For operators who burn their AI limit before lunch and still get back work they have to redo.
Claude Plan Mode:
The First Hour That Pays For The Whole Day
Two things go wrong when you use AI for real work. You run out of limit before lunch, and what comes back still is not right. Most people hunt for a bigger plan to fix the first and a better prompt to fix the second. They are the same problem, and it has one fix.
4 projects
Instead of one, head down
15 min
Running while you do other work
1 hour
Once, before the day starts
What you walk away with
A prompt you can dictate today that stops the model guessing at what you meant.
A review pass that catches the wrong direction while it still costs you one line.
A one-hour morning routine you can put in a calendar and stop deciding about.
The split that decides whether the work runs once or three times.
I have run this every working day since early 2026, across client work, team sessions and cohort workshops. It is the single biggest change I have made to how I work.
Foundation 01
You Are Not Hitting A Limit. You Are Paying Twice.
The same mistake costs ten seconds on a page and a whole job on a deliverable. Everything here is about moving it earlier.
You asked it to guess. Then you let it do the whole job on that guess. It picked a direction, you found out after it finished, so you ran it again. That second run is where your limit went, and you are still not happy with the result. Burning through your allowance and getting slop back are not two problems. They are one problem, caught too late.
What being wrong costs at each stage
Wrong while the plan is on screen
10sOne line. Ten seconds.
Wrong just after you hit go
5 minStop it, correct it, restart. A few minutes.
Wrong when the run finishes
The lotThe whole job again. Your time, your limit, your patience.
The useful reframe. The question was never whether to plan or just go. It was where you would rather be wrong. Because you will be wrong sometimes. Everyone is. Plan first and you are wrong on a page. Skip it and you are wrong on a deliverable.
What changes when you fix it. Moving the mistake earlier is the whole of it. Since I started planning first, I get around five times the work through. That is not a new model or a new tool. It is the same work, ordered so the expensive part only runs once.
Foundation 02
A Plan Is The Last Cheap Place To Change Your Mind
Plan mode reads and explores but does not touch your source. Edits stay blocked until you approve. Three ways to switch it on.
Here is the reframe everything else sits on. A plan is not a list of steps. It is the last cheap place to change your mind.
Plan mode is the setting that gives you that place. Claude reads your files and runs commands to explore, then writes a plan. It does not edit your source, and edits stay blocked until you approve. That is enforced by the tool, not by good manners, so it cannot get carried away and start building while you were still thinking.
Three ways in
Type the command
Prefix a single prompt with /plan. Simplest, and it is what I use.
Cycle the modes
Shift and Tab steps through Manual, then accept edits, then plan. Two presses from a fresh session. The bar shows plan mode on.
Use the picker
Click the mode indicator and choose Plan. Available in the desktop app, VS Code, JetBrains and claude.ai.
# prefix one prompt /plan rewrite the onboarding sequence # or start the whole session in it claude --permission-mode plan # the status bar confirms it ⏸ plan mode on
On Windows. Shift and Tab does not always register, because some Node and Bun runtimes do not enable VT input mode. Use Alt and M instead. It cycles the same modes.
That is the whole technical bit. Everything after this is about how you talk to it.
Foundation 03
The Prompt That Stops It Guessing
Dictate for five to ten minutes, hand over what only you know, then finish with the one line that makes it interrogate you instead of assuming.
Most people flip to plan mode and then write the same brief they always write. That gets you a fancier version of the same guess. Here is what I actually do.
Talk. Do not type.
I dictate. Five to ten minutes, out loud, on what we are doing and why. Typing makes you terse. You write four lines and think you have been clear. Talking makes you ramble, and rambling is exactly what you want here. You mention the client. You mention the thing that went wrong last time. You mention the constraint you would never have bothered typing. All of that is context, and context is the whole game.
Tell it what only you know
The goal, in plain words.
Who it is for, and what they actually care about.
What you have already tried, and why it did not work.
Any files, docs or notes it should read first.
What good looks like to you.
You can paste a meeting transcript straight in and say there are some good ideas in here. It will use them. The model knows a little about a lot and nothing about your business. You are the expert. Plan mode is where you hand over the part only you have.
Then the line that changes everything
Use the ask_user_question tool. Drill down and ask me everything you need. I do not want to leave any stone unturned.
One word, ask_user_question, no spaces. Left alone, the model fills the gaps in your brief by guessing. This makes it interrogate you instead. It comes back with multiple-choice questions, usually double or triple what you expected, and it asks about things you had not thought of. Answer them properly. This is the part of the job that stays yours, and every gap it finds here is a gap you do not pay for later.
Enjoying the guide?
