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: a rocket builder and a code editor meet, $60 billion flows one way and ownership the other, and every engineer's Cursor window now sits under SpaceX.
Days after its record IPO, SpaceX is acquiring Anysphere — the company behind Cursor — for $60B in stock, the largest AI-coding deal on record. The read: Musk is buying his way to the frontier of developer tooling rather than building it, and consolidating an editor that millions of engineers now depend on. If your team runs on Cursor, treat roadmap, pricing, and data-handling as open questions and make sure you have a fallback editor and an export of your configs. Watch what happens to xAI integration — that's the tell for where the product goes next.
HOW TO READ THIS Read top to bottom: Microsoft ships the 7B-parameter Fara-7B agent, it perceives the screen and acts back on it in a loop, and that same loop lets it operate across any open application.
A 7B open-weights model built to actually operate desktop and web apps — a bet that small, efficient agents can do computer-use work that today demands frontier-scale systems.
HOW TO READ THIS Read top to bottom: the agent designs an experiment, runs it alone on one GPU, then loops the result back into itself with no human step.
Agents that autonomously run ML experiments on single-GPU nanochat training — self-improving research you can reproduce on one consumer card, not just read about in a paper.
HOW TO READ THIS Read top to bottom: OpenBMB ships VoxCPM2, a reference voice waveform enters the cloning engine, the output waveform matches it, then the weights are released open.
Tokenizer-free, open-weights TTS with true-to-life voice cloning — studio-grade synthesis is now free to anyone, and so are the deepfake risks; assume voice alone is no longer proof of identity.
The high-throughput, memory-efficient inference and serving engine that's become the default way to self-host LLMs.
Open-source AI chat platform with advanced features that works with every LLM — a self-hostable alternative to closed assistants.
An agent toolkit bundling a unified LLM API, agent loop, TUI, and coding-agent CLI in one place.
A curated set of practical builds — OCR, RAG, and agents — for engineers who learn by shipping real use cases.
A growing archive of leaked system prompts behind Cursor, Claude Code, Devin, and other AI tools — useful for studying how they're actually steered.
Interconnects A framing piece on what AI governance looks like once you take the AGI scenario seriously.
SemiAnalysis A teardown asking whether SMIC's N+3 metal pitch is now smaller than Intel 18A's — China's process gap, measured.
The Sequence Last week in AI: Anthropic ships, Apple borrows, Musk lists, Bezos builds.