About this site

A neutral guide to the evidence on AI's economic impact, grown out of tracking the wemustactnow.ai open letter. Not affiliated with the letter's organizers. It takes no position on whether any claim is right.

The letter

3,508 signatures from 91 countries, tracked daily. Its signatories' universities and names seed the research collection.

"We Must Act Now: A Statement on AI's Transformation of the Economy" is an open letter published at wemustactnow.ai. The full text, quoted verbatim (retrieved 2026-07-18), is three numbered statements:

1. AI may become radically more powerful over the next 10 years.

2. This could drive an unprecedented transformation of our economy, larger than the Industrial Revolution, but unfolding over a vastly shorter time frame. It could bring risks, including large-scale job displacement, as well as opportunities such as major gains in living standards.

3. Economists, policymakers and technology leaders must act now to understand the economics of transformative AI and to build the incentives, guardrails, and institutions needed to steer AI in a direction that complements humans and benefits society.

What it says, in plain language

This summary is AI-written.

The letter makes a conditional argument, not a prediction: if AI keeps getting much more capable over the next decade, the economic change could be bigger than the Industrial Revolution while happening much faster — and nobody should assume the outcome will sort itself out. It names one headline risk (large-scale job displacement) and one headline opportunity (major gains in living standards), and its single ask is directed at economists, policymakers, and technology leaders: study the economics of transformative AI seriously, and build the incentives, guardrails, and institutions to steer it — before the transformation arrives, not after.

The letter deliberately stops short of specific policy prescriptions. What signatories agree on by signing is the urgency of understanding and preparing — not any particular answer. Where the researchers behind it actually agree and disagree on the substance is exactly what the debate and authors sections of this site map.

What signing does — and doesn't — mean

Signing indicates agreement with the three statements above, nothing more. Signatories hold a wide range of documented positions on how large AI's productivity effects will be, whether job displacement will outpace job creation, and what regulation should look like. This site never infers a person's position from their signature alone.

Where the data comes from

Papers are collected daily from OpenAlex, arXiv, NBER, and the economic reports and blog posts of AI labs and research institutes (such as Anthropic, OpenAI, Google, Ai2 and Stanford HAI). Lab and institute items are not peer reviewed and are labeled that way on every listing; news from geo-targeted news search and public RSS feeds. Each paper and article is evaluated before it is published here, so the site shows a curated subset of what is collected. Researcher positions are coded from the co-authorship network around the letter's signatories. Every database is downloadable on the data page.

How the AI is used — and labeled

Collection scripts never call AI models. A scheduled AI agent (a Claude session) runs the pipeline: it summarizes papers and news, codes researcher positions, and writes the claim notes, profiles, and per-paper explainers. Everything it writes passes a validating merge script before it can appear here, and is marked with an AI-written chip.

Editorial rules

Evidence-cited. Editorial records without sources are rejected at merge time.
Documented positions only. A researcher's stance is coded solely from what they have published or publicly said, always with a citation — never inferred from their signature alone, and never a judgment about them as a person.
Unknown by default. Absence of a coded position means we found no documented evidence — it is not neutrality, and it is not agreement. Every stance display shows its denominator.
Append-only. Records are never silently rewritten; revisions append a new dated record, so every change is in the public git history.

For AI agents and developers

Everything published here can be read programmatically, with the same labels and sources as the pages: a read-only JSON API (OpenAPI description; for example https://api.aifaultline.com/api/search?q=productivity) and an MCP server at https://api.aifaultline.com/mcp that any MCP client (Claude, for one) can add by URL. No key needed. Scores used to curate the collection are not exposed. Setup steps and the full tool list are on API & MCP, CSV and JSON downloads on Data; a plain-text overview is at llms.txt.

Newsletter and privacy

The weekly briefing is sent with Resend. We store only your email address, its confirmation status and dates, use it for nothing but the briefing, and never share or publish it. Every email has a one-click unsubscribe link, and unsubscribing stops all mail at once.

Usage statistics

To learn which pages and sources people use, the site logs anonymous activity (pages viewed, links and buttons clicked, searches, downloads) with PostHog. There are no cookies and no profiles: your browser gets a random id, and nothing you type into a form is recorded except search words. If your browser sends Do Not Track or Global Privacy Control, nothing is logged at all.

Corrections

If something about you or your work is wrong, please open an issue — corrections are applied as new dated records, and the history stays auditable.