Build with AI
Build the machine that builds the machine.
Stop using AI to get one answer. Start using it to build tools that keep working after the chat closes.
One cutter. Press Next to start (arrow keys work too).
You've done this before
Same kind of problem. Different numbers. Back to zero.
Last week you asked AI for help with a math problem. This week you asked again. Maybe for you it's spelling words every Friday, or a new story idea every day.
Each time, you started from zero.
What if the second time could be almost instant?
34 × 27
Asking from scratch…
Times started from zero: 1
The big idea
Slower at first. Much faster forever after.
You could shape every cookie by hand. Or spend five minutes making a cookie cutter, and after that each cookie takes about one second.
Making the cutter is slower for the first cookie. By the tenth, the cutter is winning. By the hundredth, it isn't even close.
Same with a game's level editor: playing a level is fun, but the level editor lets you make a hundred.
How the chart is drawn
Cutter: five minutes to make, then about one second per cookie (from the post). By hand: we assume about one minute per cookie so the lines are easy to compare.
The ring marks cookie 6, where the cutter starts winning.
Two ways to use AI
Chat mode starts over. Machine mode picks up where it left off.
- Ask a question
- Get one answer
- Chat closes, the answer is gone
- Next time, start over
Steps done so far: 4
- Ask AI to build a tool
- Test the tool
- Save the tool
- Next time, run the tool again
Steps done so far: 4
- Prompt
- The message you type to the AI.
- Engine
- A tool that does the same kind of job every time you run it: a saved prompt, a spreadsheet, a printable, or a small program. What matters is you can run it again tomorrow.
For parents and teachers: leverage
An answer gets used once. A tool gets used again and again, and each run costs almost nothing. That's leverage: effort spent once that keeps paying back. The bigger payoff is the habit. A kid who asks "will I need this again?" starts to see their work as something to improve, not just finish.
A worked example: 34 × 27
Two prompts. Watch what each one is worth over time.
"Teach me two-digit by two-digit multiplication."
"Build a math practice engine that understands K–12 math and can make practice problems for any topic I choose. Prove it works by using it to make a worksheet that teaches two-digit by two-digit multiplication."
Look at the end of the strong prompt: "Prove it works by…" It makes the AI show you real results right away, so you can see whether the engine works. And you still get the worksheet you needed today.
The Machine Recipe
A fill-in-the-blank pattern you can reuse.
The middle blank matters most. Make it big enough that the tool is still useful next week.
Try it: which middle blank would you pick for spelling practice?
For parents and teachers: sizing the middle blank
To fill the middle blank, a child has to look at one problem and name the pattern behind it: not "these 12 words" but "any word list." That move from example to category is the heart of systems thinking. It's also where kids overreach. Help them size it: big enough to reuse next month, small enough that they could tell whether it's working.
When NOT to build a machine
Two rules. Sort these four.
Rule 1: if you'll only do it once or twice, just do it yourself. Rule 2: if the point is for your brain to learn, the machine should give you practice, not do the thinking for you.
The quick check: "After this, could I do it without the machine?"
Yes, or closer to yes than before? You built the right kind of machine. No? You built a shortcut around your own brain.
For parents and teachers: practice machines vs. thinking machines
The line isn't whether AI was involved. It's where the effortful thinking happened. One question sorts almost every case: "What did the machine do, and what did you do?" If the child's half is "I solved them," the tool is working. If it's "I copied them," it isn't.
Check your machine
AI can be wrong while sounding completely sure.
A bug is a mistake hidden inside a tool. Your math engine might print an answer key with a wrong answer. Real builders test their machines:
- Check a few answers by hand. Work three out yourself and compare.
- Try tricky inputs. Zero, a really long word, a topic it hasn't seen.
- Fix it when it breaks. Tell the AI exactly what went wrong, and ask it to fix the engine, not just that one answer.
One of these is wrong. Check by hand to find the bug.
For parents and teachers: verification is the skill
AI produces fluent, confident output whether or not it's correct, so checking work is as basic as producing it. Building a machine gives kids a natural reason to check: it's their tool. Many AI tools require users to be 13 or older, and some need a parent's consent beyond that. Check each tool's terms. With younger kids, build together or supervise.
Your challenge for today
Pick something you do every week. Build the machine instead.
Before you trust it
Test your machine
Under 13? Do this with a parent or teacher.
Next week, run your machine again. That's the moment it clicks.
Askers start over every time. Builders pick up right where they left off