I'm so excited to launch prognostibench.com today.

We find ourselves in a tumultuous information environment, in which claims, takes, and predictions about how AI will or won't change the world are constantly swirling. Every week, sometimes every day, a viral Substack post or X post moves markets, launches multiple podcasts as if they exist in some infinite untapped well just beneath the internet's surface. It can feel like you're drowning.

PrognostiBench is an attempt to track the most testable claims, attach a timeframe to them, whether they are pessimistic or optimistic or neutral, and to suggest the kinds of data interested parties might start tracking if they wanted to test them. In a nutshell, it's trying to turn prognostications into hypotheses.

Confirmation bias is such a powerful force and it can sometimes feel like all of these prognostications are either foretelling certain doom or an endless age of golden opportunity. This can really drive down (or up!) one's own sense of agency over the moment in which we find ourselves. I'd prefer to keep myself grounded and I suspect you might as well. At the very least, hopefully PB is a useful resource for researchers looking for ideas.

The name is a nod to the rapidly proliferating set of benchmarks and evals that attempt to measure model behavior, including my newest favorite, BullshitBench. Importantly, I'm not actually keeping score or producing a metric. I think it's important to measure things, and for people to try and calibrate and correct their predictions, but many of these will take some years to test.

I'm launching PB first to this list and more broadly later this week. Please send your feedback, submit claims that you feel are worth tracking, and feel free to share widely!

Below are the commonplace posts you may have missed since my last digest.

Trey

This week's posts

• 2026-02-17 — The left is missing out on AI
• 2026-02-21 — Quoting json on X
• 2026-02-21 — Jared Sleeper on Which Software Companies Will Survive the 'SaaSpocalypse'
• 2026-02-23 — Getting promoted to a tech or product you don't know requiring skills you don't have ...
• 2026-02-25 — 西村直晃 Nishimura Naoaki 💤 on X
• 2026-02-28 — An AI agent coding skeptic tries AI agent coding, in excessive detail
• 2026-03-01 — We are Changing our Developer Productivity Experiment Design - METR
• 2026-03-02 — Clawed - by Dean W. Ball
• 2026-03-06 — Quoting Joey Politano 🏳️‍🌈 on X