Human-led, AI-powered

This is the line I will defend in any room. Human-led, AI-powered. A human sets the direction. AI does the heavy lifting in the middle. A human checks the result and owns it. Skip either human bit and you get slop at scale.

Think of any piece of work as three parts. Ten percent is direction, deciding what good looks like and why it matters. Eighty percent is execution, the grind of producing it. Ten percent is judgement, deciding whether the result is actually any good and taking responsibility for it. AI is brilliant at the eighty percent. It is fast, tireless, and cheap. It is hopeless at the two tens, because it does not know your business, your customer, or what you are trying to be.

Most failures with AI come from handing it one of the tens. You let it set the direction, so it writes copy that sounds like everyone else because it has read everyone else. Or you skip the judgement at the end, so nobody reads the output properly and the mistake goes out with your name on it. The middle was never the risk. The ends are.

I watched a firm publish a run of blog posts straight from a model with no human at either end. Confident, fluent, and saying nothing their customer did not already know. It was slop, and it was slop at scale, which is worse than no posts at all because now there is more of it to ignore. The model did not fail. The people who removed themselves from the work failed.

The opposite is just as common and just as bad. Owners who do all three parts by hand because they do not trust the tool with any of it. They are proud of being hands-on, but they are spending themselves on the eighty percent, the part a machine could do while they sleep. That is not craft. That is wasting your judgement on typing.

The sharper way to hold this is to ask, on every job, where am I in this. If you are not setting the direction, the work will drift. If you are not checking the result, the work cannot be trusted. If you are doing the bit in the middle by hand, you are the expensive option doing the cheap job.

The practical takeaway. Take one task you currently do end to end. Split it into the three parts out loud. Keep the first ten percent and the last ten percent firmly in human hands. Give the middle eighty to the tool and see what comes back. Done right, you do less work and the work gets better, because your attention lands where it actually counts.