What is a decision record and why should job seekers know it?

A decision record is the filed, permanent account of a single verdict an AI agent reached: what it saw, which sources it relied on, what it proposed, which bound decided the outcome, and who approved or declined it. Job seekers should know it because reading one is the daily work of AI operations, and employers test it in interviews.

What is in a decision record?

Systems differ in layout, but a governed agent's record usually carries the same core fields.

The record is append-only: entries are added, never edited away. That is what lets anyone reconstruct a decision months later, whether a manager, an auditor or a complaints handler.

Why does it matter to you as a job seeker?

Three reasons.

First, it is the work. Reviewers open a record for every held case. Knowing where to look, and what each field means, is what makes a review fast and accurate.

Second, it is how you are tested. AI operations interviews and work sample tests often hand you a record and ask why the case was held, which source is weakest and what you would decide.

Third, it is your evidence. When you approve or decline, your name and reason join the record. Good reasons build a visible track record; vague ones show too. Employers judge decision quality and the clarity of your decision rationale partly from these entries.

What does reading one look like?

Take a candidate we will call Hassan in an interview for an energy supplier. He is shown a record where a billing agent proposed writing off 90 on a customer's disputed bill. The verdict is refused, and the deciding bound is role scope: write-offs are not part of this agent's job at all.

The interviewer asks why it was refused rather than held. Hassan explains that a held verdict means the agent may draft the action for a person to decide, usually because of an amount or a weak source. Refused means the action is outside the agent's bounds altogether, so it should never have proposed it. He suggests routing the dispute to the team that handles write-offs and telling the agent's owner, since an agent proposing out-of-scope actions may need its brief tightened. That answer shows he can read a record, not just recite definitions.

What do the source labels mean?

Every fact in a record carries a provenance class, strongest first:

1. Sealed: issued by the system of record itself, such as a core ledger or policy library, and tamper-evident.
2. Signed: a named, identifiable party answers for it, such as a verified issuer.
3. Declared: stated by the customer or a person, not verified.
4. External: arrived from outside, such as an emailed invoice or a web page.

An external claim cannot authorise a payment. Spotting a decision that rests only on declared or external facts is one of the most common reasons a reviewer edits or declines.

How do you learn to read records?

Practise on real or sample records, asking the same questions every time: what was proposed, why was it held, which fact is weakest, what would I decide? How to learn to read an AI agent decision record has exercises. Pair this with what depth means and why job seekers should know it, since depth appears in every record, and with the basics in what AI literacy is and whether job seekers need it. Pryme Intelligence Academy's Operator track includes record reading in its review modules. For the full set of skills employers check, see AI skills for job seekers.