The reference for governing AI agents, one question at a time.
One question, answered in full, for the people who have to sign off on an agent.
- posts
- 31
- updated
- feed
- rss
Each entry answers one question about running AI agents under control, in full, so you can link it from a policy, a vendor review or a board pack and not need to add a paragraph of your own. They are written for security, risk, compliance and finance teams, not for the engineers building the agent.
The entries fall into four groups. Definitions: what an approval policy, a control plane or a permission manifest actually is. Controls: holds and named approvers, least privilege, segregation of duties, the two-person rule, kill switches and rollback. Risks: prompt injection through documents and invoices, retries that pay twice, drift in confidence scores. Standards: where SOC 2, ISO/IEC 42001, the NIST AI RMF and the EU AI Act reach an agent and where they stop.
Each entry is dated, shows when it was last updated, and lists the laws, standards and vendor documents it relies on, so you can check the source yourself.

PO invoice matching agent: what it should match, and where it must stop
A PO invoice matching agent finds the purchase order and receipt behind each supplier invoice and checks price and quantity against your tolerances. What it should do, the invoices it must hold for a person, and what to ask before you buy one.

RPA vs AI agents for accounts payable
In accounts payable, RPA handles the steps that never change and an agent handles the invoices that do not fit. Which part of the AP process each one should own, and the controls an agent needs before it touches a payment.

Why do AI procurement pilots fail?
AI pilots in procurement rarely fail on the model. They fail on the supplier master, on who may approve what, and on spend nobody capped. Five causes, and what to fix before the next pilot.

Agent autonomy levels: where the human hold belongs
Autonomy isn't a setting for the whole agent. It's a choice you make per action. Put the hold where a mistake stops being cheap to undo.

LangGraph vs OpenAI Agents SDK vs CrewAI vs AutoGen: who gives you an approval step
Four popular agent frameworks, read from their own docs for one question: can a person approve an action before it runs, and what happens to the run while they decide?

Confidence thresholds for AI agents: what the score means, and why it drifts
'Hold anything below 80% confidence' only works if 80% means 80%. What calibration is, why language models often are not calibrated, and how to set a threshold you can defend.

Data residency for AI agents: where your prompts, documents and records actually live
An agent sends your documents to a model, keeps state between steps, and writes a record. Each can sit in a different place. What the main providers' docs say, and what to ask.

EU AI Act obligations for deployers of AI agents, after the 2026 Omnibus
If you use an AI agent rather than build one, you are a deployer. What Article 26 asks of you, which agents count as high-risk, and the dates the 2026 Omnibus moved.

Evaluating agents in production: a golden set for back-office work
How to know your agent still works after a model update, a new rule or a new supplier: a golden set of your own labelled cases, replayed before every change and sampled every week.

Frontier model vendors as suppliers: what to read in OpenAI's and Anthropic's terms
Your agent runs on a model you rent. Treat the model vendor like any critical supplier: price changes, retirement notice, data terms and exit. What their own terms and docs say.

Hold-rate maths: what one hold in ten costs your approver
The hold rate decides whether human in the loop is a control or a bottleneck. How to work out your approver's day from volume, hold rate and minutes per case, with a worked example.

How to read the Governed Agents Reference
What this reference is, what it is not, how the pages are built, and how to tell our opinion from the sourced facts.
