A little over a year ago I wrote something on Opinion Paper entitled "Where AI Goes." I gave it great caution. In essence, what I said then was this:

If a machine does what humans cannot do, it expands human capability. If a machine does what humans can do, it replaces human capability.

It's been a year. You can barely use the Internet now without using AI. It's in every search engine. It's in your phone. Increasingly, it's in every application.

What is AI? Because "artificial intelligence" is a terrible definition. AI is a prediction engine, running at a scale no human can match. In the same way a spreadsheet calculates pages of numbers no human could work through by hand, AI reads and digests information and can actually discuss it with us. Everything else it appears to do - synthesize, simulate, advise - is prediction at that scale.

It all depends on what you give it.

Humans aren't always logical and AI doesn't always understand what we ask, so sometimes the conversation is weird. But AI has moved into the spaces defined by rigid rules and logic and started to dominate. Programming most visibly of all.

No programmer writes out zeros and ones. We use tools that create those for us when we type out code. Now we use a newer tool that creates the code when we express what we want. We call that a prompt.

If you think about it, every CEO and every manager has been prompting for years. "Susan, I need that report for Tuesday's meeting." How is that different from "Grok, create that report I need"?

By my thesis from last year, Susan isn't required any longer.

But that isn't the only possible outcome - Susan can now prompt too. She can marshal agents and produce what previously took a department. She doesn't have to be replaced - she can promote herself into the CEO's chair. That's the part I didn't see. Replacement isn't the endpoint. It's a rung for those who choose it.

The catch is what the chair requires. The new skill is management, and good management is judgment - knowing what's worth building, and telling good work from work that merely looks good.

Most people don't have experience being a manager or CEO. Judgment isn't obvious. It's built from success and failure you could trace back to your own decision. Susan can review a report generated by AI and determine its quality because she spent years writing bad ones and finding out why they were bad.

The machine can write bad output. Without judgment, nobody discovers it.

That's why social media is awash in confident nonsense, and why it keeps getting harder to spot. AI can polish a turd to a high shine - not just pictures and video, but arguments. Feed it idiocy and it will hand you back a well-organized case for the idiocy. It's a prediction engine. It has no stake in whether the prediction is any good.

Back to programming: prompting a website requires the ability to state clearly the requirement, the aesthetic, the format, and the quality. It requires testing, which requires looking for the unexpected, which only comes from experience.

Consider the person without construction experience who wants to prompt a big, open concept home with robots one day. For that big, open space, is there enough load-bearing bracing to support such a structure in a bad storm? Quality construction can only come with quality experience and judgment.

In a post-job economy, we enter the management economy. The core requirement will be a humble and careful wisdom.

Or we will build structures of confident nonsense.