Growth Academy

AI Strategy

Stop Telling AI the Answer. Give It the Problem.

By

Your AI agent can execute the exact solution you request. That does not mean your solution was right.

When I open Codex, Claude Code, or another AI agent, I rarely begin by telling it the exact solution to build.

I tell it what I am trying to accomplish.

Sometimes I give it the problem. Sometimes I give it the vision. Sometimes I share the idea I already have and say, “This is what I think, but I may be wrong.”

Then I ask for help.

That small change matters because one of the worst habits in working with AI is approaching it with an answer you have already decided is correct.

The agent may follow your instructions perfectly.

You may still get the wrong result.

Give the agent the problem, the desired outcome, and permission to challenge your first idea.

The agent becomes an order taker

Imagine a business owner says, “Build me a dashboard with four tabs, 12 charts, a daily score, and a separate page for every client.”

The agent can build exactly that.

But the real problem may have been much simpler: the owner cannot tell which three clients need attention this morning.

By prescribing the dashboard, the owner forced the agent to execute one assumed solution before diagnosing the actual problem.

That is how people get technically correct work that does not improve the business.

If you give AI the answer, it will help you execute your assumptions. If you give AI the problem, it can help you examine them.

Three better ways to begin

1. Bring the problem

Describe what is happening, who it affects, why it matters, and what has already been tried.

Our team keeps missing customer follow-ups after estimates are sent. The information is split between email, QuickBooks, and individual notes. We need a reliable way to know who is waiting, what the next step is, and who owns it. Diagnose the likely failure points and recommend the simplest solution.

2. Bring the vision

Describe the outcome you want without forcing the agent into your preferred implementation.

I want every customer to receive the right follow-up at the right time, while the team can see exceptions and overdue items in one place. Give me a practical plan for creating that outcome with the systems we already use.

3. Bring your idea, but keep it negotiable

Your experience still matters. Share your current idea as a hypothesis and ask the agent to improve or replace it.

I think we need a dashboard and a follow-up agent. I may be wrong. Challenge the idea. Tell me what is missing, what is unnecessarily complicated, and whether there is a simpler way to achieve the result.

Ask for the plan before the build

Giving the agent a problem does not mean handing it unlimited authority.

Ask it to investigate first. Have it identify what it knows, what it is assuming, what information is missing, and which options are available.

Then ask for:

  • the recommended approach
  • the alternatives it rejected
  • the tradeoffs and risks
  • the smallest useful first version
  • the actions that require human approval
  • the evidence that would prove the solution is working

Review the reasoning before you authorize the work. That lets the agent contribute judgment without pretending it should own the final decision.

Use AI to improve your thinking, not just execute it

You do not have to arrive empty-handed. In many cases, your industry knowledge will give the agent a much stronger starting point.

The difference is whether your idea becomes context or a cage.

Tell the agent what you believe and why. Then ask what would make the idea fail, which stakeholder you have overlooked, what assumption needs evidence, and what a simpler version would look like.

For an important decision, you can also ask a second model to critique the first plan. Do not choose a compromise automatically. Use the disagreement to expose assumptions, then decide which reasoning fits the business.

A better first message to an AI agent

  1. State the problem. Explain what is happening now and why it matters.
  2. Describe the desired outcome. Say what should be different when the work succeeds.
  3. Add the real constraints. Include timing, systems, budget, permissions, risk, and approval boundaries.
  4. Share your current thinking. Label your idea as a hypothesis, not a commandment.
  5. Invite disagreement. Ask the agent to identify weak assumptions, missing evidence, and simpler options.
  6. Request a plan before action. Review the approach before the agent builds, sends, publishes, deletes, or changes a live system.

The better question

Do not ask only, “Can AI build the solution I already imagined?”

Ask, “What is the best way to solve this problem, and what might I be getting wrong?”

Bring AI the problem.

Keep your answer negotiable.