AUTOMATER NEWSROOM

What’s changing.
Why it matters.

New tools, useful workflows, and consequential AI developments.

Subscribe via RSS ↗

Latest briefing

Oct 21, 2026, 10:55 PM EDT
Microsoft Says Human-Directed AI Helped Prove the 31-Year-Old KLS Conjecture — Unguided Runs on the Same Token Budget Fell Short
Microsoft Research says the 31-year-old KLS conjecture fell to a human-directed agentic search: roughly 100 AI-explored trajectories, pruned round by round, plus parallel independent proofs and a Lean 4 check. Unguided frontier-model runs on the same token budget fell short.
Oct 21, 2026, 8:59 PM EDT
Databricks AI Runtime Goes GA With Serverless NVIDIA GPUs for Training and Fine-Tuning Open Models
Databricks AI Runtime is GA: managed NVIDIA GPUs meet governed Lakehouse data, with NeMo, NGC containers and Nemotron cookbooks for enterprise fine-tuning.
Oct 21, 2026, 8:55 PM EDT
Qwen Code v0.26.0 Ships Daemon Control Commands, Revocable Web Shell Pairing and a Loop-Protection Default That Reminds Instead of Stops
Qwen Code v0.26.0 adds daemon lifecycle commands, listable and revocable Web Shell pairings, artifact records for agent-edited files, and a model.loopProtection default that reminds instead of stopping loops. What operators should check first.

Browse all →

Diagram: a Researchers box steers a fan of gray pruned agent trajectories and one indigo selected path, flowing from a key criterion through repeated recursive refinement and bounded errors to a dimension-free O(1) proof of KLS; an amber note records that unguided runs on the same token budget fell short.

October 21, 2026

Microsoft Says Human-Directed AI Helped Prove the 31-Year-Old KLS Conjecture — Unguided Runs on the Same Token Budget Fell Short

Microsoft Research says the 31-year-old KLS conjecture fell to a human-directed agentic search: roughly 100 AI-explored trajectories, pruned round by round, plus parallel independent proofs and a Lean 4 check. Unguided frontier-model runs on the same token budget fell short.

Diagram showing Claude's Reed-Solomon decoder storage reserve cut from a 665-qubit naive worst case to a 19-qubit optimized reserve, one of three combined changes that together cut the decoder's total gate count 12x versus Google's prior construction, alongside a separate 17x gate reduction for a new elliptic-curve modular inversion circuit found in a 53-hour run.

October 17, 2026

Anthropic Says Claude Mythos 5 Autonomously Designed Quantum Circuits That Beat Published Records, Cutting Gates Up to 17x

Anthropic says Claude Mythos 5 designed quantum circuits beating published records on 7 of 8 benchmark tasks, cutting gates up to 17x via a new divsteps algorithm.

Explanatory hero graphic: four Gemini 4 Argon agent boxes on the left connect through a bus to in-game chat and shared-memory boxes; a chain of dependent step boxes leads to a red goal box reading Ender Dragon defeated in 27 h 14 m; a green panel reads live dashboard, pause and stop available, no intervention required; and a stats strip lists 2,783 programs, 605 deaths, four agents and no human-written game-specific tools.

October 17, 2026

Four Gemini 4 Argon Agents Beat Minecraft End to End in 27 Hours

Google DeepMind researchers report that four Gemini 4 Argon agents, coordinating only through in-game chat and shared memory, defeated Minecraft's Ender Dragon in 27 hours — 2,783 self-written programs, 605 deaths and no human-written game logic.

Our AI-assisted editorial method · Explore our in-depth Intel guides