What's new in DevEx - January 22

Cursor ran an internal experiment to build a browser from scratch, generating ~3 million lines of Rust code in a week. “It kind of works”, but it’s far from a production-ready browser. What matters isn’t the output quality, but the speed and coordination of agents, as this would normally take teams months to get done.

OpenAI says it’ll begin internal testing ads in ChatGPT for Free + Go users in the US, with higher tiers staying ad-free and promises that prompts won’t influence the ads. This is clearly incentivized by their push for higher revenue amid rising costs and stronger competition. Time will tell if Claude and others follow this trend.

Ryan Dahl's recent statement isn’t that engineers are obsolete, but that AI already handles a large share of production code (around 30% at Google and Microsoft, and most of Claude Code’s own codebase). What changes is the work itself: engineers must now review AI output, shape system architecture, and decide what should be built.

Stable release just got shipped after months of production use. The update adds server adapters to run agents inside existing web apps and composite storage to separate memory, workflows, and observability backends. AI SDK v6 support and a unified observability schema make long-running agent systems easier to deploy and debug.

Railway raised a $100M Series B as its platform grew to over 2 million users through organic developer adoption. Teams use Railway to avoid traditional cloud overhead and ship faster as iteration speeds increase. Its growing enterprise footprint shows developer-first infrastructure is rapidly moving into mainstream engineering orgs.

The integration combines durable execution with LLM observability for production agents. Temporal keeps multi-step workflows alive through crashes and retries, while Braintrust adds tracing, prompt versioning, and cost visibility across LLM calls. Together, they make long-running agents easier to debug, evaluate, and operate.

Rafter provides a simple way to find and fix vulnerabilities in your code before they reach production. It runs fast, requires no setup, and focuses on the kinds of mistakes that show up when teams move quickly or rely on AI-generated code.

Rafter started when a friend vibe-coded a great project, but hesitated to ship it because he was worried it would get hacked. Today, it gives developers actionable feedback in under 30 seconds, without tuning rules or learning new jargon. It’s already being used by solo builders and fast-moving teams, and is backed by Mucker Capital. Worth a look if security keeps getting pushed to “later.”

After visiting Toronto, SF, and LA, we’re bringing Future of DevEx back to New York City on February 24! We’re partnering with PostHog and Arthur to explore how AI is changing the way we build, and share insights we’re seeing across the industry.

If you’re an engineer or a technical founder in New York, this is THE PLACE for you to learn where developer workflows are headed. Spots are limited, so sign up fast.

If you’re not in New York, stay tuned as we’ll be announcing more cities in the coming months!

Got a burning engineering question? Just hit reply and we’ll tackle it. And if you enjoyed this issue, consider sharing it with your friends and teammates.

Want to share your dev tool, research drops, or hot takes?

Submit your story here - we review every submission and highlight the best in future issues!

Till next time,