Managers Don't Need More AI News. They Need Precedent.
So we pulled 159 real AI deployments out of a year of reporting, named every organization, linked every source, and kept the six that got cancelled.
Every week we ship what's new in AI. New models, new funding rounds, new launches. It's the job and I'm not knocking it. But new is the wrong axis for the person who actually has to decide something.
A lot of our readers are the ones being asked to put AI somewhere real: a claims desk, a loading dock, a support queue, a grading workflow. When that person goes looking, they're not asking what's new. They're asking a much older question. Who already tried this in my industry, and what happened to them?
That question has been weirdly hard to answer. The information exists, but it's scattered across a year of headlines, most of them written to sell the future rather than report the result. So we built the thing we kept wishing existed.
What it is
The AI Use-Case Library is a searchable database of real, named-organization AI deployments. 159 of them so far, across 21 industries. Each record names the organization, the function, the tools and vendors, the reported outcome where there is one, the current status, the date, and a link to the source it came from. You can filter by your own industry and job function and read only the deployments that came with a measured result. 77 of the 159 did.
Two rules kept it useful. Every entry is a named organization doing a real thing, so a vendor announcing a product with no customer attached doesn't make the list. And outcome numbers are quoted from the source, not estimated by us. When a record says 97.9% of OpenAI's staff now use its Codex agent, or that Uber burned through its entire 2026 AI budget by April, that's the source's figure, and the source is one click away in the row.
We kept the failures
Six of the entries are projects that got halted or reversed, and they're in the library on purpose.
Ford is rehiring 350 quality inspectors after automated inspection missed defects that only experienced humans reliably caught. Waymo paused robotaxis in four cities after vehicles kept driving into flooded roads. Meta pulled back its automated hate-speech moderation and abusive posts targeting legislators tripled within six months. A North Carolina school district banned AI detectors after a flagged student appealed a zero and had the grade changed to 100. A community college's AI name-reader skipped dozens of graduates and got switched off mid-ceremony. And a federal judge ruled DOGE's use of ChatGPT to cut about $100M in grants unconstitutional after the model tagged a Holocaust literature anthology as DEI-related.
A library of only the wins would be a brochure. The cancelled projects are where the real lesson lives, because they show you which deployments looked clean on a slide and came apart in contact with the actual work.
Where it goes
159 is a start. The corpus we pulled from supports several hundred, and we're adding new deployments every week as they cross the wire. Per-industry pages and a few other cuts are next.
For now, go find your own row. If you run operations in insurance, or QC in a plant, or a support team, there's a decent chance a named company in there already did the thing you're being asked to scope, with a link to how it went.
Browse the AI Use-Case Library →
— Alexis