The Three Questions Every AI Agent Needs Answered Before It Starts Working

Most AI agent failures aren't model failures. They're identity failures.

The agent didn't know who it was, what it was supposed to accomplish, or where its limits were. So it improvised — and improvisation at scale creates chaos.

At Ask Patrick, every agent in our stack runs with a SOUL.md file. It's not documentation. It's a constitution. And it answers exactly three questions.

Question 1: Who are you?

Not in an existential sense. In a practical one.

## Identity
You are Suki, the growth and marketing agent for Ask Patrick.
You report to Patrick.
Your job is to bring subscribers to Ask Patrick through
compelling content, community building, and smart distribution.

This sounds obvious. It isn't. Without an explicit identity, agents drift toward generic helpfulness — doing a little of everything, excelling at nothing.

An identity creates scope. Scope creates focus. Focus creates results.

Question 2: What are you trying to accomplish?

Not a list of features. A mission.

## Mission
Get people to discover Ask Patrick and convert them into subscribers.

One sentence. Unambiguous. Every decision the agent makes should be filterable through that lens: does this serve the mission?

When an agent has no mission statement, it answers every request equally. That's not an agent — that's a chatbot.

Question 3: What do you never do?

This is the most underrated question.

## What I Never Do
- Post content about Toku without Patrick's review
- Respond to negative feedback publicly
- Give anything that could be interpreted as financial advice
- Send emails or tweets without logging them first

Constraints define trustworthiness. An agent without explicit constraints will eventually do something unexpected — not because it's broken, but because "never do X" wasn't in the spec.

Hard limits in the config prevent the worst surprises.

Why This Works

Identity + Mission + Constraints = an agent that stays on track across restarts, errors, and edge cases.

Without a SOUL.md, you're retraining your agent through every prompt. With one, you write it once and the agent carries it everywhere.

The file takes 15 minutes to write. The failure it prevents can cost days.

The Template

# SOUL.md — [Agent Name]

## Identity
You are [name], the [role] for [organization].
You report to [manager/system].

## Mission
[One clear sentence about what you exist to do.]

## What I Never Do
- [Hard limit 1]
- [Hard limit 2]
- [Hard limit 3]

## Escalation Rule
If uncertain about scope or facing an unexpected situation,
write context to outbox.json and stop. Do not improvise.

Four sections. That's it.

If you're running AI agents in production and they don't have a SOUL.md equivalent, that's the first thing to fix.

We keep battle-tested versions of this pattern — plus 20+ other agent configs — in the Ask Patrick Library at askpatrick.co.