What is a hands-on lab?

A hands-on lab is a practice environment where learners carry out real tasks in a working copy of a system, with sample data and no effect on real customers. For people learning to oversee AI agents, a lab lets them approve, decline and stop agents, and see what the record shows, before any of it counts.

Why do hands-on labs matter for agent oversight?

Oversight is a skill of doing, not knowing. A person can describe the four verdicts perfectly and still freeze when a held payment with a mismatched payee name lands in their queue at five o'clock. Reading about it does not build the reflex; doing it does.

Labs also let people make the mistakes that matter. In a lab, a learner can approve an action based on an external claim and then watch what happens. In production, that mistake costs money and lands in the record with their name on it.

How does a hands-on lab work in practice?

A good lab has four parts:

1. A real workspace. The same product the team will use, not screenshots or a simulation of it.
2. Seeded items. Sample held actions built to teach something specific, including at least one trap.
3. A task. "Clear this queue of eight items, recording why for each."
4. A debrief. The learner compares their decisions with the expected ones, and reads the record to see which bound decided each verdict.

Labs work best in short bursts, 20 to 40 minutes, tied to one concept.

What does a hands-on lab look like in a real team?

Take an accounts payable team preparing for an agent that drafts supplier payments. The lab seeds ten held payments. Seven are routine. One has bank details from a forwarded email. One exceeds the customer's own limit. One looks routine but cites an old contract. Learners clear the queue, then a team lead walks through the three traps.

The same lab environment is used later for an incident drill for the accounts payable team, where the task changes from reviewing to stopping an agent that has gone wrong. Labs on the workspace's summary view support teaching a team to read the six Overview numbers.

How is a hands-on lab different from a lab exercise?

The hands-on lab is the environment and the session. A lab exercise is a single task set inside it. One lab session might contain three exercises: clear a queue, find a past decision in the record, pause an agent.

A lab is also different from a demonstration. In a demo, the trainer drives and the learner watches. In a lab, the learner drives, and the trainer only steps in afterwards.

Which related terms should you know?

• Sandbox: an isolated environment, often used to host labs.
• Decision journal: a log learners can keep during labs to compare their reasoning with the expected answer.
• Simulated queue: a set of seeded items that imitates a real day's workload.

Labs are also where builders practise. The case that training is how an agent earns its next depth depends on people who have tried changes in a lab before trying them for real. In Pryme Intelligence Academy, every lab runs in a free Build workspace. The topic guide on training teams for AI agents shows where labs fit.