OpenAI is buying Pierre Computer, a Git storage startup it says is already inside Codex
Pierre's Code Storage hosts the code AI agents write for Lovable, Bolt and Amp, and the deal comes two days after OpenAI showed its own Git backend. Three more stories inside.
Today's lead is an acquisition by the company behind Codex. It leads as the day's one advancement for a tool many of you run, and it's hours old, so its Hacker News thread is still small. Also today: a GitHub issue says Postman's desktop app put its plugin into Claude Code and Codex without asking, on the day's biggest Hacker News thread. Codex users say a training opt-out they'd turned off is back on, again. And Cursor changed which cheap models skip its $0.25 fee on business plans. Three of the four reached us through users or a pricing-page change, not a vendor post.
What changed this week
OpenAI is buying Pierre Computer, whose Git storage already runs inside Codex
OpenAI says it will acquire Pierre Computer Company, "expanding Codex’s tools for building, understanding, and reviewing code," in a post published at 14:30 UTC. It had told Axios that Pierre is "a developer tools startup whose technology is already embedded in Codex," and Techmeme's summary of that report says the price wasn't disclosed. Pierre sells Code Storage, Git hosting built for agents: its site says it "powers the largest agentic workloads in the industry" and names Lovable, Bolt and Amp as users. Two days ago OpenAI described Pando, its own in-house Git backend, in a post we covered yesterday. Pierre announced the deal on X; its own site doesn't mention it yet.
Postman's AI plugin installs itself into Claude Code, Cursor and Codex by default
A GitHub issue says Postman's Mac desktop app installed its plugin into the user's Codex and Claude Code on launch, with no command run. Postman's own docs say its installer targets every coding agent it finds by default, and on Codex it "approves the hook for you", a start-up instruction steering the agent to Postman's tools. Commenters on the 1,393-point Hacker News thread say it also adds an outbound MCP server, a remote tool connection. Check yours: npx @postman/postman-plugin status.
indicava
Codex users say the data-sharing toggle they turned off flipped back on
A Hacker News poster says the Codex data-sharing setting their team had switched off was "silently reenabled," and none of them had opted back in. Their company keeps it off so its code doesn't reach AI labs through prompts. Several commenters report the same, and one calls it "illegal behavior." It repeats a September thread, "OpenAI keeps re-enabling the 'allow training' setting," which this newsletter covered then, and no OpenAI statement turned up either time. The toggle is "Improve the model for everyone" in ChatGPT's data-controls settings.
Source: news.ycombinator.com
Cursor's $0.25 fee now skips cheap models by price, not by a named list
Cursor charges Teams and Enterprise plans an extra $0.25 per million tokens on third-party models, and just changed which cheap models skip it. A list of 15 became a price test: any model "whose cache-read price is strictly below $0.10 per million tokens across every pricing tier." By Cursor's own price table, that adds DeepSeek V4.1 Flash ($0.006 per million cache-read tokens), which forum users asked for on Oct 28. The changelog is silent, and the old list was last seen Nov 2.
Also worth your time
- Simon Willison finds Qwen 3.8 27B can interpret model weights too, running it on his laptop
- ffmpeg-skill lets Claude Code, Cursor or Codex edit video with local FFmpeg (early, unverified)
- Mesh Avatar Studio has a coding agent rig a 2D avatar from one illustration, per AlphaSignal
Know someone who'd want this in their inbox? Forward it — that's how this grows. And if we got something wrong, or you think we buried the real story today, hit reply. A person reads every one.
The New Way is written with AI. It gathers the day's stories, checks them against their sources and drafts every summary. A person decides what runs and reviews every issue before we hit send.
