The Rule of 7 P's, for AI.

Running a business on your own means agreeing a job with a client, writing the proposal and doing the work. You may hand some of that work to an AI agent. There is still nobody beside you reading the proposal before you send it.

The seven P's stand for “Prior Planning and Preparation Prevents Piss-Poor Performance”, an old British Army expression. For a client job, that means settling what finished looks like before anybody starts. The agent's brief needs to say what finished looks like too.

Other hands have a reason to ask

In a team, a proposal can expose a vague promise because other people have to work from it. Someone needs to estimate the effort. Someone else has to arrange the work. Each has a reason to ask what a line actually commits them to doing.

Suppose a proposal offers to “prepare a report on customer feedback”. The person planning the work needs to know whether the feedback already exists. Reading supplied interview notes is a different job from arranging interviews. The proposal sounds reasonable either way, right up until somebody tries to organise it.

Colleagues miss things too. When the job feels familiar, people fill in the bits nobody wrote down. What helps is having to make a practical decision from the words on the page. A document passed round for a quick nod can remain just as vague.

Working alone, you already have an answer in your head. You can move from writing the proposal to preparing the report without noticing that the answer never reached the page. The client may discover the difference only when the report arrives.

By spec, I mean the written description of the job. Trying to write that description in full can expose a decision you've only made in your head.

The spec is your only colleague.

Done belongs in the client documents

The statement of work is the document that says what you will deliver. The proposal and this document need to say what the client gets, what is excluded and how you will both recognise finished work. The description of done belongs there before production begins.

I'd say writing down what finished looks like is part of agreeing the job. Leave that vague and you're starting with a decision still open, even if you feel ready to get on with it.

In the hypothetical feedback job, suppose the client wants an account of interviews they have already conducted. The proposal can describe the result plainly.

“The report groups the supplied interview notes by theme, links each finding to the relevant notes and marks any disagreement between interviews.”

That line gives the client something specific to change before work starts. Perhaps they wanted advice on which complaints to tackle first. That would need agreeing, because a report of what customers said does not settle what the business should do.

In this example, recommendations are excluded. No new interviews either. What is excluded belongs beside the description of the report. Otherwise “customer feedback” still leaves room for a much larger job.

The statement of work carries the same wording, with the supplied notes identified. It also says how the finished report will be checked: the findings must be linked to those notes, and conflicting accounts must remain visible. A reader can compare the report with the material it covers.

Writing the statement of work can reveal an awkward admission. You might not yet know which notes the client intends to supply. That is a useful thing to discover while agreeing the work.

An agent can fill the silence

An agent given an unclear brief can proceed on its own interpretation, depending on its instructions. Anthropic's prompting guidance includes an instruction to infer the likely action when intent is unclear and continue, using tools to seek missing details. The agent needn't ask.

In the hypothetical report job, “prepare a report on customer feedback” leaves the agent to guess what the report should contain. Suppose the agent groups the notes into themes without linking findings to the notes. The document can look useful and complete while missing something the client agreed to receive.

Unless asking first is part of the brief, the question can come back as a finished report.

An agent can also be directed to question you before proceeding; the guidance on clarifying requirements describes asking it to interview you about unresolved requirements. If the brief leaves a decision open, though, you can't count on the agent asking you to settle it.

The same description of done belongs in the agent's brief. For the report, that includes using the identified notes, keeping findings linked to the notes and leaving recommendations out. The agent may need extra instructions about file formats or where the notes live. Those extra details still need to fit the job you've agreed.

Shortening the brief to “analyse these interviews” would throw away the decisions made in the proposal and statement of work. You would be leaving the agent to choose again, after already settling the choice with the client.

The next proposal starts with what finished looks like

The next job needs a sentence describing its finished result before you start preparing that result. The sentence needs enough detail for the client to say what they disagree with. It's still easy to change the wording before work starts.

In the report example, “links each finding to the relevant notes” is a decision about what the client receives. The client can say they need that connection shown differently. The wording can change before anyone has arranged the findings into a document.

There may still be unknowns. A client might need to locate the material before the report's coverage can be agreed. An unfinished description is useful here because it shows exactly which part of the job is waiting.

Before the next job begins, write the line that says what finished looks like. Put that same line in front of the client and the agent.