Framework
Your kids will pay to think
On steering, taste and the thinking tax.
Three things are true and most people have not connected them yet.
- Your kids will pay for AI-assisted thinking the way you pay for internet access - not as a luxury, but as a basic cost of functioning.
- Most of them will waste that money. The default answer is the average one. The valuable answer takes steering: taste, depth and the nerve to ask "where am I wrong?" Nobody is teaching this.
- The gap between those who can steer and those who cannot will be the next literacy divide. It is already opening.
01The first answer is always free
You need to price a new product: a compliance automation tool for Brazilian trade operations. Niche market, complex regulation, buyers who have seen every vendor pitch twice.
You ask the AI. You get a perfectly adequate answer about value-based pricing, competitor benchmarking and three-tier plans - the kind of answer on the first page of any SaaS pricing blog. Competent. Generic. The answer it would give anyone who asked the same question.
That is the default: the answer that would satisfy the largest number of people. The consensus. The peak of the distribution. Useful the way an encyclopaedia is useful, for getting oriented. Nobody builds a business on encyclopaedia summaries.
The default output clusters at the peak. The value lives in the tail - and you have to steer there.
So you push back. What happens next is the entire point of this piece.
02What steering actually looks like
Back to the pricing problem. Here is the conversation, compressed. Watch what happens at each turn.
1The generic question
You ask
How should we price a compliance automation product for trade operations?
It answers
Consider value-based pricing. Research competitors like Descartes and MIC. Offer three tiers based on transaction volume: Starter at $X, Professional at $Y, Enterprise at custom pricing.
Median response. Could apply to any product. Zero insight.
2First steer: add context
You ask
These companies currently pay customs brokers $800-2000 per shipment to handle compliance by hand. A single mistake causes a $15,000 fine and a three-week delay. Our tool removes 90% of the manual review. How does that change the pricing model?
It answers
With $800-2000 per shipment at risk, pricing should anchor to cost avoidance. At 200 shipments a year and $1,200 average broker cost, the tool saves $216,000 annually; capturing 15-20% of that puts you at $32,000-43,000 a year. But the real anchor is the $15,000 fine - one prevented fine a year justifies the whole subscription.
Better. Specific numbers. Still conventional pricing theory.
3Second steer: cross-domain
You ask
Interesting, but churn worries me. These companies have used brokers for decades - it is a relationship, not just a service. What would a behavioral economist say about the switching cost? And what does insurance pricing teach us about pricing risk-prevention products?
It answers
Two frames. From behavioral economics: you are fighting loss aversion in reverse. The broker is a known cost - predictable, comfortable. Your tool is an unknown saving - uncertain, uncomfortable. People weigh potential losses about 2.3 times more than equivalent gains, so reframe from "save money" to "remove your largest uncontrolled risk". From insurance: risk-prevention products have an asymmetric value curve. The customer does not pay for the 999 shipments that go fine, they pay for the one that does not. Price like an insurer: a base premium for coverage, with the automation as the mechanism.
Now we are in the tail. Behavioural economics plus insurance pricing. No generic post covers that combination.
4Third steer: pressure test
You ask
You are being too safe. What is the strongest argument against everything you just said? Where does this whole approach fall apart?
It answers
The strongest counterargument: your customers are Brazilian importers, and there the broker relationship is not a service contract - it is a risk-sharing partnership with implicit guarantees. The broker absorbs regulatory ambiguity. Software cannot absorb ambiguity; it only handles the cases it was built for. The moment a new regulation lands that your system has not learned, the customer is exposed with no human backup. You are not replacing a cost - you are asking them to replace a safety net with a tool. The insurance frame only holds if you also provide the equivalent of reinsurance.
Deep tail. This identified a structural flaw in the positioning, not just the pricing - and it changed the whole go-to-market.
Four turns. The first gave you what a search engine gives anyone. The last found a structural risk in the positioning that would have taken months to discover by trial and error.
