AI Weekly · The public archive
Field notes for people who ship AI.
Every issue turns research into one architecture decision and one experiment you can run this week. Subscribers read it seven days first.
Latest public issue
The future of AI isn't bigger models, it's finer control over the ones we already have.
New research shows how to make models more efficient and predictable without retraining.
Read the issueAll issues
Page 1 of 1
- 02The evaluation gap is where AI agents go to dieWhy your metrics look good but your agents fail in productionRead issue
- 03The best AI systems don't predict, they intervene.From protein folding to traffic lights, one primitive keeps showing upRead issue
- 04The era of blind compute is overMoving from idealized math to real-world efficiencyRead issue
- 05We are optimizing for speed while the foundation rotsYour eval pipeline is broken and your model is blindRead issue
Read the decision while it is still timely.
The website is the library. The inbox is the early-warning system—every Tuesday, seven days ahead of the archive.