AI made "good enough" free. Here's what it still can't hand you.

By Marc Compeau

Imagine that overnight, every kid on Charlee's street woke up to find the same machine sitting on their porch. Free. Perfect lemonade, every time — the right ratio, the right sweetness, never a sour cup. By noon there are nine stands on one street and the lemonade is identical at all of them. Now tell me which kid goes home with money in a coffee can.

That's the year we just had. The pitcher got handed out to everybody.

The tool was never the edge

There's a moment in the life of every powerful tool when it stops being a differentiator and becomes table stakes. Knowing spreadsheets was a job title in 1990. Having a website was a competitive advantage in 1998. Being the person in the office who could edit video was a whole career in 2010. Each of those was a genuine edge — right up until everyone had it, at which point the edge quietly became a baseline, and nobody got a raise for clearing it.

AI is moving through that same arc, only faster. Two years ago, "I use AI" was interesting. Today it's the equivalent of putting proficient in email on a résumé. If your entire advantage is that you know how to prompt well, you have built your business on a thing your competitor will have by Thursday.

If you can use AI, you're average. If you can solve problems and communicate — and use AI to do both faster — you're extraordinary.

I want to be careful here, because that first sentence gets misread as "AI doesn't matter." It matters enormously. It's just that it matters the way electricity matters. Nobody wins on having it. People win on what they run with it.

It can answer the question. It can't find it.

Here's the part the machine on the porch doesn't do: it doesn't know that the mailboxes at the top of the hill are where everyone stops on a Saturday, or that the soccer game lets out at two, or that Mrs. Kelleher will buy four cups if you ask about her dog first. It makes lemonade. It has no opinion about where the thirsty people are.

That's the first half of the whole thing — problem-finding. AI is extraordinary at answering a well-posed question and completely indifferent to whether it was worth asking. It will produce a beautiful, confident, well-organized solution to a problem nobody has. Which is, if we're honest, what most failed businesses are: an excellent answer to a question no customer was asking.

Knowing which problem is real comes from somewhere the model can't reach. It comes from standing at the stand. From the slow afternoon where nobody stops and you have to actually look around and figure out why. From the customer who frowned for half a second before she said "sure, I'll take one." That half-second is data no machine gave you, and it's the whole business.

And then you have to say it out loud

The second half is communication, and it just got strange. Competent prose is now free. Anyone can produce a clean, grammatical, perfectly organized paragraph about anything, instantly, forever. Which means competent prose is worth roughly what perfect lemonade is worth on a street with nine machines.

What isn't free: knowing what this particular person needs to hear. Knowing what to leave out. Having the nerve to say the true, plain, slightly uncomfortable thing instead of the smooth thing. A price is a conversation, and conversations are still won by people who can read a face.

Use the machine for the draft. Absolutely use it for the draft. Then do the part that's yours — cut it in half, put your own voice back in, and make sure it says the thing you'd actually say if the person were standing in front of you. The draft was never the hard part. Knowing what deserved to be said was.

A multiplier, not an addition

The most useful way I've found to think about it: AI is a multiplier. And the thing about multipliers is that they are ruthless about what you bring them. Ten times a real insight is a very good year. Ten times nothing is still nothing — just produced faster, in higher volume, with better formatting.

So the practical move for anyone building something is not "adopt AI." Everybody's doing that; it buys you nothing. The move is to be deliberate about which half you hand over. Give it the parts that were never your advantage — the first pass, the summary, the research sweep, the boilerplate, the seventeen versions of the subject line. Then take every hour that buys back and spend it on the two things that don't scale and never commoditize: talking to actual customers, and deciding what's worth building.

Hire the same way. I'd take the person with judgment and a clear voice who's never touched a model over the world's best prompt engineer with nothing to say. The first one learns the tool in a month. The second one has a much longer road.

Back at the stand

Nine identical stands on one street, all with the same perfect pitcher. The kid who wins isn't the one who figured out the machine — they all did that before breakfast. It's the one who noticed the shade on the far corner, who set a price that felt fair on a hot day, who looked people in the eye and said something worth stopping for.

The machine makes the lemonade. It doesn't pick the corner. It doesn't set the price. It doesn't know your neighbors' names.

That was always the job. It just used to be hidden underneath a pile of work that isn't hard anymore.

Marc



Short Takes are the weekly notes from Charlee's Lemonade. The book — The Lemonade Stand: What Every Entrepreneur Already Knew at Age Nine — is available now.