Library
Where the work sits
A map of the field. Shelves follow the work, not a single launch.
Start here
One worked check, four list prices, then three questions — before you trust what changed this week.
Pocket list (November 1, 2026)
- Measurement — which named test, on which date, scored by whom.
- Bill — list input and output per million tokens on the vendor page.
- Door — live API, lab claim, replay, or a downloadable file — keep them separate.
Models · Agents · Fine-tuning · All chapters
Start here
Ideas that outlast a single launch. Each page is tied to developments already on the briefing.
Model evaluation
A number belongs to the test, the date, and the person who measured it. A lab's score and an independent rerun are different objects. Leave naming the suite, who ran it, and whether you can call the model that produced the score.
Inference
Where the model runs, and what a million tokens cost, once the launch post is over. Leave naming input vs output, whether the model is on a public API, and what a loop multiplies.
Frontier models
The models a lab sells as its strongest general system, usually through an API or an app, with a published price or a named tier. Leave naming the door (API, app, or waitlist), the price, and what the lab's rank does not prove.
Open-weight models
A file a team can download and inspect. The license and the machine decide whether that file is usable. Leave naming the license, the active parameter count, and whether you have the machine to run it.
Agent harnesses
The model writes the next step. The harness decides which tools, memory, and checks surround that step. Leave asking which piece moved: model, tools, loop, or the check.
Models
The systems themselves: assistants, open weights, and specialists. Name the layer before you repeat the claim.
Agents & Harnesses
The model is one part. The harness is the tools, the memory, the context, and the check that decide what that model can finish.
Developer Tools
Software people run. We file a tool when a workflow changes: how a model is served, indexed, or evaluated.
Research
Papers, system cards, and technical reports. A product post, a system card, and a paper are different evidence.
Infrastructure & Chips
Chips, clusters, inference servers, and the cost of running a model. Filed when a spec or a price comes from the company that sells it.
Enterprise AI
AI inside a company: admin controls, data boundaries, and the difference between a demo and a rollout.
Security & Safety
What a model can be talked into, what an agent can touch, and what was actually measured.
Industry & Companies
Labs and the companies around them. A map of who builds, who hosts, and who applies. Not a stock note.
Policy
Rules, reporting duties, and liability. Filed when we can name the institution and the document.
Definitions
Kept off the main shelves, including superintelligence. Open when you want a definition, not the day’s ranking.