AI that matters, from the architect's desk. Curated and engineered by Saaket Varma, PhD — no hype, just signal.
HOW TO READ THIS Read top to bottom: the lab ships Muse, which merges a friend's face into your photo, then that spreads to three apps.
Meta shipped Muse Image, the first model out of its Superintelligence Labs division, and wired it into the Meta AI app, Instagram, and WhatsApp — including the ability to pull other Instagram users into generated photos. This is the first concrete answer to what Meta's superintelligence bet actually ships as product, and the answer is consumer image generation with your social graph as raw material. The consent model here is the real story: your likeness can now appear in someone else's AI output by default of a platform setting, which is exactly the kind of design that draws regulator attention in the EU and state privacy suits in the US. If you build on Meta's platforms or advise anyone who does, go read the likeness controls now and assume the defaults favor generation, not protection.
HOW TO READ THIS Read top to bottom: gameplay footage becomes training data that teaches a robot model to act.
General Intuition is training foundation models for physical AI on millions of hours of video game footage, claiming robotics is near its ChatGPT moment. The bet is that game data can substitute for real-world robot data, which is the single scarcest resource in embodied AI — every serious robotics lab is bottlenecked on it. If sim-to-real transfer from game footage holds up at scale, the cost curve for robot capability starts looking like the LLM cost curve circa 2022. Watch for actual transfer benchmarks before buying the framing; the claim is cheap, the evidence isn't public yet.
HOW TO READ THIS Read top to bottom: Facebook ships the Astryx repo, its design-system layers assemble, agents read the structured specs directly instead of guessing, and adoption spikes to 7,236 stars.
Facebook open-sourced Astryx, a fully customizable design system built to be consumed by AI agents, and it's climbing GitHub trending fast (7,236 stars and accelerating). The interesting part isn't the components — it's the premise that design systems now need a machine-readable contract so coding agents can compose UI correctly without a human in the loop. That's a structural signal about where frontend work is headed: the design system becomes the API surface, and the agent becomes the consumer. If your org maintains a component library, agent-readability is about to become a requirement, not a nice-to-have.
HOW TO READ THIS Read top to bottom: Menlo Ventures backs Lovable, which closes a $300M round, and that round doubles Lovable's valuation to $13.2B.
Lovable is reportedly raising a $300M round led by Menlo Ventures that would double its valuation to $13.2B. Treat this less as a company story and more as a price signal: investors are still paying runaway multiples for AI app builders, which tells you the market believes vibe coding captures durable workflow share rather than being a feature the model providers absorb. The risk case writes itself — every frontier lab is shipping app-generation natively — so the valuation only works if distribution and iteration speed are the moat.
Office suite built for AI agents — lets agents read, edit, and automate Word, Excel, and PowerPoint from the command line.
Open-source agentic browser positioning itself as the alternative to ChatGPT Atlas, Perplexity Comet, and Dia.
Open-source toolkit for building your own AI SRE agents — incident response as an agent workload.
MCP server giving Claude terminal control, file system search, and diff-based file editing on your machine.
Coding-assistant skill that turns any folder of code or SQL schemas into a navigable knowledge graph — works across Claude Code, Codex, Cursor, and more.