The AI did not get smarter between turn one and turn four. You steered it. Each turn needed something no prompt guide teaches: the taste to recognize that the first answer was generic, the depth to know that behavioral economics and insurance pricing were relevant, and the courage to ask "where does this fall apart?" when you might not like the answer.
Here is the paradox: the people who extract the most from AI are the people who already know the most. You cannot steer toward knowledge you do not know exists.
The gap between turn one and turn four is not technology. It is the operator.
03The skill your kids need
Your daughter is 14 with a school project on climate change. She opens an assistant, types "tell me about climate change" and gets a perfectly organized five-paragraph summary: causes, effects, solutions. She copies it, changes a few words and submits it.
She got an A. She learned nothing. She used AI the way most adults do - as a vending machine.
Now imagine she reads the summary and thinks: everyone in my class will write exactly this. So she starts pushing.
1Reject the obvious
She asks
Everyone already knows about CO2 and rising temperatures. What about climate change would actually surprise people?
It answers
Ocean acidification: CO2 dissolves into seawater and breaks down the shells of the tiny organisms at the base of the food chain. The part of climate change that has nothing to do with heat and everything to do with chemistry.
Not what she expected - which is the point.
2Bring in another angle
She asks
My teacher keeps saying nuclear energy is too dangerous. Is that actually true, compared with coal?
It answers
Deaths per terawatt-hour put nuclear as the safest energy source per unit of power generated. Coal kills hundreds of times more people.
The data contradicts what she was taught. She had no idea.
3Challenge the answer
She asks
Are you just being pro-nuclear because that is what the internet says? What is the actual strongest argument against it?
It answers
Waste storage is genuinely unsolved over thousand-year timescales. Plants take fifteen years to build and cost billions. And they need the kind of stable, long-term government commitment democracies are bad at holding. The real argument against nuclear is not safety - it is politics.
It walked back its own framing. That only happens when you ask.
Same three moves. She recognized the generic answer and rejected it, pulled in a different angle, then challenged the position. A 14-year-old running the same playbook as the pricing example - not because someone taught her a framework, but because she refused to hand in something boring.
Nobody taught her that at school. School is still optimizing for the skill AI already does perfectly: recalling and organizing existing knowledge. The bottleneck moved and education has not caught up.
Teaching a child to say "that is too obvious, go deeper" is worth more than teaching them to code. Code is the commodity. Judgement is the differentiator.
And nobody measures this. We test reading comprehension, mathematical reasoning, spatial awareness. We do not test whether a student can recognize a median answer, push past it and pressure-test what comes next. That ability will shape their career more than any grade they get today.
04The divide is already opening
Every literacy transition follows the same arc: novelty, advantage, expectation, invisible. Nobody lists "can use email" on a CV anymore. Nobody will list "uses AI" in ten years.
But "uses AI" was never the skill. The skill is steering. And unlike previous divides, this one compounds: every year you develop it you extract more from better models, and every year you do not, the gap widens.
The same four stages, every time
- Novelty - "interesting, but I do not really need it."
- Advantage - "the people who use it seem to move faster somehow."
- Expectation - "you do not know how to use this? How do you get anything done?"
- Invisible - "nobody mentions it because everyone does it. Like breathing."
A pricing decision. A school project. A medical diagnosis. A product strategy. Different contexts, same underlying skill. The person who stops at the first answer and the person who pushes to the fourth are holding the same tool. They are not getting the same results.
Your kids will pay to think. The question is whether they pay for average answers they could have found themselves, or for the lateral, synthesised, pressure-tested kind that makes them genuinely dangerous in their field.
That depends entirely on what you teach them now.
Steering is how we build
Every product decision here - architecture, pricing, go-to-market - goes through the process you just read. That is how a very small team competes with companies ten times its size.
Essay
AI isn't replacing you - it's revealing your standard
When someone says "AI will replace developers", that is not a prediction. It is a confession about their quality bar.
Essay
Why companies are wrong about AI layoffs
AI does not replace people. It replaces the average. The companies cutting headcount are optimizing for the wrong thing.