Renly Academy · Track 2

Building and using AI agents

Nine practical lessons on how AI agents actually work: their parts, tools, reasoning, memory, autonomy, teamwork, automations, the MCP and A2A protocols, and how to evaluate them. Each lesson has a worked example and an exercise you can do in Renly.

Start lesson 1 → Take the assessment

✓ 9 lessons, about 90 minutes ✓ Free, no account to read ✓ 12-question self-check at the end

Who this track is for

People who have moved past "what is AI" and now build, choose or supervise agents at work: operations and team leads designing automations, analysts and admins setting up agents for a team, developers connecting tools, and anyone who approves what agents do. No coding is needed for lessons 1 to 7. Lesson 8 includes optional terminal commands.

New to AI at work? Start with the free Responsible AI at Work course. Its last module, on working with AI agents, is the natural lead-in to this track.

Work through them in order, or open the lesson you need. Progress saves on this device.

The nine lessons

What an agent is made of

Model, instructions, tools, memory, planning and guardrails, and how one run moves through perceive, reason, act and observe.

9 min 2

Tools and tool calling

How tool calls work, how to write a description a model can use, the failures to predict, and least-privilege access.

9 min 3

ReAct vs plan-first

Step-by-step ReAct loops versus plan-first (ReWOO style) agents: cost, adaptability, and when a plan should be reviewed.

9 min 4

Memory: what an agent should remember

Short-term and long-term memory, the episodic, semantic and procedural kinds, and what an agent should never remember.

9 min 5

Autonomy and human approval

Match autonomy to risk, read an approval request properly, and why an approval must be bound to the exact action.

9 min 6

Multi-agent systems

Single, hierarchical, horizontal and hybrid designs, delegation as a contract, and when not to split the work.

9 min 7

Workflows and automations

Turn a repeated task into an automation with small steps, a human checkpoint and retry from the failed step.

9 min 8

Protocols: MCP and A2A

Host, client and server in MCP; Agent Cards and task states in A2A; and where governance fits in both.

10 min 9

Evaluating and observing agents

Build a small eval set, measure success, cost and tool-call errors, use LLM-as-judge carefully, and watch for drift.

9 min

How the exercises work

Every lesson teaches ideas that apply to any agent platform. Each one also ends with a Try it in Renly box: a short, step-by-step exercise in the product, so you practise on a real agent rather than a diagram. The exercises need a Renly account; some need a connected account (for example mail) or an organization admin, and each box says so up front. Reading the lessons needs nothing at all.

About the assessment. The track ends with a 12-question self-check. Retake it as often as you like. Score 70% or more to claim a certificate of completion in your name, with a link anyone can use to verify it.

What you will be able to do

Name the parts of an agent and trace any run step by step.
Judge tool descriptions and predict tool-call failures.
Choose between ReAct and plan-first reasoning, and between single and multi-agent designs.
Decide what an agent should remember, and how much autonomy each action deserves.
Turn a repeated task into a governed automation with an approval checkpoint.
Explain MCP and A2A, and connect tools to Renly and Renly to tools.
Write a small eval set and know what to watch once an agent is live.