A PreToolUse hook is the guardrail that survives bypass mode
Permission prompts depend on what mode you're in. A PreToolUse deny hook doesn't. Here's a small one that blocks force pushes, and where it falls short.
3 min read
AI for people who write code.
News on coding assistants, frontier models and GitHub workflows, plus practical ways to get more out of them. Written for working developers who've heard enough hype.
Latest
All postsA separate evaluator agent is what makes long agent runs worth the money
Anthropic's harness splits planning, building and testing across agents. The cost jump was 20x, and the reasons it paid off are specific.
3 min read
Google's Agent Development Kit hits 1.0 for Kotlin, with Android support
Compile-time tool schemas, human-in-the-loop flows and resumable sessions, for teams that build agents in Kotlin.
2 min read
Cursor Projects puts a coordinator agent in charge of the work
The agent that plans and delegates never writes the code. Cursor says it can keep a project's context for months.
3 min read
Build a Claude Code subagent that reviews your diff with fresh eyes
A reviewer that never saw the code being written catches different things. Here's the file, the invocation and the over-reporting trap.
3 min read
What 16 parallel agents building a C compiler teach about running agents at scale
Anthropic's engineering team spent nearly $20,000 on it. The lessons about tests, locks and duplicate code transfer to much smaller jobs.
3 min read
Run Codex from a script: codex exec, structured output and CI
The non-interactive mode turns Codex into a pipeline step. The defaults are conservative, and one warning matters a lot.
3 min read
Let the agent interview you, then write the spec, then start fresh
A three-step workflow for features too big for one prompt, and what makes a spec worth handing over.
3 min read
That 2-point benchmark lead might just be a bigger server
Anthropic measured how much infrastructure settings move agentic coding scores. It's more than many leaderboard gaps.
3 min read
Claude Code's auto mode catches most risky actions. Anthropic's numbers show what it misses
A 0.4% false positive rate sounds great. The 17% miss rate on real overeager actions is the number to plan around.
3 min read
Topics
- Assistants12
Coding assistants and agents, and how to get better results from them.
- Code review3
AI in code review and pull requests.
- GitHub1
Actions, Copilot and AI integrations on your repos.
- Models2
Frontier model news from Anthropic, OpenAI, Google and xAI.
- Tools3
New AI developer tools and notable releases.