Topic
Model choice
Routing across frontier models instead of betting on one.
Anthropic's S-1: a preliminary read
Anthropic filed publicly to go public on Oct 26. A first read of the numbers, the margins, the compute commitments, the governance and the risk factors, and what the filing still leaves blank.
Early October 2026: AI gets a budget
The month in AI, first half of October. What business leaders should take from two weeks of AI news.
The model is eating the harness
Work that AI systems once had to build around the model keeps becoming part of the model. What that pattern looks like, why it keeps happening, and how to build systems that get better with every release.
How to measure applied AI
A model's benchmark score says little about how it will do on your work. A practical playbook for measuring AI inside an enterprise: real cases, outcome metrics, repeated runs, one scorecard and a loop that keeps measuring after launch.
The Bending Spoons playbook
Bending Spoons runs more than 50 acquired software businesses on one shared platform with a core team of about 600. What the platform contains, why AI changed its economics, and how any roll-up or private equity firm can build its own.
September 2026: the month agents moved in
The month in AI. Always-on agents went mainstream, frontier models got cheaper and harder to get at the same time, and safety stopped being theoretical.
Building an AI roll-up platform
Private equity invented buy-and-build. The AI roll-up is being built mostly outside it. How a private equity firm can run the model, from choosing the vertical to the platform, the people and the exit.
The private equity AI playbook
What a fund should actually do with AI at each stage of an investment, from diligence through exit, and how to build AI EBITDA that survives the next buyer's diligence.
Choosing an AI partner for your portfolio
Lab-owned deployment firms, consultancies, integrators, platform partners or building it yourself. Four questions that separate them, and why model choice matters over a seven-year hold.
Your advantage is already on campus
Access to capable AI is now ordinary, which moves the question from what an institution can buy to what it already has. Four things a university holds that nobody can copy, and how each becomes a system.
What to build, and what to buy
Enterprises are buying most of their AI and worrying about what they bought. The build-or-buy line does not run between layers of the stack. It runs through every one of them.
Why evals matter
The public benchmarks stopped measuring anything useful this year. The instrument you build for your own work is the one that decides everything else.
Should I train my own model?
Where the model market stands this month, what each class is for, what the industry and the law constrain, when post-training pays, and how Fig thinks about it.
Flash or Thinking: how Fig routes every task
Two lanes that outlast any model, the signals the router reads, how a model earns traffic, and when to overrule it.
Sovereign AI: the operating playbook
Sovereignty is a capability, not a location. Five rights, a ladder of infrastructure routes, and four proofs to run before production.