Mike Darlington ▲

Your agent crashed halfway through a turn. What happens to the work? durable-harness records each step, so a new function picks up the interrupted turn instead of starting over. Works with Claude Code, Codex, OpenCode and Pi. eve and fx already run this way.

Terence Tao on Numberphile, after OpenAI's dump of AI-generated proofs: mathematics is "transitioning from an era of proof scarcity to proof abundance." "Now we're drowning in proofs, but we're not getting all the understanding that we used to get." Swap proofs for pull requests and it's the same story in software. Code is abundant now. Understanding of the code is the scarce thing. Full video 👇

Your coding agent can now make images, video and audio. AI CLI lets any agent use any generation model: Claude Code with Flux, Codex with Gemini, opencode with MAI, fx with Grok. One install covers every pairing, so you choose the agent and the model.

How to make your coding agent nearly 3x faster in one line: /model deepseek/deepseek-v4.1-flash-1017-ultrafast Works in fx on @vercel AI Gateway. Same prompt, 1.7s vs 4.7s 🖤

Three services on port 3000. No collisions. portless now supports Docker Compose. portless compose up gives every service a named URL. Routes survive container restarts, and compose down cleans them up. No more remembering which port is which.

Boris Cherny (@bcherny), creator of Claude Code, on Lenny's Podcast: a year ago a model could run 15 or 30 seconds before going off the rails. Now some Opus runs go for hours or even days. Once an agent runs for hours, a single request is the wrong shape. You need durable execution: steps that checkpoint, retry and resume after a crash or a deploy. That's what durable workflow engines like Vercel's Workflow SDK are for. Full ep 👇

agent-browser, for your whole computer. agent-desktop gives any coding agent computer use on macOS, Windows and Linux. It runs in the background while you keep working, and agents can work in parallel, each with its own cursor.

Aaron Levie (@levie) on the role he sees every enterprise needing: an internal forward-deployed engineer. Is this what AI adoption inside a company actually looks like. Someone who understands both the work and the tooling, sits with a team, and wires agents into how they really operate. If you're an engineer who likes talking to the business, this is the most leveraged seat in the building right now. Full ep 👇

AI is writing more of your code. Who's checking it for security? Vercel Security Review scans every push for vulnerabilities, and Vercel Agent can open a PR with the fix. It only reviews what changed, so it finishes in seconds and runs on every deploy.

Your agent can now say "I'll do that tomorrow at 9" and actually do it. Vercel Schedules lets apps and agents create, change or cancel one-time and recurring jobs while they run. No redeploy, no database, no polling job. The request arrives when it's due.