THE NOTEBOOK / THREE SMALL PRINCIPLES

Thoughts in
the margins.

A few principles behind the way I want to build. Short enough to keep in mind when the work gets complicated.

01 / PRODUCT THINKING

Start with the next action.

Before choosing the model, the framework, or the screen, I want to know what someone is trying to do next.

“Answer questions about our documents” is a capability. “Help a new teammate find the right onboarding step” is a direction. That difference changes what needs to be retrieved, what an answer should look like, and whether AI is useful at all.

The first version can be small: one person, one recurring question, one answer they can check. Watch what happens after the answer. That is where the product starts to become clear.

Try this in the workbench

02 / HUMAN × AI

Give automation an exit.

An automation should have a good answer to “wait, that’s wrong.”

That might mean an original document beside an extracted field, a review queue for uncertain messages, or a way to undo a changed label. The useful part is giving the person enough context and control to recover.

I like separating a suggestion from an action. Suggesting a reply is different from sending it. Showing a possible answer is different from updating a record. Each boundary deserves a deliberate decision.

03 / THE SMALL DETAILS

Make the waiting useful.

A system can be doing the right thing and still leave someone wondering whether anything is happening.

A useful waiting state explains what is underway, preserves what the person entered, and gives them a way to stop. If there is a delay, the interface should help them decide what to do with it.