← All resourcesResources

Agent labs

Reading about agents only gets you so far. Start by testing whether the task suits an agent at all. Then the three labs let you build one, then a team, then attack it, with no code and nothing saved. Each ends with the governance record an IMDA or ISO 42001 review would expect.

Usually opened by Managers who will approve agents, and teams about to build their first one.

4 tools in this group, all free with a printable result

  1. Assessment

    TRACE Agent Evaluation

    Is this task safe to hand to an autonomous agent at all?

    When to use it
    Before you build. Run it on the task before any of the labs.
    What you get
    A deployment, conditional or not-yet verdict across five criteria.
    Effort
    5 criteria
    Last updated
    29 Sept 2026
    Evaluate your taskBackground reading: From Human-in-the-Loop to AI-on-the-Loop: Redesigning Oversight Architectures
  2. Lab

    Build Your Own AI Agent

    How does an AI agent actually work, and what makes one safe to launch?

    When to use it
    Start here. No technical background needed.
    What you get
    A working agent with a job, tools and rules, checked against IMDA and ISO 42001.
    Effort
    7 guided steps, about 10 minutes
    Last updated
    29 Sept 2026
    Build your agentBackground reading: Critical Tasks Suitable for Agentic AI
  3. Lab

    Advanced AI Agent Builder

    What changes when agents work as a team?

    When to use it
    After the first lab, when multi-agent designs are on the table.
    What you get
    A Supervisor and Specialist team that plans, pauses for approval, recovers from failure and leaves a governance record.
    Effort
    5 guided steps, about 10 minutes
    Last updated
    2 Oct 2026
    Build your teamBackground reading: Eight Practical Types of AI Agents Emerging in Real Systems
  4. Lab

    Expert AI Agent Builder

    Would our guardrails hold against a real attack?

    When to use it
    Before production, when security and audit need to see the defences tested.
    What you get
    Guardrails red-teamed with five attack techniques, and a governance record to hand to an auditor.
    Effort
    5 guided steps
    Last updated
    2 Oct 2026
    Harden your teamBackground reading: How to Test an AI Agent Before You Trust It With Real Work

Other groups

Work through it together

The workshop turns these diagnostics into a half-day session with a benchmarked scorecard and roadmap. Or book a private session on your specific situation.

Book the WorkshopBook a 1:1 Session
Terence Kok
Before You Go

I built these labs for the managers who will be asked to sign off an agent without ever having built one. Twenty minutes in the first lab changes the questions you ask.

Terence Kok