I just say, Devin, handle it for me, right? No other further communication is needed. I can just forward the email. I can paste the screenshot, whatever.

9:40

SPEAKER_00

And that kind of volume—that's us as a small team of nine people managing a thousand person conference, right? We're going to manage 6,000 people in San Francisco this fall, this summer. And I'm pretty sure we can stay the same size. It is incredible the amount of productivity that you can get once you're sufficiently onboarded and you have the workflows hired out. We have agents for ETL. We deal with an external vendor system that has data that we don't have in a central source of truth. So I need to get the API key to sync over data and make sure there's a single source of truth. These are very boring routine tasks.

10:08

SPEAKER_00

There's another fun story that I can tell you: agents for buying. So I saw this viral tweet about how somebody put a claw in Wall Street next to the Wall Street bull. And I was thinking, oh, that's funny. We should put a claw in front of our conference. And that's exactly what happened. And so I asked Devin to research where I can get a lobster in London. Devin comes back with phone numbers and email addresses and websites. And I just click through and think about it and ask for more research. And I'll pop this guy. That's literally the lobster that you had was bought from Devin.

10:48

SPEAKER_00

And I think this kind of personal automation for everything else—it just matters that you have an agent that has web access, that has a smart enough model. So this is effectively a claw, right? An open claw, nano claw, whatever you call it. It doesn't really matter. It matters that you're using agents for things that you would otherwise have spent knowledge work on. I might have had an executive assistant. I might have had a junior employee do these things for me. But now I can do it serverless on demand with a coding agent. I'm not here to only show Devin. I just advise with the company now. But I started exploring town.

11:30

SPEAKER_00

Because I think what's happening here is coding agents are breaking containment, right? There's all these other more fit for purpose knowledge management tools, like the wikis that Andre Kaprati is talking about, that OpenClaw is now adopting as well. You're going to see an explosion of this this year. This is probably in the top three to five trends of 2026 that I want to alert you to. So here is me managing the World's Fair this summer. Here are all the tracks I'm planning. Here's my Apple Notes. On the left is my Apple Notes of all the people. It's intentionally small.

12:34

SPEAKER_00

And I threw it into Town and out pops a nicely formatted Notion doc with research on all the speakers that I intend to solicit and think about curating. And then obviously once you get enough conviction, you are thinking about replacing entire pieces of SaaS. Here's me arguing with my employees about kicking out a SaaS tool and building it ourselves because we can. So I clearly have the most conviction. I think one of the annoying things is if you are in a position of power management and you deal with employees who are not as much in conviction, try to bring them along the journey and not talk down or ignore their concerns, right?

13:00

SPEAKER_00

Because they are very valid concerns because they are exactly the people that will have to deal with your bullshit when you get it wrong. And we do get it wrong. And then obviously once you get enough psychosis, you are thinking about replacing entire pieces of SaaS. Here's me arguing with my employees about kicking out a SaaS tool and building it ourselves because we can. So I clearly have the most psychosis. I think one of the annoying things is if you are in a position of power management to deal with employees who are not as much in psychosis and try to bring them along the journey and not talk down or ignore their concerns, right?

13:17

SPEAKER_00

Because they are very valid concerns because they are exactly the people that will have to deal with your bullshit when you get it wrong. And we do get it wrong. So one method I am approaching the AI replacing SaaS concept, which I think should be relevant for a lot of you, is well, let's identify the top three concerns and let's systematically reduce them. And that's the process that we are going through right now. So I just wanted to give you a little bit of that taste of how AI is changing our business as managing the conference. It's come really a long way. It's a consistent theme I'm seeing even among our speakers.

13:41

SPEAKER_00

This is Malta opening keynote talking about how 60% of the user base of Vercel now is bots, is agents. It's not humans. So actually your dashboards don't matter. Your APIs matter. Your CLIs matter. Your MCPs matter. Here's the MCP apps guys, Ido and Liad, who spoke today, speaking on ETN, about how basically your custom UI is going away. You should ship UI to somebody else's app. And I think this pattern of how your primary user is changing is really shifting towards what people are calling agent experience.

Reading tools

Type to find a passage

Appearance
Ask this transcript

Notebook

Export Markdown

Add a note