An intelligence system for the AI economy, built on a map of the argument.
The Atlas is nine portals over one body of material: the Signal Board files each tracked development under the audiences it matters to and the claims it bears on; the News Blotter is the Daily Edition, a newspaper written each weekday from what the engines collected; Savant is the Atlas’s own weekly research report, with a point of view and a ledger of open hypotheses; Claims & Theses is the argument map itself; the Report Portal holds every report the Atlas writes, each a cited PDF; the Data Portal offers the corpus as datasets (26 today, 13 of them behind an access key); the Research Portal reads new AI research against the map; Startup Scout follows young AI companies; and the Tooling Monitor scans the market for AI tools each week. Beside them, Ask answers questions over all of it with citations you can open, and Education keeps a shelf of guides.
The AI Atlas collects what happens in AI every weekday, from news wires, regulatory filings, research papers and the tool market, and places each development on a map of the open questions: what is being claimed, what would settle it, and which evidence moves it. Engines gather and structure the record; a daily edition and a weekly research report read it; Ask answers over the whole corpus with citations you can open. It is written for people carrying AI transformation inside regulated financial institutions, where a claim has to be checkable before anyone acts on it, and where the honest answer most weeks is that nothing settled.
Everything else hangs on one structure. An open question about AI and the economy holds two to four stances, each labeled with who holds it. Each stance rests on claims, and every claim carries a test: what would have to be true to stop believing it. Evidence attaches to a claim as supporting or contradicting it. A claim that links two domains, say compute costs to labor markets, is pulled out as a bridge-claim and tested on its own. The maintainer’s confidence in each claim, and the written reason for every change, sits in a private layer; the public view is the same map with that layer removed. The map is how a new development gets placed: not "is this big news" but "which argument does this move, and which way".
Collect: collection engines run on weekday schedules (a news scan, the Signal Board’s discovery pipeline, company intelligence, new research, and a weekly tooling scan), and every item is kept with its source. Structure: models turn raw text into records under written rules (a summary, tags, extracted facts, a relevance score, a source-reliability tier), and every record is indexed for search by words and by meaning. Place: a development becomes a signal that touches named claims; publishing it, by a person or by a 48-hour policy for high-significance drafts that nobody archived, is what writes its evidence onto the map. Read: the Daily Edition writes the day up each weekday afternoon; on Fridays the research roundup and Savant write the week, and Savant poses a new hypothesis and revisits the open ones. Ask: questions get answers drawn from the records and the Atlas’s own reports, each cited, with a line marking where an answer goes beyond what the Atlas holds. Decide: nothing the machines produce moves a confidence. A person moves it, and has to write down why.
Most weeks nothing happens that should move the map, and the Atlas says so. Success is being able to place a new development in a minute and know what it touches, and to hand a colleague a report where every sentence opens onto its record.
The deep argument map covers the market and economics of AI. The collection engines range wider, from bank regulation to the tool market, and the Signal Board files what they find under six audience lenses. The whole thing is a personal project and a running record of how the maintainer reads the field, not a consensus and not advice.
Every portal: what it does, what it looks like, and the pages inside it.
The engines, the gates between a model and the public page, and how Ask finds things.
What it does not do, what is not built, and the ways it can be wrong.
Where the data comes from, which services see it, what is stored, and who can read what.
Every term defined: question, stance, claim, test, bridge-claim, signal, Savant, and the rest.
What this does that a general chatbot or a research platform cannot.