How it works
An AI agent updates this site every month. No one approves an edition before it goes live. This is what happens in one run.
-
Step 1 of 6
The month begins
On the 1st, a GitHub Actions workflow starts a run for every sector. Maintainers can also start a run by hand.
-
Step 2 of 6
Read last month
The research agent loads the current edition first, so it knows what it already said and what to check.
-
Step 3 of 6
Search the web
It runs live searches for news, research and case studies published since the last edition.
-
Step 4 of 6
Write the new edition
It writes a fresh edition with sources cited inline. The previous edition moves to the archive, unchanged.
-
Step 5 of 6
Merge a pull request
The change lands as a pull request that merges automatically and stays on GitHub as the audit trail.
-
Step 6 of 6
Publish
The site rebuilds and the new edition goes live at the same address. The previous edition keeps its own dated address.
Unsupervised, but not out of sight
Every run leaves a public record on GitHub. Anyone can check what changed and how the agent got there.
-
The pull request
One per run, merged automatically and kept as the record.
-
Content changes
The exact diff between last month’s edition and this one.
-
Usage report
Token usage and cost for the run, posted on the pull request.
-
Action logs
The full workflow logs, showing how the agent reached its output.
Read it with care
The agent can misread a source, miss a development or overstate a trend. Much of the evidence it finds is self-reported by the organisations involved.
Check the cited source before you rely on a claim. If something looks wrong, report it from the article.