Just Build It Back to Plan It First: AI's Full Circle
When many first started using AI, planning felt obsolete. Why spend an hour outlining ideas and clarifying objectives when a minute or two could produce an acceptable report, email, or even a full project?
Just do this became the entire instruction.
Sometimes it worked. Often it didn't.
When it didn't, hours would vanish fixing problems that 10 minutes of planning would have prevented.
Worse - sometimes the problems went unnoticed, resulting in embarrassing errors (Deloitte submitted a government-commissioned report with fabricated quotes created by the undisclosed AI that actually wrote it).
There's a better way.
The Difference Between Solving and Teaching
Prompting without planning: Write me an email to decline this meeting.
Prompt. Hope. Revise when it misses the mark.
Planning with AI: Review my previous emails to this person. Consider that they're senior to me and I've already declined twice this month. Check what tone I typically use for delicate situations. Then draft three versions with different approaches and tradeoffs.
One approach produces an email.
The other produces an email and teaches the system how you communicate.
The difference isn't about better prompting. It's about deciding what you actually want before asking for it.
What Planning Looks Like Now
Planning with AI isn't the same as planning before AI. The sequence has changed.
Before: Plan extensively, then execute.
After: Plan enough to start, execute with AI, refine based on output, plan more precisely, execute again.
The best results come from alternating between planning and execution - using AI's output to sharpen your thinking, then using sharper thinking to improve AI's output.
This is faster than traditional planning. It's also faster than no planning at all, because you avoid the spiral of fixing things that shouldn't have been built that way in the first place.
Where People Get Stuck
The most common mistake isn't skipping planning entirely. It's planning the wrong things.
Time spent planning:
- What tool to use - useful
- What outcome you want - essential
- How the AI should phrase things - usually not worth it
- What constraints matter - critical
- What format you need - depends
The leverage is in clarity about outcomes and constraints. The details of execution can often be left to iteration.
A Simple Framework
Before any significant AI task, answer three questions:
- What does done look like? Be specific. Not a good report but a 2-page analysis that a board member can read in 5 minutes and make a decision.
- What constraints matter? Tone, length, format, audience, things that must be included, things that must be avoided.
- What context does the AI need? Previous work, examples of good output, background information, your preferences.
You don't need to write a detailed brief. You need clear answers to these three questions.
The Return on Planning
Ten minutes of planning typically saves an hour of iteration. Not always - sometimes you're doing something straightforward and can skip to execution. But for anything that matters, the math is clear.
More importantly, planning changes what's possible. Without planning, you get AI's default interpretation of your request. With planning, you get output shaped by your specific needs, constraints, and context.
The difference between good enough and exactly what I needed usually comes down to the thinking that happened before the first prompt.
The Irony
AI makes execution so easy that planning becomes more valuable, not less.
When creating something took significant effort, you had to plan - you couldn't afford to waste the execution time. When execution becomes nearly free, the temptation is to skip planning. But that just means you execute the wrong thing faster.
The people getting the most from AI aren't those who prompt faster. They're those who think clearly about what they want before they start prompting.
The old advice still applies: measure twice, cut once.
AI just changed what we're measuring and cutting.