Enter your details to unlock the remaining insights. We'll also send you the complete guide as a reference.
The Method
Read The Plan Like A Junior Wrote It
This is where most people fall over. They get a plan, glance at it, and say go. That is the same as not planning.
This is where most people fall over. They get a plan, glance at it, and say yep, go. That is the same as not planning. You know your business better than it does. You know what good looks like. So read the plan and mark it up like a junior wrote it.
You have missed this
This bit is great, keep it
Drop that entirely
Wrong order
Get the plan right first. Then hit go.
If it comes back fast, it is thin
A real plan takes time. Fifteen minutes is normal. Twenty is common. Sometimes longer. We got used to instant answers, and genuinely useful AI work is not instant. So if a plan lands in thirty seconds, do not accept it. Send it back.
Have you been thorough? Is this as detailed as you can go?
You will be surprised how often the second version is twice as good.
Then take your hands off, but not off the wheel. Once you like the plan you are not stepping away from the car. You are putting it into cruise control. You still check in. You still steer. You just stop doing every single input yourself.
The Method
When It Earns Its Time
Not every job needs a plan. Four always do, and the fourth is the one everybody misses.
Not every job needs a plan. The short answer is anything significant. These four always get one.
Starting something from scratch
A new build, a new document set, a new system. Nothing exists yet, so the shape is entirely up for grabs.
Designing something
A dashboard, a structure, an architecture. Anything where the shape matters more than the words.
Adding a big feature
Bolting something substantial onto a thing that already works. The risk is what it breaks, not what it adds.
Something is wrong across the whole thing
This one gets missed. When I find a mistake repeated everywhere, I do not fix it file by file. I go into plan mode and say: you have not done this properly across all of them, here is how it should be, give me a plan to revise it everywhere.
Plan the part, not the whole project. You do not have to plan an entire project in one go, and on anything complex you should not. Plan the architecture. Then plan the database. Then plan the mapping. Do it in the order you would do it if you were doing it yourself. Pick up a new part of the project, run a new plan.
Start fresh chats. The longer a conversation runs, the worse the output gets. Long context means more mistakes and more made-up answers. Start a new chat for a new piece of work. Do not drag a four-hour conversation into a fresh job.
The Routine
The First Hour
Plans are slow, so you cannot fire one at the moment you need the work. Fire all of them before the day starts instead.
Plan mode is slow. That is the feature, not the bug. But it means you cannot fire a plan at the exact moment you need the work done. So do not. Fire them all first thing.
Where the tasks come from
I do not pick tasks off vibes. Everything traces back up that chain. Every Sunday, sometimes Monday morning, I run an accountability session. It goes through my goals, my data and my calendar, tells me what I did and did not do, and asks me questions I would rather avoid. Then I set the week. Every morning I check the agenda. Some days it tells me I have forty hours of tasks in one day, which is not real, and asks me what is actually highest priority.
Then one question. If I need to hit this goal this week, what are the most important tasks I need to do today?
That is your MITs. Most Important Tasks. Three or four, not ten. Ten is not a priority list, it is a wish.
Make it a habit, not a decision
Mine is a recurring block in my calendar. It fires every morning when I wake up and it is called MIT Planner. If it is a decision you make each day, you will skip it on the days you are busy. Those are the days you need it most.
The hour itself
Check the week's goals.
Pick today's three or four MITs.
Open the project for the first one. Switch to plan mode.
Dictate five to ten minutes. Finish with the ask_user_question line.
Answer its questions.
Move to the next project and do the same. Do not wait.
By the time the last one is running, come back to the first.
Read it, mark it up, hit go.
Go and do everything else.
Four projects, about fifteen minutes each. One hour. All I am doing in that hour is getting plans. Morning works best because nothing has gone wrong yet and nobody is bothering me. By midday there are meetings and things on fire. If you are a night owl, do it the evening before. What matters is a block where nobody can reach you.
The Routine
Half-Hour Slots, Not Deep Work
The part nobody warns you about. It feels wrong for a week, then you are moving four projects a day instead of one.
You do not watch the potatoes boil
You are cooking a few dishes at once. You do not stand at the hob watching one pot. You put the potatoes on, then you chop something, then you get something in the oven. Everything cooks at once and it all lands together. Same thing here. Fire all four plans, then go do your other work while they cook. Sitting and waiting for a plan is throwing away the exact productivity you just bought.
Half-hour slots, not two-hour blocks
This is the part nobody warns you about, and it changes the shape of your whole day.
Two hours, head down, one thing. Deep work. You finish one project and start the next tomorrow.
Half an hour on this one. Refine it, push it forward, set the next thing going. Move to the next project. Come back round later.
