I think about AI as another power up. Power ups don’t pass levels. Players do.
Sometimes the level calls for the cape. Sometimes the fire flower. Sometimes the mini mushroom, because it’s the only thing that fits the pipe. Which one helps depends completely on where you’re standing and what’s in front of you.
The part I keep chewing on is what happens before any of that. In the game, the player picks. You see the block, you hit it, you take whatever comes out and you work out what it’s for. That’s fine in a game, because working it out is the game.
It goes differently in software somebody uses at work.
The person filling in an inspection form at six in the morning didn’t come to evaluate tooling. They have a walk-through to finish and a shift starting behind them. If I hand them four ways to do the same thing plus a help article, I’ve made their morning worse and called it flexibility.
So reading the level is my job. Which power up, whether one is needed at all, and whether the person in front of it has any appetite for a new thing today.
That last one gets skipped a lot. There’s an assumption baked into how we ship that people are sitting there hoping for capability. Mostly they’re hoping to finish. The window for handing somebody something new and having them take it is much narrower than the roadmap suggests, and it closes fastest for the people with the least slack in their week.
What this looks like in practice is unglamorous.
It means the assistant opens with the thing they were already doing, filled in from what the system already knows, and it asks before it acts. It means one obvious path rather than a menu, and the menu tucked behind the path for the small number of people who want it. It means the feature works on the first screen, before anyone configures anything, because if the first thing somebody meets is a setup step then whatever I built is already losing to the way they did it last week.
And it means being honest that some levels don’t need a power up. Sometimes the answer is that a field should be pre-filled, or removed, or that the form was three questions too long. I’ve watched a lot of effort go into making something intelligent that would have been better made shorter.
None of that means the capability isn’t worth building. I want the capability. I’d just rather the person using it never had to become a student of it.
When I’m unsure whether something is worth shipping, the check I keep going back to is whether it moves any of four things: agency, so somebody can choose or shape or own the result; clarity, so they can tell what happened and what matters next; capability, meaning they get better at the work rather than routed around it; or belonging, where the system recognizes both the context and the contribution of a particular person.
Progress in any of those is usually a solid sign of being on the right track. Speed on its own has fooled me before.
Because the level still has to get passed by a person. Everything else is inventory.