Skip to content

Notes from the work.
Field-tested takes on what we're seeing.

Strategy

The two kinds of AI use (and why companies confuse them)

Half the AI arguments inside a company are two people using the same word for different things. One kind helps almost anyone tomorrow; the other is software.

Evaluation

What's changed in our recommendations this quarter

The densest quarter on the model release calendar since GPT-5 shipped. One read of what shifted, what stayed, and what it changes for anyone making AI decisions right now.

Agents

What 'agentic' actually means in 2026

The word lost most of its meaning in 2024 and has been slowly recovering it. A working definition, three failure modes, and a checklist.

Agents

The agent question is the wrong question

The most common opening question we get from buyers in 2026 is 'should we build an agent?' It's the wrong question, and the right one is harder.

Local AI

Why local AI is real now (and where it isn't)

The honest answer about local AI changed sometime in the past year. Here is what runs locally now, what doesn't, what 'local' really means, and who actually benefits.

Strategy

What we tell first-time AI buyers when they ask where to start

The conversation we have with first-time AI buyers on a discovery call. Start with one bounded workflow: not a tool, not a budget, not an agent.

Adoption

The unglamorous half of every AI engagement

The demo is always good. Half of every AI engagement is the unglamorous work nobody demos, and that half decides whether the system survives.

One useful email per week

Get new writing in your inbox.

New essays, field notes, and short observations on what we're seeing in applied AI. One email a week, no filler, unsubscribe with one click.