I keep having the same conversation.

Someone tells me they’ve been going back and forth with an LLM for an hour. The output is wrong. It keeps misunderstanding what they want. They’re ready to write the whole thing off as overhyped.

So I ask what they started with. And it’s always some version of “I told it to build the thing.” No requirements. No constraints. No questions asked or answered. Just “go.”

And then they’re surprised when it goes in the wrong direction.

The problem isn’t the model. The problem is nobody planned anything.

The SEAL Mantra⌗

There’s a saying in the Navy SEALs: “Slow is smooth, and smooth is fast.”

The idea is simple. If you rush, you fumble. You miss things. You backtrack. You waste time fixing mistakes that shouldn’t have happened. But if you slow down - move deliberately, think through the steps - everything flows. You end up faster than the person who sprinted out of the gate.

This applies directly to working with LLMs.

The people I see getting the most out of these tools aren’t the ones firing off prompts as fast as they can type. They’re the ones who spend the first hour not building anything at all. They’re asking questions. Defining the problem. Getting the LLM to poke holes in their assumptions before a single line of code or a single paragraph gets written.

Plan Mode Exists. Use It.⌗

Most serious AI tools now have some version of a planning mode. Claude has it. Codex has it. The feature is right there. Almost nobody uses it.

In plan mode, the LLM doesn’t execute. It thinks. It asks you questions. It maps out approaches. It identifies edge cases and tradeoffs before anyone commits to a direction.

This is the part people skip. And it’s the part that matters most.

Here’s what a good planning session looks like:

  • You describe what you’re trying to accomplish - not how to accomplish it
  • The LLM asks clarifying questions you hadn’t considered
  • You go back and forth on scope, constraints, and priorities
  • You end up with a clear plan that both you and the LLM agree on
  • Then you execute, and the execution is clean because the thinking already happened

Compare that to the alternative: you type a vague prompt, the LLM guesses at what you want, you correct it, it guesses again, you correct it again, and thirty messages later you’re frustrated and starting over.

The first approach feels slower. It’s not.

This Isn’t Just About Code⌗

I’m using development examples because that’s where I see it most, but this applies to anything complex you’re doing with an LLM.

Writing a content strategy? Plan first. What’s the audience? What are the constraints? What does success look like? Get all of that nailed down before the LLM writes a single word.

Researching a topic? Plan the research. What do you already know? What are the specific questions? What would change your mind? Let the LLM help you build a framework before it starts filling in answers.

Building a spreadsheet model? Plan the structure. What are the inputs, what are the outputs, what assumptions are you making? Hash that out before any formulas get written.

The pattern is always the same. Invest time in the plan. Get the LLM to challenge your thinking during the plan. Then execute against a shared understanding of what “done” looks like.

When You Don’t Need This⌗

Obviously, not everything needs an hour of planning. “Add a date filter to this screen” doesn’t need a strategy session. Quick, well-defined tasks with clear inputs and outputs - just do them.

The threshold is roughly: if the task has ambiguity, it needs a plan. If you could explain the task to a junior developer in one sentence and they’d get it right on the first try, skip the ceremony. If you’d need to sit down and talk through it for a few minutes, that conversation should happen in plan mode first.

The Real Payoff⌗

A good plan doesn’t just make the output better. It makes the output unsupervised.

When an LLM has a clear plan with defined success criteria, you can hand it the work and walk away. Go get lunch. Context switch to something else. Come back to a finished result that actually matches what you wanted.

Without a plan, you’re babysitting. You’re watching every step, correcting every wrong turn, re-explaining what you meant every third message. You’re not saving time. You’re spending your time differently - and worse.

That’s the real tradeoff people miss. The hour of planning isn’t just about better output. It’s about getting your time back. You front-load the thinking, and then the execution runs without you.

The Uncomfortable Part⌗

Planning isn’t fun. Nobody opens an AI tool excited to spend an hour asking questions and defining scope. You want to build the thing. You want to see output. Planning feels like you’re not making progress.

But every minute spent in plan mode saves you multiples on the other end. The back-and-forth disappears. The “that’s not what I meant” disappears. The starting-over-from-scratch disappears. And increasingly, your presence during execution disappears too.

Slow is smooth. Smooth is fast. The hour you think you’re saving by skipping the plan is the five hours you’re about to waste sitting there, correcting an LLM that never understood what you wanted in the first place.