Blog Posts

The other day I wrote about the npm worm that learned to trust your AI agent — a keyv-adjacent...

Someone built a browser game where you play the human-in-the-loop for an AI coding agent. Commands...

Build a Tencent RTC conversational AI companion whose listening, exploring, and advising modes are enforced by application state—not entrusted to an increasingl

When people discuss AI agents escaping their boundaries, the mental image is usually dramatic: a...

Agent incidents rarely look like a movie villain. They look like a helpful assistant combining two...

Last week a post on DEV asked a question I can't stop thinking about: we keep handing AI agents more...

A few months ago I wrote about regression-testing prompts before shipping them. The responses that...

A lot of us are now running coding agents that can read files, write files, and execute shell...

Last week I read yet another thread about AI agents getting more and more tools — shell access, file...

The European Union's AI Innovation Package is designed to give AI startups and small and medium-sized...

The counterexample first A planning agent decides it needs the latest pricing sheet, so it...

Enterprise AI automation is not scaled simply by giving more employees access to a workflow platform....

The Silent Tax: Calculating the Real Cost of Vendor Lock-In in Your AI Stack Vendor lock-in in AI...

When AI agents ask for permission to act, how often do humans actually catch the dangerous ones? A...

A multi-agent development loop run from one laptop: an orchestrator that writes no code, coders in isolated worktrees, and rules each earned by a failure.

Your agent runs as a service account that can do everything. Scoping its authority to the user it acts for, as a TypeScript capability set the tools enforce.

SIGTERM arrives at step 14 of 20. What has to be true for the agent to continue rather than restart, and the schema-version trap on the way back.

Rewind a run to any step, change one thing, and replay from there. Turning a checkpointer into the only debugger that works on a nondeterministic system.

A 40-second agent run with no output is indistinguishable from a hang. A typed event stream, an SSE endpoint, and a client that survives a reconnect.

An engineering guide to implementing NIST ML-KEM and ML-DSA in multi-agent and RAG architectures.

As autonomous AI agents evolve from simple chat-based assistants into fully autonomous digital...

Persona-based AI avatars (teacher, coach, psychologist, etc.) raise a real engineering question that...

I spent two weeks with a CLAUDE.md and an AGENTS.md saying different things about the same...

Someone on your team pastes a stack trace into an assistant, gets a plausible diff back, and hits...