Ideas for building
better software.

Introducing Run SDK: secure eval for your agents

Agents increasingly write TypeScript programs to coordinate tools and process their results. Once those programs touch real applications, some steps require authentication, while others need human approval. Executing that code with gives it the same access as the application around it, including its secrets and internal services, and leaves no durable way to pause at those boundaries.eval Today, we're releasing the , a package for executing untrusted JavaScript and TypeScript without giving it

The end of credential sprawl for agents

Every useful agent reaches beyond your codebase. It posts to Slack, opens pull requests, queries Snowflake, or calls an internal API. That reach is what makes it valuable, and it's also where the risk lives, because for years, granting it meant provisioning a long-lived token and hoping it never leaked. replaces long-lived tokens with ones your code requests at runtime, scoped to the task and expiring on their own.Vercel Connect During the , we've grown the ecosystem past 100 connectors, unifie

Sleepwalker: Passive Backdoor with Its Own Command Language

Comments

ISC Stormcast For Tuesday, August 25th, 2026 https://isc.sans.edu/podcastdetail/10066, (Tue, Aug 25th)

52 Days, 2,340 Rows, Every Cost Logged as Zero: The Stop Hook Trap

Going from a $700/month student side hustle to a real business in six months came down to one thing: I stopped instructing Claude and started letting it run the whole environment autonomously. That environment then spent 52 days writing 2,340 log rows where every single cost was zero — and it never once complained. Most people who start with Claude Code use it as a convenient chat AI. But once monthly revenue crosses a certain threshold, your thinking shifts. Instead of "issuing instructions and

The Upload Succeeded, the Record Did Not

Originally published on hexisteme notes. I built a YouTube upload stage for a video pipeline, and the flow looked clean enough on paper: start a resumable session, PUT the file, get back a video ID, verify the upload actually landed the way it was supposed to, then write a local record marking the episode as uploaded. Four steps, each one depending on the last. It was the dependency between the last two that turned out to be the problem. Verification here means re-querying the video through vide

Ensuring Safety in the Generative AI Ecosystem: Protecting Users from Non-Consensual Intimate Content

Disrupting a new covert influence campaign from Russia

OpenAI banned Russia-origin accounts using AI to promote a fake Israel-based think tank and a “sovereignty” index praising Russia and criticizing the West.