Who Actually Bought Windsurf? 🌀

Three companies walk into a bar. They all claim they bought the same startup.

Right. So here's the week in AI coding acquisitions, and I need you to stay with me because it gets properly daft.

Monday: OpenAI's $3 billion bid for Windsurf collapses after Anthropic yanks their Claude API access. Brutal move, that. Like cutting off your rival's electricity mid-negotiation.

Also Monday: Google DeepMind swoops in and hires Windsurf's CEO Varun Mohan, co-founder Douglas Chen, and the key researchers. Not an acquisition. A reverse acqui-hire. Price tag: $2.4 billion. For the people, mind you. Not the product.

Tuesday: Cognition (the Devin lot) scoops up what's left: 250 engineers and $82 million in ARR.

So in the space of 48 hours, Windsurf got rejected, stripped for parts, and sold off at a car boot sale. If you're a Windsurf user wondering what's happening to your IDE, the honest answer is: nobody has a clue, least of all Windsurf.

Feels like: Watching a pub quiz team implode mid-game, with three other teams fighting over which one gets to adopt the remaining members.

---

Steve Yegge Says What Everyone's Thinking

The Pragmatic Engineer ran an interview with Steve Yegge this week, you know, the bloke who's been ranting about platforms since before most current SWEs had GitHub accounts.

His take on vibe coding: it's deceptively hard. Not the coding itself (the AI handles that bit). The hard part is knowing whether what the AI produced is any good. He reckons there's an emerging "AI Fixer" role inside companies, someone whose entire job is reviewing and correcting AI-generated code.

Which... yeah. That tracks. I've been saying this for months. Vibe coding is mint when you're prototyping or bodging together a script. But the second you need it to work in production, at scale, with actual users? You need someone who understands the code the machine spat out. And right now that someone is still a human.

The hot take from the same week, "all models are the same", is doing the rounds on tech Twitter. And I reckon it's about 60% correct. For most tasks, swapping Claude for GPT for Gemini gets you roughly comparable results. The differentiation is in the tooling, the UX, the ecosystem. Not the raw model output.

📚 Geek Corner

The "AI Fixer" pattern: Yegge's observation maps onto something systems thinkers have known for ages. Every wave of automation creates a new class of worker whose job is managing the automation's mistakes. Assembly lines created quality inspectors. Automated trading created risk managers. AI coding is creating AI fixers. The interesting bit is where these fixers sit: inside the IDE (like Cursor's diff review), in CI (automated code review bots), or as an actual human role. My bet: all three, layered. The tooling will catch the obvious stuff, CI will catch the structural stuff, and humans will catch the "this technically works but is completely wrong for our domain" stuff.

---

AWS Kiro: Another IDE Enters the Chat

Amazon launched Kiro this week. An "agentic IDE" because apparently we needed another one of those.

Look, I haven't given it a proper go yet so I'll reserve judgment. But the timing is... interesting. Windsurf just got dismembered. Cursor is the darling of the vibe coding crowd. VS Code with Copilot is the safe enterprise pick. And now AWS thinks what the market needs is their IDE?

The pitch is all about "spec-driven development", where you write a spec, Kiro builds it. Which sounds ace until you remember that writing a good spec is the hard part of software engineering. We didn't solve that problem when we had humans writing code, and bolting an AI on top doesn't make the requirements magically clearer.

Still. Amazon has a habit of shipping things that look naff at launch and then quietly becoming essential three years later. See: Lambda, ECS, CDK. So maybe this time next year I'll be eating my words.

---

China Drops a Trillion-Parameter Model (For Free)

While everyone in the West is arguing about who owns Windsurf, China quietly dropped a trillion-parameter open model. Free to use. Free to fine-tune.

The "AI is a commodity" crowd is having a field day. And honestly? The commodity argument gets stronger every week. When China can ship models this big, open-source, at no cost, what exactly is the moat for commercial model providers?

I'll tell you what it is: integration and trust. Nobody's running a trillion-parameter Chinese model on their healthcare data or financial systems. Not because the model isn't capable, but because compliance, data residency, and the fact that your CISO will have an actual cardiac event.

But for hobbyists, researchers, and anyone mucking about with prototypes? Absolute gift. The walls around premium AI are getting shorter by the month.

---

Claude Code: The Main Character of Week 29

Every other newsletter this week had a Claude Code tips article. Claude Code Router (route your requests to different models). Claude Code user-friendly wrappers. Claude Code best practices. Claude Code this, Claude Code that.

It's reached that point in a tool's lifecycle where the cottage industry of tips and tricks is bigger than the tool's actual documentation. Which is either a sign of massive adoption or a sign that the thing is too fiddly to use without a guide. Probably both.

The Claude Code Router bit is interesting though. It lets you intercept requests and route them to cheaper models for simple tasks. Which is exactly the kind of pragmatic cost-saving hack that makes the difference between "fun experiment" and "thing I can actually run in CI without my manager asking why the bill doubled."

---

DOGE Leaks an xAI API Key 🔑

And because no week is complete without a government-adjacent security cock-up: DOGE accidentally exposed an xAI API key. In the clear. On the internet.

I'm not going to pile on because frankly, everyone has committed credentials at some point. But when you're a government entity with "efficiency" literally in the name, leaking API keys is a proper bad look. Especially when the key gives access to Grok, and the whole thing gets picked up by every security newsletter in existence.

Here's what gets me though: we keep shipping AI tools faster than we ship the security practices to use them safely. API key management is a solved problem. Secret scanning is a solved problem. git-secrets, trufflehog, pre-commit hooks, all been around for years. And yet here we are, watching a government agency faff about with credentials like it's their first day on GitHub.

---

So Where Does This Leave Us?

Zooming out from the wreckage of the week. Three companies fighting over one AI coding startup. China giving away trillion-parameter models like free samples at Costco. Newsletter hot takes about all models being the same. Every cloud provider launching their own IDE.

We're watching AI coding become a commodity in real time. The models are converging, the features are all copying each other, and the pricing is in a death spiral towards zero.

The winners won't be the ones with the best model. They'll be the ones with the best workflow, the tightest integration between the model, the IDE, the CI pipeline, and the developer's actual habits. That's why Cursor is winning right now. Not because their model is better (they use other people's models). Because the experience of using it is better.

Bottom line: Vibe coding hit peak hype this week and I'm not sure the hype is wrong. The tools are getting proper good. But the messy acquisition drama, the security leaks, and the "just ship another IDE" energy all tell the same story: the AI coding market is moving faster than anyone's ability to make sense of it. If you missed the Claude Code vs Warp showdown, that tells you where the real differences lie. Strap in, crack on, and maybe rotate your API keys while you're at it.