Anthropic bought Stainless, the New York startup that generates and maintains software development kits for OpenAI, Google, Cloudflare, and dozens of other companies. The acquisition was announced Sunday.

If you’ve used the OpenAI Python SDK or the Anthropic TypeScript client in the past two years, you’ve used Stainless’s output. The company’s core product automatically generates idiomatic client libraries from API specs, then keeps them in sync as the API changes. It handles type definitions, error handling, retry logic, pagination, and all the other unglamorous details that make the difference between a usable SDK and one developers hate.

Stainless was founded in 2022 by Alex Rattray and Colin Kennedy. Before starting the company, Rattray worked on infrastructure at Stripe, where he saw firsthand how much engineering time goes into maintaining high-quality SDKs across multiple languages. Stainless automated that work using a combination of code generation and language-specific templates.

The timing matters. AI companies ship fast and break APIs constantly. Manual SDK maintenance can’t keep up. Stainless solved a real problem at exactly the moment when every AI lab needed the solution.

The acquisition creates an obvious conflict of interest. Anthropic now owns the tooling that generates its competitors’ client libraries. That’s not inherently bad, but it requires trust. Will Stainless continue to support OpenAI’s SDK with the same priority? Will new features ship to all customers equally? The answers depend entirely on how Anthropic runs the business.

The company could keep Stainless independent, treat it as neutral infrastructure, and benefit from the revenue and strategic position. Or it could slow-walk competitor support and use access as leverage. We don’t know yet.

From a technical perspective, nothing changes immediately. Existing Stainless customers should see the same service. The generated SDKs will keep working. But long-term, this is a bet that Anthropic can operate developer tooling as a separate business without letting competitive dynamics poison it.

There’s precedent for this working. GitHub is owned by Microsoft but still hosts OpenAI’s repositories. Hugging Face takes money from every AI lab and maintains neutrality. On the other hand, there’s also precedent for acquisitions killing trust. If OpenAI or Google starts looking for alternatives, we’ll know the answer.

For developers, the practical impact depends on scale. If you’re just using the OpenAI SDK, nothing changes. If you’re building on multiple AI platforms and relying on consistent SDK quality across all of them, watch how this plays out. Stainless’s value was neutrality. That’s harder to maintain when you’re owned by a player.

Cursor Ships Composer 2.5

Cursor released Composer 2.5, the latest version of its multi-file editing mode. The update focuses on better context management and fewer hallucinated edits.

The headline feature is “codebase-aware” context selection. Instead of requiring you to manually specify which files the agent should look at, Composer now searches your project automatically and pulls in relevant code. In practice, this means fewer failed edits because the model didn’t know about a dependency or type definition.

Cursor also added better diff previews and smarter merge conflict resolution. If Composer suggests changes to a file you’ve edited since the agent started running, it’ll attempt to reconcile both sets of edits instead of just failing or overwriting your work.

This is incremental progress, not a breakthrough. Cursor is still the best AI code editor for agents that edit multiple files at once, but the core problems haven’t changed. The model still guesses wrong sometimes. It still makes changes you didn’t ask for. You still need to review every diff.

If you’re already using Cursor, the update is worth installing. If you bounced off Composer because it kept breaking things, try it again. The error rate is lower.

OpenAI and Dell Partner on Enterprise Codex Deployment

OpenAI and Dell announced a partnership to bring Codex, OpenAI’s code generation model, into hybrid and on-premise enterprise environments. The deal targets companies that won’t send proprietary code to the public cloud for regulatory or security reasons.

Dell will handle the infrastructure. OpenAI provides the model. Enterprises get a deployment option that keeps everything inside their own network perimeter.

This matters less for startups and more for banks, healthcare companies, and government contractors. Those organizations have been stuck using worse models because the good ones only run in OpenAI’s cloud. Now they can use Codex without changing their security posture.

Pricing wasn’t disclosed. Neither was the exact model variant Dell will deploy. If it’s a smaller, faster version of Codex tuned for on-premise hardware, that’s useful. If it’s the full cloud model with worse latency because it’s running on local servers, the value proposition is weaker.

The partnership is a hedge. OpenAI gets enterprise revenue it couldn’t access before. Dell gets to sell AI infrastructure. Enterprises get a compliance-friendly option. Whether it’s fast enough and cheap enough to actually use is the question no press release answers.