← Dailies

Daily · 2026-11-05 00:00 UTC

OpenAI buys Pierre, Meta buys Conductor

OpenAI buys the git plumbing for coding agents, Meta buys agent orchestration, and Uber admits almost nobody books a ride with an AI agent yet.

10 items · ~3 min listen

1. OpenAI is buying Pierre Computer Company — Pierre runs a cloud service AI agents use to store the code they generate, and its open-source Trees and Diffs libraries already underpin OpenAI's Codex interface. Terms were not disclosed; OpenAI says it will keep both libraries open source and fold Pierre's paid hosting into Codex. SiliconANGLE · Nov 4
2. Meta acquires Conductor, an AI coding startup — Conductor deploys teams of agents into parallel workspaces so engineers can work on separate parts of a project at once. Its agents come from Anthropic and OpenAI rather than in-house. Conductor's cloud service shuts down November 20 and cloud data is deleted after that; founder Charlie Holtz promises "something new" early next year. TechCrunch · Nov 4
3. Uber: four rides booked by AI agents in its busiest week — On the Q3 earnings call, Uber said it handled around 320 million trips and orders during Halloween week, of which only four rides and roughly 500 Uber Eats orders came through a personal AI agent. CEO Dara Khosrowshahi: "very, very early and very limited traction." DoorDash reported a few hundred agent orders. TechCrunch · Nov 4
4. ChatGPT now generates animated GIFs with Images 2.6 — Users can turn a text prompt or a photo into a flipbook-style animation, then adjust playback speed and individual frames. OpenAI says character consistency holds well enough across frames for the format to work, and that users generate more than four billion images a week. The Decoder · Nov 4
5. MIT shows GPT-6 Astra reading and rewriting other models' weights — In experiments the model identified which of four games a network had been trained to play from raw weights alone, and raised a Qwen model's math accuracy from 7.9% to 48.1% by changing 0.016% of its weights. The researchers call it "Weight-Space Programming"; it failed to spot a sycophancy fine-tune without also seeing attention patterns. The Decoder · Nov 4
6. Nothing teases a screen-free AI device for December — CEO Carl Pei said on X that the company's first AI device arrives next month, hinting that it will not have a screen — just over a year after announcing plans for AI-native devices and an AI operating system. The Verge · Nov 4
7. Adobe puts its marketing agent inside ChatGPT, Claude and Copilot — CX Enterprise Coworker is now available in ChatGPT, Claude, Databricks Genie, Gemini for business and Microsoft Copilot, with Amazon Quick, IBM and Snowflake expected soon. Adobe also shipped prebuilt Model Context Protocol connectors for Amazon Ads, Meta campaigns and Snowflake Cortex AI. PYMNTS · Nov 4
8. Microsoft Clarity adds tools to see what shopping agents do on your site — Agentic Analytics shows merchants what AI agents attempt on their pages and simulates whether those agents can complete a checkout; Brand Agents push products into recommendations inside Copilot, Bing and MSN. Ask Clarity, a natural-language analytics query tool, is in closed beta for retailers. PYMNTS · Nov 4
9. Qualcomm pivots to data centers as phone sales slide — With smartphone sales down around 20% year over year, Qualcomm forecast about $650 million in data center revenue for its next quarter, $5 billion by the end of 2027 and $15 billion by 2029 — eventually around 30% of the business, against $10.4 billion in total revenue for 2026. The Verge · Nov 4
10. Ricursive releases Ripple-65B Preview, a dense model that punches up — A 65B dense Transformer pretrained on 12 trillion tokens with up to 1M context scores 80.52 on Terminal-Bench 2.1, ahead of open-weight models several times larger, including a 501B model at 80.1 and a 975B model at 63.8. Apache 2.0, with an NVFP4 build that fits a DGX Spark. Hugging Face · Nov 4

Try This Next

• 🔧 Pull Ripple-65B Preview and run it on a single workstation — a 65B dense model that fits an NVFP4 build on a DGX Spark is a cheap way to test agentic coding locally.
• 🎥 Read the MIT weight-space programming writeup before your next fine-tune; reading a model's numbers is a different skill from training it.
• 📝 List the tasks where an AI agent could still save you time — the Uber and DoorDash numbers suggest booking and ordering are not among them yet.

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.