Start with three ideas: why AI agents must be governed, how the decision chain decides what an agent may do, and what depth means. Agents differ from chat tools because they act, so learn first how their authority is limited and checked. Prompting and technical detail can wait.
Why learn governance before anything else?
An AI tool that drafts text leaves you to decide what to do with it. An agent goes further: it can change a record, send a message, issue a credit or move money. Mistakes have consequences for real people, and some cannot be undone.
So the first question about any agent is not "how clever is it?" but "what is it allowed to do, who decided that, and how do we know it decides well?" Governed agents answer those questions by design. Authority is granted per agent, proven on the business's own cases, enforced at the moment of action, and it expires. Once you see agents this way, job adverts, interview questions and the work itself make more sense.
What is the decision chain?
Every action a governed agent proposes meets five bounds at once:
- Role scope: is this part of the agent's job?
- Certificate: has the agent earned the depth this action needs?
- Sources: are the facts strong enough? Sealed, signed, declared and external sources carry different weight.
- The business's own limits: amounts, customer types or actions the business always wants a person to see.
- Platform guardrails: the floor no business can lower.
The lowest bound wins, and the verdict names which one decided. There are four verdicts: allowed, held for a named person, refused as outside the agent's bounds, and withheld at the platform floor for every agent. Every verdict is filed, so any decision can be reconstructed later.
What is depth?
Depth is how far an agent may act with a capability: 1 read, 2 draft for a person, 3 execute routine work within bounds, 4 move money or take hard-to-undo actions, 5 coordinate other agents. It is earned on evidence, not switched on, and it can drop: certificates lapse after 90 days and a run that clears no depth sends the agent back to depth 1 the same day.
Depth tells you where people fit. At depth 2, every draft reaches a person. At depth 3, people sample and handle exceptions. Depth 4 actions above a limit are refused: the agent may only draft them for a named person.
What comes next, in what order?
| Order | Topic | Why then |
|---|---|---|
| 4 | Source classes | You need them to judge any held case |
| 5 | Reading a decision record | It shows the chain and depth in action |
| 6 | Red lines and saying no | The core of review work |
| 7 | Writing reasons | How your decisions become useful to others |
| 8 | Prompt injection and manipulation | How agents are attacked through what they read |
| 9 | Your own field's rules | Where your judgement applies |
How to learn to read an AI agent decision record covers step 5. Learn to say no to an AI agent before you learn to use one explains why step 6 comes before tool skills. How to learn to spot prompt injection covers step 8.
What happens if you learn in the wrong order?
Take a learner we will call Ahmed, an administrator who wants to move into AI operations. He spends a month on prompting courses and can get impressive drafts from a model. In his first interview he is shown a record of a held payment and asked why it was held and what he would do. He talks about how he would rephrase the agent's instructions.
The interviewer wanted to hear that the payee was added from an emailed instruction, an external source that cannot authorise a payment, and that he would decline and confirm with the supplier. Ahmed's month was not wasted, but it answered a question nobody was asking. Starting with governance, the chain and depth would have prepared him for the one they did ask.
How long does the first stage take?
Not long. The three core ideas can be learned in a few hours, and a structured course covering them, such as Pryme Intelligence Academy's free Foundations track, takes about four. How long it takes to learn AI agent skills gives a fuller picture, and how to learn AI agent skills for free lists ways to do it without cost. The skills employers then ask you to prove are in AI skills for job seekers.
Questions people ask
Do I need to understand how models work first?
No. It is interesting, but it does not tell you what an agent is allowed to do or how to review it. Learn the governance first and the technology later if you want it.
Is prompting worth learning at all?
Yes, for using AI tools in your own work. It is a different skill from supervising agents that act, and less often what operations employers test.
What if my employer's system uses different terms?
The ideas carry over even when names differ. Look for how the system limits authority, what it holds for people and what it records.