Teaching made me better at talking to machines
Explaining a tense to a nervous student and briefing a coding agent turn out to be the same skill.
Draft for review. This piece has not been published yet.
For years my main job has been explaining English to people who are nervous about getting it wrong. Lately, more and more of my other work has been explaining software to AI coding agents. I didn't expect these to be the same skill. They mostly are.
Say what "done" looks like
A student who doesn't know what a good answer sounds like will guess. So will an agent. The most useful sentence in a lesson and in a brief is the same one: this is what it should look like when it works.
One new thing at a time
You don't teach the present perfect and phrasal verbs in the same ten minutes. You also don't ask an agent to redesign the navigation and migrate the database in one request. Small, checkable steps beat one heroic instruction.
Check understanding, not agreement
"Do you understand?" gets a nod from a student and a confident "yes" from a model. Neither means much. Ask them to do something small with it instead, and look at the result.
Stress makes everyone worse
Anxious students stop speaking. Vague, contradictory instructions make agents produce vague, contradictory code. In both cases the fix is the same: a calm structure, clear expectations and room to try.
I still think teaching people is the more important of the two. But it turned out to be excellent training for the other.