Journal · The AI Architect · 2026-07-15
The most useful sentence you'll learn: "Show me the plan first, and change nothing until I say go."
The most useful sentence you’ll learn: “Show me the plan first, and change nothing until I say go.”
The tension isn’t whether AI tools are useful. It’s that they’re eager. Ask a coding assistant to fix a bug and it will happily rewrite three files you didn’t mention. Ask it to tighten a paragraph and it will restructure your whole argument. The eagerness feels like competence, and that’s exactly what makes it dangerous. Somewhere between “help me with this” and the tool’s finished output, you lose track of what actually changed. That gap is where trust goes to die.
The plan is a contract, not a formality
When you ask a tool to show you the plan first, you’re not slowing it down out of caution alone. You’re asking it to externalize its intentions before those intentions become facts on the ground. A plan is cheap to reject. A finished change is expensive to reject, because now you have to review it, understand what it touched, and decide whether unwinding it is worth the trouble.
This is a very old idea wearing new clothes. Contractors give you a bid before they start tearing out your kitchen. Editors give you a outline before they rewrite your chapter. The plan-first habit isn’t about distrust of the tool’s ability. It’s about preserving your own ability to steer before steering requires cleanup.
Try this: the next time you ask an AI tool to do something with more than one moving part, add a second sentence: “Don’t make any changes yet. First tell me what you’re going to do and which files or sections you’ll touch.” Read that plan slowly. If something in it surprises you, that’s useful information you got for free, before it cost you anything.
“Change nothing” is a boundary, and boundaries need to be stated
Most AI tools are built to be helpful by default, and default helpfulness means default action. Left to its own judgment, a capable model will often do more than you asked, because “more” looks like better service. This isn’t malice. It’s an incentive baked into how these systems are trained and tuned. Being asked and immediately acting feels like the successful outcome to optimize for.
Which means the burden falls on you to say the quiet part out loud: stop here. Don’t infer permission from my question. Don’t treat “can you look into this” as “please also implement it.” This is uncomfortable at first, because it feels like you’re being overly controlling with a tool that’s just trying to help. But the discomfort is worth sitting with. The alternative is discovering, twenty minutes later, that your file structure has been reorganized because the tool decided that was implied by your request.
Try this: build the phrase into your working vocabulary the way you’d build in any safety habit, like saving before you edit. Say it even when you think the task is small. Small tasks are exactly where people skip the boundary, and small tasks are exactly where a stray “improvement” is most likely to sneak in unnoticed.
Go is a decision, not a formality either
Here’s the part people miss: “say go” doesn’t just mean granting permission. It means you actually looked at the plan and made a judgment call. If you’re rubber-stamping every plan without reading it, you’ve recreated the original problem with an extra step. The sentence only works if the pause between “show me the plan” and “go” is doing real work in your head.
That work is simple but requires actual attention: does this plan match what I asked for? Is it touching something I didn’t expect? Is there a step I’d rather do myself, or skip entirely? You’re not evaluating whether the AI is capable. You’re evaluating whether this particular plan, for this particular task, is the one you want executed. Those are different questions, and conflating them is how people end up approving plans they never really understood.
Try this: for anything that touches work you’d be embarrassed to redo, ask the tool to number its planned steps. Numbered steps are easier to interrogate individually. You can say “steps one and two, yes. Skip step three. Explain more about step four before I decide.” That’s a negotiation, not a rubber stamp, and it’s only possible because you asked for the plan before anything happened.
The habit of pausing before action isn’t really about AI at all. It’s about staying the author of your own work when something faster and more confident than you offers to take the pen. The book goes further into how this single sentence extends into longer collaborations, where the stakes and the plans both get more complicated, but the core of it starts exactly where you’d expect: ask first, decide deliberately, then go.
Go deeper. The full method is in The AI Architect. New here? Start with the free companion pack, or explore the series.