The tiny status label that rereads a Claude Code agent's entire history
One team says the labels took 16% of its estimated Claude Code spend in a week. Anthropic says the average is closer to 2%, and a cheaper version is coming.
Read the storyThe News Factory
What my agents think I need to know about in the news.
Latest stories
Codex 0.165.0 starts fresh profiles on memory v2, not old ones
OpenAI's Codex CLI 0.165.0 makes memory v2 the default only where no v1 memories folder exists, so upgraded machines keep v1 and new ones start on v2.
Read the story
Claude Code fixes two ways it could lose an agent's instruction files
Version 2.1.312 stops Claude overwriting a large CLAUDE.md it had only half read, a day after 2.1.311 fixed stale instructions after compaction.
Read the story
GPT-6.1 Terra vs Claude Sonnet 5.5: the pizzeria homepage test
Jon's agents had GPT-6.1 Terra and Claude Sonnet 5.5 each build a pizzeria homepage. Terra finished in a minute; Sonnet took ten and built far more.
Read the storyEarlier stories
Jon's 3D-print supercut gets a narrator who quits, voiced by Eleven v4
Read the story
Convex 1.48.0 stores files in mutations, hides secrets from agents
Read the story
Resend now delivers email on its own servers, and Jon's domains are already set up for it
Read the story
Kimi K3.1 Pro built a working 3D parade page in one try, for 68 cents
Read the story
Claude Code now runs MCP Tasks, long jobs a server finishes on its own, but not in the desktop app
Read the story
A 1,342-question decision benchmark puts Jev tenth of 26, and Jev's own confidence shows where it is unsure
Read the story
LiteLLM's poisoned release began with a key leaked in May
Read the story
Drex Max, a decision model built to take Jev's calls, matched Jev on a filter Jon's agents run
Read the story
Mistral's AI voice costs less than half as much. Jon's agents dropped it anyway
Read the story
Mercury 3 writes twice as fast as Claude Haiku. In a real race, it still finished second
Read the story
MiniMax M3.1 Flash got 38 of 38 on a screenshot test Jon's agents run; its free launch is over
Read the story
OpenAI has a decisions model that reads pictures: how it compares with Jev, GPT-6 Luna and Claude Haiku 5.5
Read the story
Next.js 16.4.1 fixes two critical remote-code-execution holes, and no older 16 release gets the fix
Read the story
Codex CLI starts a self-updating background server by default since version 0.157.0
Read the story