How to train your team to direct AI agents

Most AI training teaches people to use a tool. The capability that compounds is teaching them to direct agents — to delegate real work, set the guardrails, and review the output. Here's a practical way to build that across a team.

Using AI vs. directing AI

Using AI is typing a prompt and copying the answer. Directing AI is delegating an outcome to an agent that holds your context, takes the first pass, and routes anything consequential back to you. The first is a personal productivity trick; the second is an organizational capability — and it's the one that keeps paying off after the workshop ends.

The skills that actually transfer

• Writing directives — plain-English instructions an agent can act on repeatedly, not one-off prompts.
• Scoping — breaking a goal into a workflow with a clear definition of done.
• Sign-off — deciding what an agent may do alone and what needs a human gate.
• Review — reading an agent's reasoning and evidence, not just its answer.

A progression that builds capability

People learn fastest when training mirrors how responsibility grows: an individual delegating one workflow, a team lead deploying a shared memory, then leadership standing up ownership roles. That's the shape of the Academy tracks — Spark, Build, and Architect — each ending in a working agent, not a certificate nobody opens again.

Make the learning stick

A one-off ChatGPT session leaves nothing behind. Training sticks when each person leaves with an agent they built and keep using, the accounts stay live after the programme, and the team can run the next session themselves — building internal capability instead of vendor dependence.

Where to start

Pick one workflow with an obvious payoff and a bounded risk — a weekly report, a triage queue, a follow-up cadence — and have one person direct an agent through it end to end. Build trust, then widen. The people side matters as much as the software.