One prompt now builds a whole app, a whole video, a whole game. That is exactly why the fussing I cannot stop doing is about to be worth more, not less.
Type a sentence, get a working web page. Type another, get a dashboard. Another, a small game. It works and it looks fine.
There is also nothing particularly right with it. A model trained on everything hands back the middle of everything. That is not a knock on the machine. Averaging is the job we gave it, and it does the job beautifully.
The trouble is that the middle is where everyone else's one prompt lands too.
Watch how fast each one went from clumsy to good.
Every one of those climbed steeply and then flattened into ordinary. The flattening is where a thing stops being impressive and starts being expected.
The models landing right now, OpenAI's Astra and Anthropic's Claude Fable among them, are being handed longer jobs with less instruction and coming back with more finished work. Video will get longer, quicker, and need less steering. Three-dimensional worlds are next in the same way pictures were.
That word has already spread further than most people realise. Steam, the biggest store for computer games, added a disclosure box for content made with AI. Roughly one in five new releases now carries the label, and more than 7,300 games on the store had disclosed by March of this year. Plenty of players use the label as a filter and never open the page. It is not a quality judgement, because they have not seen it yet. It is a sameness judgement.
The right-hand column is not a list of skills. It is a list of opinions, held stubbornly, over time. That is a different kind of thing, and it is the kind that does not arrive in a software release.
In an afternoon I can now find out what sells best in almost any market. Which categories lead and by how much. Which tribes exist inside them, and how the whole picture shifts depending on whether the person is twenty-two or sixty, male or female, in Ohio or in Osaka. That kind of research used to cost a firm real money and take a month. It is close to free now, and I am good at it. Thirty years of marketing will do that.
It makes no difference whatsoever to what I end up making.
Build horror. It has led its category three years running. Huge audience, spends real money, tells its friends. On the numbers it is not a close call.
An odd, gentle, faintly ridiculous browser arcade game called Alderhoard. A horror game made by me would be a horror game made by a man impersonating someone who likes horror, and you would be able to tell.
I drifted back to my own corner, the way I always do, and all that lovely research just watched me do it. The longer version of that argument is here.
Most advice assumes the opposite, that with better information you would behave differently. For plenty of people that is true. If you are wired the way I am, information is not the lever, and the comfort zone beats the spreadsheet every single time.
So the better question is not what should I be making. It is who else is standing in the corner I keep drifting into. Still a research question. Just a smaller and far more interesting one.
I spent a while watching videos of people building things in a single prompt that would take me two weeks of picking at. Bigger, flashier, done in an afternoon. And the whole time I am sitting there thinking: why am I not doing this? What is wrong with me?
I could not stop micromanaging. Nudging a colour. Rewriting a line for the ninth time. Playing the same twenty seconds of a level over and over because the timing was very slightly off.
It felt like a defect. Everyone else had found the fast lane and I was still hand-sanding.
Then I looked at what I had actually made, next to what the fast lane produces, and the thing I had been apologising for turned out to be the only reason you could tell them apart.
Past a certain point I cannot steer my attention at all. It goes where it wants and I follow it there. No productivity poster covers this bit.
Where it went recently, all of it inside Alderhoard:
Then it flips, with no warning and on nobody's schedule. A day arrives when I go through the whole game from one end to the other, asking whether a stranger could pick it up and follow it, repairing everything that only ever made sense inside my own head.
That second mode is not discipline arriving late. It is a breaking point. The pull toward actually finishing gets strong enough to overrule the fiddling, and in my experience it always turns up eventually.
Anything built on stopping the fiddling. It assumes I can hand my attention a schedule. I cannot, and forty years of evidence says so.
Keep the finish line written down somewhere I cannot avoid seeing it, so that when the focus does snap back, it has an obvious place to land.
Ten small decisions repeated across everything you make add up to a style. Nobody can prompt their way to your style, because they would need your ten decisions and they do not have them.
Every fussy choice narrows who the thing is for. That feels like shrinking your audience. It is actually finding it. A thing made for everyone is made for nobody.
When the average gets free and infinite, being distinguishable is the whole ballgame. The tweaking is not a delay before shipping. It is the part that makes it worth shipping.
It is not the ability to produce work. Production is the part that got cheap, and it is getting cheaper by the month.
The scarce part is having a view about how the work should be, and refusing to leave it alone until it is. That comes from decades of noticing when something is off, from watching formats arrive and die, from opinions you did not get out of a survey.
So the advice is not to speed up and become a one-prompt person. It is to let the machine be the one-prompt person, and to spend the hours it saves on the parts you cannot leave alone.
The pointless animals are at alderhoard.com. Four arrows and a space bar.