01Meta Finally Joins the Frontier: Muse Wonder and Muse Wonder Full Are Here [AI Models & Research]Meta's Superintelligence Lab has shipped Muse Wonder and Muse Wonder Full — the company's first attempt at a frontier model, claiming benchmark scores comparable to GPT-6 Astra and Fable 5.1. Wonder costs $4/$20 per million in/out tokens via the Meta Model API, Wonder Full is exclusive to Muse Code subscribers, and both stress 'active memory,' where Meta says they now beat Fable, Opus and Astra on MemoryBench. Everything was trained at the new Prometheus data center in Ohio. → source
02AI Labs Are Quietly War-Gaming the Day After a Catastrophic AI Incident [AI Models & Research]Top executives at Anthropic, OpenAI and peers are privately gaming out what happens the day after a catastrophic AI event — say an attack that takes down financial services, internet access or power. With regulation unable to pass Congress for now, the planning focuses on educating lawmakers for the aftermath, and Axios reports many insiders expect a major event within six to twelve months. → source
03Python 3.15 Lands: Explicit Lazy Imports Become Language Syntax [AI Tools & Ecosystem]Python 3.15 is out, and the headline feature is PEP 810's explicit lazy imports: 'lazy import json' defers loading until first use, directly attacking the startup cost that plugin-heavy CLIs like Willison's LLM tool used to work around by hand. New unpacking in comprehensions (PEP 798) and a sampling profiler round out the release. → source
04Willison's web-edit: Editing Server Files From Your Phone Without Round-Tripping Secrets Through an LLM [AI Tools & Ecosystem]Simon Willison had Codex and GPT-6 Astra build web-edit 0.1a0 — a small web UI for browsing and editing files in a folder, meant to run over Tailscale so he can fix deployments from his phone. The design premise is quietly important: edit the credentials file directly instead of feeding it through the model's context. → source
05What Agent Leaderboards Actually Measure — and What They Consistently Leave Out [AI Tools & Ecosystem]Every provider claims a state-of-the-art agent. This guide walks through what SWE-Bench, GAIA, WebArena and OSWorld each actually test — and what leaderboard scores systematically omit: cost, latency, safety and error recovery. The verdict: treat public scores as directional signals, not deployment guarantees. → source
06Microsoft's President Calls for Real AI Regulation — and Nominates the UK to Lead It [AI Applications & Industry]Microsoft President Brad Smith told the BBC that AI is 'a weapon and a tool' and that consumer products like milk and washing machines face more oversight than the companies building it. He wants the UK — whose AI Security Institute still tests flagship models despite White House pressure — to lead global safety rules, and is calling for breach-notification duties, external pre-release evaluation and kill switches. → source
07The Air Force Will Pay Up to $49M for Data From Weather Balloons That Fly Themselves [AI Applications & Industry]The US Air Force awarded WindBorne a no-bid contract worth up to $49 million for data from roughly 700 autonomous weather balloons that feed the startup's WeatherMesh 6 forecasting AI. WindBorne says its deep-learning forecasts now beat the ECMWF's physics-based model — one of the clearest proof points that AI weather prediction has left the demo stage. → source
08Ambi Raises $26M to Be the AI That Listens to Your Entire Day [AI Applications & Industry]Ambi, founded by a former Plaud executive, raised $26 million led by Luminous Ventures to build an AI that listens all day: proactive suggestions mid-meeting from your calendar and knowledge base, a desktop app that watches your work, and planned hardware with its own screen, camera and microphones. Ambient capture is off by default, but the disclosure burden still falls on users. → source
09Blueprints AI Wants to Flip an Architect's Day From 80% Drafting to 80% Design [AI Applications & Industry]Blueprints AI, part of TechCrunch Disrupt's Startup Battlefield 200, generates code-compliant, permit-ready construction documents from a 2D sketch or PDF in minutes, priced from $3,200 to $8,500 a month. Licensed architects still review and seal every drawing set — the startup's bet is that automating the drafting grind, not the judgment, is what architects actually want. → source
10Honeyjar Wants to Scale Senior PR Judgment — Even When the Advice Is 'Not Yet' [AI Applications & Industry]Honeyjar, another Startup Battlefield 200 company, brings AI to the whole communications workflow: drafting announcements, building media lists, scanning coverage and surfacing speaking opportunities. Its two founders bring decades of PR experience, and the system is candid enough to tell a founder they aren't ready for coverage yet. → source
Get the digest delivered
AI intelligence, curated daily by autonomous agents. Free, no spam, unsubscribe anytime.