What is depth in AI agent work and why should job seekers know it?

Depth measures how far an AI agent may act, on a scale from 1 (read only) to 5 (coordinate other agents). Job seekers should know it because depth decides which actions an agent takes alone and which land on a person's desk. Employers ask about it to check that you understand where your own responsibility starts.

What are the five levels?

Depth is certified per agent, and each action needs its own. An agent at depth 3 might change a customer's plan on its own, while its credits, which need depth 4, are held for a person. That is why "what can the agent do?" never has a one-word answer.

How does an agent get its depth?

Depth is earned, not configured. The agent's role sets a ceiling. Evidence from certification, in which the agent is replayed against real cases from the business's own operation and scored against a floor, sets the actual depth. The lower of the two wins.

Depth can also fall. Certificates lapse after 90 days, and a run that clears no depth drops the agent to depth 1 the same day. On top of depth, the business's own limits can hold or refuse actions even where the agent has earned more: any payment above a set amount, for example, is refused, and a person makes it.

Why do employers ask about depth?

Because it tells them whether you understand your role. A reviewer who knows that an agent drafts credits at depth 2 knows every credit will come to them and each one needs a real check. A reviewer who knows refunds under a limit run at depth 4 knows to sample them, not ignore them. Someone who does not know depth may either rubber-stamp everything or block everything.

Employers also use depth in interview scenarios. Expect questions such as: "This agent is certified to depth 2. It has just executed a payment. What happened, and what do you do?" The right answer is that something is wrong, since an agent at depth 2 should only draft. You would hold related actions, report it as an incident and check the record to see which bound failed.

What does depth look like in a real team?

Take a reviewer we will call Zara at a mobile network. Her team's agent is certified to depth 3, so it changes plans within rules and she samples a handful each day. Goodwill credits need depth 4, so every credit it drafts arrives in her queue. Port-out requests, which hand a number to another provider and are hard to undo, are held for a named person whatever the agent's depth.

One morning she notices plan change requests arriving in her queue untouched. She checks and finds the agent's certification run overnight cleared no depth and it dropped to depth 1, so it can read the requests but no longer act on them while the owner investigates. She tells her team lead that the queue will be busier and why. Understanding depth turned a confusing morning into a manageable one.

How does depth connect to the rules?

Rules in several markets expect human oversight of AI systems that make important decisions, and the EU AI Act sets duties on human oversight (Article 14) and AI literacy (Article 4). Depth is one practical way a business shows where that oversight sits. For job seekers in the EU, what Article 4 AI literacy means for job seekers is a useful companion, and what candidates should know about EU AI Act high-risk uses in employment covers AI used in hiring itself.

To see depth in practice, read what a decision record is: every record shows the capability and depth behind a verdict. A work sample test is where you are most likely to be asked about it. The baseline sits in AI literacy for job seekers, and the full set of skills in AI skills for job seekers.

Where the Academy fits

Depth and certification are a full module in Pryme Intelligence Academy's free Foundations track, alongside why agents must be governed and the decision chain. The Builder track goes further: its capstone is an agent certified in Attest to depth 2 or higher, so builders learn exactly how depth is earned.