It feels wrong for about a week. Then you look up and you are moving four projects a day instead of one.
Why it actually matters
Some days I do not get to my last MIT until quarter to four. School pickup is half four. Forty five minutes. If I had to start planning at 3.45 it is not happening: fifteen to twenty minutes of planning and I am out of time before I have done any real work. But the plan has been sitting there since eight in the morning. So I open it, iterate, and get the thing out the door before I leave.
That is the difference. Not that AI made me faster. The thinking was already paid for, hours ago.
Why It Works
The 10 / 80 / 10 Split
Get the first tenth right and the middle runs once. Skip it and the middle runs three times.
Once this is running, your work splits three ways.
Yours. The planning.
This is what makes the output yours and not generic. You bring the client knowledge, the context, the standard, the direction. Nobody else can do this bit.
The AI's. The doing.
Drafting, building, restructuring, first passes. Let it run.
Yours again. The review.
Critical, not a rubber stamp. Is it in your voice? Is it over-delivering or just delivering? Does it line up with everything else you have done? There is usually a round or two of do this better here.
Get the first 10% right and the 80% only runs once.
Skip the first 10% and the 80% runs three times.
That is the whole economics of it, and it is why both problems at the top of this guide have the same fix.
Why It Works
Pick The Model, Own The Handoffs
Thinking work gets a thinking model. Volume work does not. And when the AI hands off to itself, it cuts corners.
Planning is thinking work, so give it a thinking model. Drafting and volume work can run on something faster. The lazy version is to run everything on the heaviest model, and that is the most common way people blow their limit. It is not thoroughness, it is waste.
You do not have to manage it by hand any more. Set the model to opusplan and it splits the job for you. Opus does the planning, Sonnet does the execution, and it switches over automatically at the moment you approve the plan.
/model opusplan # plan mode -> opus (reasoning, architecture) # after approval -> sonnet (code, drafting, volume)
Check what you are on. The default model depends on your account. Max, Team Premium, Enterprise pay-as-you-go and the API all default to Opus 5. Pro and Team Standard default to Sonnet 5. So if you are on Max and have never touched your model setting, you are already running the heavy model for everything, including the volume work that does not need it. Max is 100 US dollars a month for the 5x tier and 200 for the 20x, so that setting is worth two minutes of your attention.
The bit almost nobody does
When the job is big, ask the plan to cover the handoffs too. Everyone wants the AI to spawn its own agents and run the whole thing alone. I have tested it plenty, and here is what I keep finding: the output is better when you do the handoffs yourself. When the AI hands off to itself, it cuts corners. So I ask for the handoffs in the plan.
Which step uses which model.
What can run in parallel.
What depends on what finishing first.
The AI does the work. The handoffs are mine. You stay the driver. That is the job.
Field Notes
Seven Ways People Get This Wrong
Every one of these produces a session that looks like planning and costs the same as not planning.
Every one of these produces a session that looks like planning and costs you the same as not planning.
Treating the plan as a formality
Getting a plan and immediately saying go is the same as not planning. The review is the point.
Accepting a fast plan
Thirty seconds means it guessed. Ask whether it has been thorough and watch the second version.
Typing instead of talking
Typing makes you terse. Five minutes of rambling out loud beats four typed lines every time.
Skipping the ask_user_question line
This is what turns a guess into a plan. Do not drop it, and answer properly when it asks.
Ten MITs
Three or four. If everything is important, nothing gets planned properly.
Planning at the moment of need
Plans are slow. Fire them first thing so they are waiting when you get there.
Sitting and watching it think
That is the productivity you just bought, thrown in the bin. Start it and go do something else.
BurningBurningyouryourlimitlimitfeelsfeelslikelikeaapricingpricingproblem.problem.GettingGettingslopslopbackbackfeelsfeelslikelikeaapromptpromptproblem.problem.NeitherNeitheroneoneisisit.it.ThatThatisisaadirectiondirectionproblem,problem,andanddirectiondirectionisischeapcheaptotofixfixwhilewhilethetheplanplanisisstillstillononthethescreen.screen.
Your first week
Do not rebuild your whole day. Do this.
Write down your three MITs for tomorrow. Just three.
Open one project. Switch to plan mode. Dictate five minutes on the goal and the context. Finish with the ask_user_question line. Answer the questions. Read the plan. Mark it up.
Leave it and go do something else.
Come back. Notice the thinking is already done.
Do it for all three MITs.
Put it in your calendar as a recurring block so it stops being a decision.
Compare what you got through against a normal week.
One habit. One hour. At the front of the day.
Jump back to a section
Read next
Primary sources