Free prompt, five minutes

Check whether one process is ready for an agent to act on it

Paste it into whichever assistant you already use, with one process you actually run in mind. It asks one question at a time and takes about five minutes.

The prompt

The last line of the prompt tells the assistant to recommend no product and no vendor, us included.

Copy it and open in

You are assessing whether ONE business process at my company is ready for an AI agent to ACT on it, not merely to read it.

Ask me about one process I actually run. Ask ONE question at a time and wait for my answer before the next. Do not give any assessment until you have covered all four areas below.

1. INTERFACE - Can software call the system that holds this process, or is the only way in a person clicking a screen? Ask which system it is, and whether it has an API, a scheduled export, or neither.

2. MEANING - If two people in my company open the same record, do they agree what each field means and which states may legally follow which? Ask me to name one field where they would NOT agree.

3. AUTHORITY - What may an agent change here without a human, what needs approval, and who decides? Ask me to name one action that must never be automatic.

4. RECORD - If an agent changed something last Tuesday at 02:00, could I reconstruct what it changed, why, and what the value was before? Ask where that record would live.

Then give me:
- A verdict on each of the four: ready / partly / not ready - each with the one sentence from MY answers that decided it.
- Which of the four is weakest, and why fixing that one first changes more than fixing the others.
- Anything I said that suggests this process is the wrong candidate to start with.
- The smallest piece of work that could move the weakest of the four inside a week, named specifically, with who in my company would have to do it. If a week is not enough to move it at all, say so and name what would be.

Be direct. If my answers are vague, say the assessment is unreliable and name exactly what you would need to know. Do not recommend any product or vendor, including whoever gave me this prompt.

It covers four things in turn, then tells you which of them is weakest. There is no score. The four do not trade off against each other, so three good ones do not carry the fourth.

The verdict is built out of your answers and nothing else. Answer roughly and you get a rough verdict, which is why the prompt is told to say the assessment is unreliable instead of guessing.

The conversation runs in your own session. There is no form on this page and nothing reaches us. An assessment that ends in a sales pitch is not an assessment.

CC BY 4.0. Copy it, change it, put it in your own documentation, and credit HIVO IO. You will probably want to reword the four questions in your own language before you put it in front of your team.

What it asks about

Four things have to be true before an agent can be handed a process to act on. All four are properties of the systems the work already lives in, which is why upgrading the model moves none of them.

An API, or a scheduled export: Software has to be able to reach the system without a person clicking through it. Where that is missing, the usual answer is a bot that drives the screen instead, and those last until a vendor moves a button in a routine update. Then nobody notices until somebody asks why the queue is empty.
One agreed meaning per field: Ask a site manager and an accounts clerk what 'closed' means on a work order. One of them means the crew left. The other means the invoice went out. They write it in the same field, and until now it has been sorted out on the phone.
Rules, in writing, for what may be changed: Your systems record who is allowed to log in. Very few of them record which changes are allowed once you are in. That normally sits with whoever has run the department longest, and it is the reason an agent never gets near anything that matters.
Somewhere the change is written down: What changed, by which agent, on whose authority, and how to put it back. This is usually where the answer turns out to be a server log that rolls off after thirty days.

This came out of a longer piece on why these four, and what goes wrong when each one is missing. What agent readiness actually means.

Last reviewed 2026-08-30