· 1 min · tooling
Your agent is only as good as your repo
Teams keep grading the model on codebases that would fail any new human too. The preparation argument, with the checklist I actually run.
i ship llm features for a living. i write down what breaks, what it costs, and what i'd do differently.
· 1 min · tooling
temperature is a TypeError now, and the API sends a 400 for it anyway
Anthropic's Python SDK v1.0 shipped August 20 and deleted temperature, top_p and top_k from Messages calls. On Opus 4.7 and later the API already rejected them. Migration notes.
· 3 min · agents
You probably don't need multi-agent
The orchestrator, the critic, the researcher and the planner walk into your architecture. One model with good tools was already doing the job. The math and the exceptions.
· 5 min · tooling
agents can't read your website, and the fix is one Accept header
Plain fetches get div soup and cookie walls. Stripe, Cloudflare, Vercel and the Claude docs now answer Accept: text/markdown, and copying them takes an afternoon.
· 2 min · architecture
The price crash quietly turned your 2024 cost hacks into debt
Frontier output pricing is down roughly 88% since early 2023. The clever cost engineering from two years ago is now complexity you pay to keep.
· 2 min · tooling
MCP went stateless and my infrastructure diagram got boring
The 2026-07-28 spec release candidate kills sticky sessions, adds real auth, and makes remote MCP servers deployable like any other web service. Finally.
