# Agents for Everything Else

swyx, AI Engineer | AI Engineer Europe 2026 | 14:10

Source: https://www.youtube.com/watch?v=zepu8Kk6FBQ
Channel: AI Engineer (https://www.youtube.com/@aiDotEngineer). Summarised by AIE Talks.
Page: https://aietalks.com/talks/agents-for-everything-else
Published: 2026-05-01
Tags: agents, coding-agents, mcp, team-adoption

## TL;DR
- Agents helped AI Engineer's nine-person team run conference operations that previously required more manual coordination.
- The main gain was that employees could act on ideas immediately, which led them to do more design, polish, and experimentation.
- Coding agents are spreading into data management, research, purchasing, and other knowledge work, so teams need to rethink their products around agent users.

## Summary
swyx explains how AI Engineer uses agents to run its conferences with a nine-person full-time team. The first success was connecting Cognition's Devin to Figma and turning designs into a working website. The larger change came afterward. Designers could annotate work, test ideas, add animations, and polish pages without waiting for a developer or for swyx to unblock them. Agents also manage conference data, update the schedule from forwarded emails or screenshots, handle ETL tasks, research purchases, and turn notes into organized documents. swyx argues that agents are useful because they remove dependency work and allow people to work in parallel. He also describes moving conference data out of a CMS and keeping it in code that an agent can manage. He is cautious about replacing SaaS, since employees must live with mistakes, and recommends reducing their main concerns one by one. He expects agents to become primary users of software, making APIs, CLIs, and MCPs more important than human dashboards.

## Key ideas
### Agents remove the dependency work that slows ordinary tasks
[04:19](https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=259s)
swyx says one underappreciated benefit of agents is that they save people from "random yak shaving." Instead of repeatedly discovering that one dependency must be handled before another, an agent can crawl the dependency tree and deal with installation or Python dependency problems. He also says productivity models need to account for parallel work, since autonomy alone does not capture the benefit. This changes the practical value of an agent before it produces any visible feature. It can clear the setup work that otherwise blocks a person from starting.

### A coding agent let a designer's Figma page become a working website
[03:42](https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=222s)
The first major workflow change came when swyx connected Devin to Figma. A designer could show a Figma page that might previously have taken one, two, or four weeks to implement. After a coworker handled the Figma integration, Devin produced a functioning website that was pixel-perfect to the design. swyx says he was surprised because he had not experienced that workflow before. The designer was happy with the result, and the site shown in the talk was the live AI Engineer website.

### Agents gave nontechnical employees room to act on their own ideas
[05:25](https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=325s)
After the first success, swyx's designer in Indonesia began working with Devin while swyx was asleep. The designer used red-line annotations, a communication style swyx had not taught him. The team also started making things it would normally have skipped, such as an Easter egg, a visual treatment inspired by a viral tweet, animations, and extra polish. The reason was simple: the feedback loop no longer depended on swyx or a contracted developer. Employees could have an idea and try it immediately, which made the work more enjoyable and produced work swyx had not previously received from the team.

### The biggest gain was more output from people, rather than more generated code
[07:25](https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=445s)
swyx deliberately moves away from talking about coding-agent output or lines of code. His claim is that agents increase the productivity of the humans around them. When employees do not have to wait for a manager or developer, they can explore more ideas and finish more small improvements. This is why he calls the theme "agents for everything else." The agent is part of a broader workflow that changes what the team chooses to do, not only how quickly it writes software.

### Conference operations can be managed from code and forwarded requests
[08:00](https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=480s)
AI Engineer treats its conference as a data-management problem involving speakers, sponsors, and attendees. swyx says the important unlock was discarding the CMS and committing the data to code as the source of truth. Devin manages the schedule there. When a speaker sends a change by email, swyx can forward the message or paste a screenshot and ask Devin to handle it. He connects this workflow to the team's ability to run a thousand-person conference with nine people, with a larger San Francisco event planned while keeping the team the same size.

### Agents can handle routine integrations and real-world purchasing research
[09:18](https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=558s)
The team uses agents for ETL work when an external vendor holds data that needs to be brought into a central source of truth. The agent can help obtain an API key and sync the systems. swyx also describes asking Devin where to buy a lobster in London after seeing a viral idea involving a claw. Devin returned phone numbers, email addresses, and websites, and the lobster at the event was bought through that research. His point is that an agent with web access and a capable model can take on knowledge work that might otherwise go to an executive assistant or junior employee.

### Teams should reduce employee concerns before replacing SaaS tools
[11:14](https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=674s)
swyx is exploring whether agents can replace whole SaaS products, including tools his team currently debates removing. He admits that managers have to bring employees along without dismissing their concerns. Those concerns are valid because employees will deal with the consequences when an internally built replacement fails. His proposed method is to identify the top three concerns and reduce them systematically. The talk presents this as an active process, with mistakes expected rather than hidden.

### Software needs to account for agents as primary users
[12:46](https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=766s)
swyx connects his experience running AI Engineer to changes in product design. He cites Malte's opening keynote saying that 60% of Vercel's user base is now bots or agents rather than humans. He says dashboards matter less when agents are the users, while APIs, CLIs, and MCPs matter more. He also points to MCP apps and the idea that custom interfaces may be replaced by interfaces delivered inside another application. His own behavior has changed accordingly: he no longer cares about the Figma dashboard and sends work into Cloud Co-work instead.

## Notable quotes
- "Anytime there's random yak shaving, I think one underappreciated benefit of agents is that they save you the yak shaves." (04:19)
- "I'm getting more productivity out of my humans." (07:25)
- "I might have had 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." (10:30)
- "Your dashboards don't matter, your APIs matter, your CLIs matter, your MCPs matter." (12:46)
- "Agents for everything else are coming. Wake up. Use it. Bring it home to work." (13:27)

## Tools & references mentioned
- Cognition
- Devin
- Coworker
- Figma
- React
- Supabase
- Tido
- Google Sheets
- Sessionize
- Sanity
- TLDraw
- Steve Ruiz
- Town Assistant
- Apple Notes
- Notion
- Cloud Co-work
- MCP
- Vercel
- Malte
- Edo
- Liad
- Andre Karpathy
- ETM

## Who should watch
- You run a small operations or events team and want examples of agents handling coordination, data updates, and research.
- Your employees wait on developers or managers before they can try small ideas, and you want to see how agents change that feedback loop.
- You build software for people who increasingly work through agents, APIs, CLIs, or MCPs instead of a product dashboard.

## Related talks

- [Agentic Engineering: Working With AI, Not Just Using It](https://aietalks.com/talks/agentic-engineering-working-with-ai-not-just-using-it) (Brendan O'Leary, Kilo Code, 27:03)
- [Software Development Agents: What Works and What Doesn't](https://aietalks.com/talks/software-development-agents-what-works-and-what-doesnt) (Robert Brennan, OpenHands, 16:46)
- [Building an Autonomous Engineering Org](https://aietalks.com/talks/building-an-autonomous-engineering-org) (Angie Jones, Agentic AI Foundation, 17:36)
- [Collaborative AI Engineering: One Dev, Two Dozen Agents, Zero Alignment](https://aietalks.com/talks/collaborative-ai-engineering-one-dev-two-dozen-agents-zero-alignment) (Maggie Appleton, GitHub, 17:43)
- [Dispatch from the Future: Building an AI-Native Company](https://aietalks.com/talks/dispatch-from-the-future-building-an-ai-native-company) (Dan Shipper, Every, 17:58)