Writing
Guides to the questions teams ask about agentic engineering, and essays on what agents change about software work and how anyone can tell whether the work is right.
Guides
What is agentic engineering?
Agentic engineering is building software with AI coding agents that do real work, while the team's own tests, gates and review rules decide what merges.
4 min read
How do you make sure AI agents build what your company decided?
Write decisions down as a cited record your people approve, check agent work against it before it acts, and let drift be proposed back to a person. Passing tests is not enough.
3 min read
How can a venture fund tell whether its portfolio is getting faster with AI?
Read every company under the same rules, from a frozen baseline, combined by default, with company-level numbers shared only by opt-in. Self-reported figures cannot be compared.
3 min read
How do you measure whether AI coding agents are actually working?
Freeze a baseline of delivery before agents change anything, then read agent-native measures against it under rules that do not move. Usage counts alone cannot tell you.
6 min read
What should be in place before AI coding agents merge code?
Tests agents cannot talk past, gates that hold their changes to those tests, written conventions, small changes, matching review rules, a safe way back and a spend budget.
4 min read
How do you measure what AI coding agents cost per change?
Divide agent cost by the pull requests that merge, then read it beside cycle time and rework. Spend per merged pull request measures spend that lands, not spend per seat or per line.
3 min read
Essays
The Rails Were Never the Railroad
After the railroad booms, the fortunes went to the people who bought bankrupt lines at the foreclosure sale, tore out the boom-era iron, and kept the route. What held value through that cycle, and which part of an AI stack is actually yours.
9 min read
Who Got Rich Off the Punch Cards?
The card-punchers earned wages. Jacquard got a pension. Hollerith owned the reader, the format, and the count. He became IBM. Where value actually settles in a translation layer.
5 min read
Specs Live in Incidents
The real specification of an operational business is not in its documentation. It is in its postmortems, its exception queues, and the heads of its tenured operators.
4 min read
The Pattern Cards: How the Spec Became the Source Code
Amazon wrote user manuals for software that didn't exist. Cursor's agents rebuilt SQLite from its manual alone. The spec was always the source code; humans compiling it by hand was the temporary part.
12 min read
Get the next essay
New writing, straight to your inbox. No spam, unsubscribe anytime.