Notes on work · July 20, 2026

How do you decide what not to automate?

The most useful question in automation isn't what a tool can do. It's what your work is actually made of.

In the clinic, I learned to separate what a test could measure from what only a conversation could surface. The measurable part was never the whole patient. The same split runs through every workflow I've mapped since.

What are the three kinds of work inside every workflow?

Every workflow contains three different kinds of work. Judgment: decisions that depend on context, trust, and consequences, like pricing an unusual job or telling a client hard news. Exceptions: the weird cases that don't fit the normal path and need a human to notice they're weird. And repetition: the copying, reformatting, re-entering, and routing that happens the same way every time. Automation is for the repetition. Judgment and exceptions need people, and the workflow should be built to hand them to people quickly.

Automating judgment is how teams end up with systems nobody trusts.

Why do automated systems lose a team's trust?

A system loses trust the first time it confidently does the wrong thing with a case that needed a human. One bad automated email to an upset customer costs more than a hundred correctly routed ones saved. The people who do the work remember it, and then they quietly stop using the system and go back to the spreadsheet. The automation didn't fail because the technology was weak. It failed because someone automated a decision instead of a task.

I'd rather ship a smaller system your team actually trusts than an impressive one they route around.

So what should you automate first?

Start with repetition that happens weekly or daily, follows the same steps every time, and doesn't require anyone to weigh consequences. Data moving between two tools by hand. The same status update assembled from the same three sources. Intake information retyped from a form into a record. These are boring wins. Boring wins compound.

Before you buy anything or build anything, sit with the people who do the work and sort one workflow into those three piles: judgment, exceptions, repetition. If you want a second set of ears for that conversation, that's exactly what we do. Book a workflow conversation and bring the messiest workflow you've got.

Book a workflow conversation

Start with the work, not the software.

Book a workflow conversation