Insights

Notes from the work, not the conference circuit.

Short pieces on decisions we have had to make in production: what to automate first, what to leave human, and what breaks after launch.

Latest · AI & automation9 min

The first workflow you automate should be the boring one.

Teams usually pick the workflow with the most visibility. The one worth automating first is the one that runs the same way every time, has a clear owner, and fails loudly when it goes wrong.

Read the note

6 notes

AI & automation
Human-in-the-loop is a product decision, not a safety net
Where the review step sits determines whether people trust the system or quietly route around it.
August 2026 · 6 min
Product engineering
Ship billing before you ship the dashboard
An MVP that cannot take money is a demo. The order in which we build the first four systems, and why.
July 2026 · 7 min
Scale & modernization
The deploy nobody wants to press
Release fear is a measurable problem. What we instrument first when a team has stopped shipping weekly.
July 2026 · 8 min
AI & automation
Evaluation sets are cheaper than incidents
Forty labelled examples written in an afternoon catch more regressions than a month of prompt tuning.
June 2026 · 5 min
Product engineering
What we put in a written scope
The exact sections of the document a client gets after discovery, including the out-of-scope list.
June 2026 · 6 min
Scale & modernization
Decompose the part that blocks the deploy, not the ugliest part
Legacy rewrites stall when they start with taste. Start with whatever gates the release instead.
May 2026 · 9 min

One note a month, when there is something worth sending.

No newsletter cadence for its own sake. We write when a project teaches us something we would have wanted to read first.

Notes from the work, not the conference circuit.

The first workflow you automate should be the boring one.

All notes

Human-in-the-loop is a product decision, not a safety net

Ship billing before you ship the dashboard

The deploy nobody wants to press

Evaluation sets are cheaper than incidents

What we put in a written scope

Decompose the part that blocks the deploy, not the ugliest part

One note a month, when there is something worth sending.