MC.43: What Investors See in Devin That We Don’t — A $2B Question

Hey everyone,

I watched the talk from Scott Wu at the AI Engineer World’s Fair about the making of Devin, a software engineering agent valued at $2 billion. I have to say, it got me thinking...

Devin’s Resources Map

So, Scott Wu shows off the resources map for Devin and something really struck me. The core feature - the one that's supposed to solve software engineering - is this tiny blue rectangle surrounded by a bunch of other stuff. Take a look:

You'd think most of the resources would be about solving and planning code engineering tasks. But the map shows mostly user-facing features and reimplementations of existing tools. So why are investors hyped about this?

Breaking down Devin’s Features

Product Features

Let’s break it down:

• Fork, Rollback, and Resume: Basically, it's git.

• Interactive Browser, Shell, and Editor: Free stuff with lots of options already out there.

The interesting part is the tools to verify Devin’s Work. Continuous evaluation of an AI system is crucial for success in production. But what's actually in this block? Automated tests? Static analysis? An integrated code review process? All of the above? It's a key part, but I couldn’t find a lot of details apart from the history of commands used by Devin on a session and a global diff.

Overall it feels a lot like reinventing the wheel and/or just wrapping existing tools. Nothing to be super excited about.

Devin’s Tools: AI using human tools?

Devin has access to a shell, an editor and a browser. I get the shell and browser part, but a custom editor? Seems overkill when LLMs output whole files or edits.

Maybe they’re using “editor” as a name for a smart pipeline with languages servers, tree-sitters and the likes to provide a smarter context and a better feedback loop?

It's not super clear how it helps and feels like a marketing gimmick: anthropomorphizing their AI agent to make it look smarter than it is. It has a name and uses the tools you use for development, just as our imagination pictures it.

Again this part is not exciting.

Machines, Knowledge and Playbooks

They built an infrastructure similar to platforms like Replit so their AI agents can develop on remote code box. You can access the machine to run commands to help your AI if it’s stuck.

Besides infrastructure, they have a RAG (retrieval augmented generation) system for knowledge and memory. And then there are these "Playbooks" - kind of like instructions to follow. Not sure why this isn't just part of the knowledge block.

User Interface and communications

They've got integrations with Slack, GitHub, and their own mobile/desktop apps. Nothing groundbreaking, but it shows they want control over the user experience. That's a counterpoint to other players like Fumedev, which chose to rely on GitHub and GitLab's pull request system to interact with an AI agent.

So, what makes Devin so special to investors?

Here’s my wild theory:

The value is not in the AI agentic part. I think it’s a big marketing bet with a long term vision to capture and expand the existing market of hosting:

1. People are dreaming of AI building software for them.

2. Cognition AI (Devin’s creator) wants to be THE brand for this dream.

3. They’re probably faking it till they make it: build UX and infrastructure while slowly showing off capabilities.

4. If (when?) the AI gets good enough, developers and non-techies will flock to build apps faster.

5. Once they have the development environments, hosting services are the next logical step.

6. Boom! They’ve just sneaked into the hosting market under everyone’s noses by creating their own space.

Conclusion

It's pure imagination, but this is the only explanation I have for the funding they got. Unless they're hiding something exceptional, the current tech isn't that impressive, and there's lots of competition from big players like Microsoft and open source projects which share similar benchmark results. So it must be a business play more than a technological shift.

This post is not a rant against the overhyped side of AI, it’s about my understanding of their valuation. I'm happy they work hard to fulfill the dream of many where a software engineer AI agent drafts 80% of the work. I hope they'll get there and share their results with the world.

Let me know your thoughts,

Sam

Enjoyed this newsletter? Forward it to a friend and have them signup here.

Until next Thursday 🎉