A brief that writes itself

A weekly executive brief built from live government economic data by a team of AI agents, with a validator whose only job is to check the others, and no human involved after setup.

Why I wanted it

I talk a lot about organised data with intelligence on top. I wanted a demonstration you could visit: a polished brief that changes every week because the world changed and the system noticed.

What happened

Every week a job pulls public numbers from the Federal Reserve and the Bureau of Labor Statistics: interest rates, inflation, growth, business formations, job openings, quit rates, wages, layoffs. It saves a dated snapshot and works out what moved since last time.

Then a small team of AI agents reads it. Two write from different perspectives. An orchestrator combines them. A validator reads the result with one job: check that the others did not contradict themselves. Then it publishes.

There is no server running. Everything happens at build time, once a week, on a schedule. Because every snapshot is saved into the project, the project has quietly become its own history of the economy.

A cousin of this project is a newsletter with seven columns, each written by its own specialist author agent, illustrated automatically, then reviewed by an editor agent that can send rewrites back to individual authors. That one keeps a single human gate: me, before it goes out.

What it looks like

A clean page of charts and a few paragraphs that read like a person wrote them. The page also shows the pipeline itself, because the process is the product.

What’s next

The newsletter’s third phase. And a hard look at whether a validator agent is enough, or whether I want a human reading every issue anyway.