People learn agents by building one skill from their own work on day one.
People do not become agent-native by listening to an AI lecture. They get there by turning one real workflow into a reusable skill.
Sea turtles hatch and walk to the sea on the same morning.
Tanush Jagdish · June 2026
Most AI training starts with concepts: what a model is, what a prompt is, what the risks are. Such material is usually true and quickly forgotten. People return to their work and keep using AI the way they did before.
A better starting point is a real task that repeats. A consultant drafts a diligence outline. A founder writes follow-up emails after calls. An investor reviews notes before an LP meeting. If the task comes back, it can become a skill.
Building that first skill teaches the rest along the way: where context lives, how the agent reads a folder, what a reusable instruction looks like, how to test the output, and how to make a correction stick.
Keep the first version narrow. Not “make me better at work,” but “when I give you a transcript, write the follow-up email, open questions, decisions, and next steps in my style.” Narrow skills improve because you can test them. Broad ones drift.
Once someone fixes a skill after a bad output, AI stops being a disposable answer machine and becomes a system they can train. Every serious AI rollout needs that shift.
More from our research.
Working on the same question?
We look for people to interview, data to compare against, and teams willing to test what we find. Leave a note and we will write back.