Notes on expertise, work, and the machines reshaping both.
Where we think the value is moving as knowing gets cheap and work gets atomized — and why human judgment becomes more valuable, not less.
Follow new releases
Trust, but Triangulate: Validating AI When the Web Echoes
A model trained on the internet inherits the internet’s failure modes: ideas that are true because everyone repeats them, and questions where the honest answer is that qualified people disagree. Here is how we decide what to trust.
From the Right Question to a Simulated Outcome: The Decision Loop
Most tools stop at retrieval — they fetch an answer and call it done. The harder, more valuable system runs a full loop, from generating the right questions to simulating what happens if you act on them, and re-running as reality returns signal.
AI Is Redrawing the Line of Expertise
A model can now recall, summarize, and reason across more material than any human alive. So what, exactly, is an expert for? The answer is not less — it is sharper.
From Careers to Jobs, From Jobs to Tasks
Each decade, the container we pour work into gets smaller. Understanding that trajectory is the only way to see where durable value is actually pooling.
Expertise Is the Thread
Automation pulls work apart into pieces. Something has to run through those pieces and make them cohere. That something is expertise — and it is the most valuable thread in the system.
The Verification Economy
A model can generate a confident answer to anything. That makes the answer worthless on its own — and makes the ability to prove it the thing worth paying for.
Tacit Knowledge Doesn’t Live in Documents
Train a model on everything ever written and it still won’t know what your best operator knows. The decisive knowledge was never text.
The Shrinking Half-Life of a Skill
The specific thing you learned to do is obsolescing faster than ever. The durable asset is the judgment underneath it — and that’s worth investing in.
Why “I Don’t Know” Is an AI’s Best Output
Confident fabrication is worse than no answer, because it costs you the most exactly when you can least afford it. Honest uncertainty is a feature, not a failure.
The Return of the Generalist
Specialization was a response to scarce knowledge. When recall becomes free, the advantage swings back to the person who can connect across domains.
Pricing a Minute of Expertise
The old way bundled an expert’s decisive minutes with hours of overhead you didn’t need. Unbundle it and the valuable minutes can finally be priced for what they’re worth.
What we’re writing next
The Expert as API
What changes when a person’s judgment can be called like a function — on demand, for one question.
From Headcount to Knowledge-on-Demand
Why the next org chart has fewer full-time hires and a deeper bench of reachable experts.
What Provenance Looks Like in Practice
Source-tagging every claim sounds simple. Here is what it actually takes to build.
Owning Your Own Knowledge
A consent ledger for experts: how to share what you know without losing control of it.
When Should an Agent Refuse to Answer?
Drawing the line between helpful inference and irresponsible guessing.
The End of the Billable Hour
If value concentrates in minutes, the hour stops being the right unit to sell.
Follow the thinking as it develops.
One essay at a time, no noise. Just where we think expertise and work are headed.