Software Gets Installed. Intelligence Gets Onboarded.
We bought artificial intelligence and have been using it like artificial efficiency. The real work isn't prompt engineering — it's onboarding something smart and unfinished, the way good managers always have.
Photo Credit: Rob Grzywinski
Most people are adopting AI the way they'd adopt a new app — install it, learn the buttons, expect things to go faster. That's why so much of it feels underwhelming. AI isn't artificial efficiency; it's artificial intelligence, and intelligence has to be onboarded. The people and companies getting real value out of it figured that out somewhere along the way. The ones who didn't are stuck asking it to rewrite emails and wondering what the fuss is about.The fuss is real but the category is wrong. Software does what you tell it to do. Intelligence does something stranger. It misunderstands you, overreaches, gives you something almost-right in a way that forces you to know what right would have been. That isn't a bug to be patched in the next release. That's what intelligence is and working with it is a different skill than clicking buttons.
The Intern Test
Imagine hiring a smart intern. On their first day, you hand them a messy email and say "make this better." They come back with something too formal, too long and slightly wrong. You read it and decide interns are overhyped.Everyone can see the problem. Maybe the intern did mediocre work — fine. But the deeper failure was managerial. You gave no context, no examples, no sense of audience or stakes. You didn't say what better meant. You didn't read the draft and say "good instinct here, wrong move there." You didn't onboard them.AI is the first thing most of us have ever bought that gets better when you manage it.
That is how most people are using AI, and nobody should feel bad about it. Why would they? Every tool we've ever owned just did what it was told. Word didn't get better when you mentored it. Excel didn't surprise you. AI is the first thing most of us have ever bought that gets better when you manage it, and managing isn't a skill software ever required.
The Organizational Version
This gets bigger than prompts in a hurry, because companies also have a relationship with intelligence. You can see it in how they treat interns. In what happens to the consultant's report after the consultants leave. In whether bad news moves toward a decision or toward a scapegoat. AI didn't create those patterns. It just made them cheap to observe.Some companies know how to work with intelligence before it's convenient. They know it arrives half-formed, creates work before it creates value, and needs translation and protection. Others only want intelligence after it's been cleaned, approved and made non-threatening — but by then it isn't intelligence anymore, it's compliance.That's why so many AI rollouts feel strange. The company thinks it's testing a tool; it's actually testing whether it knows what to do when something smart shows up unfinished. Companies have been failing that test with humans for decades. AI just brought the bill forward.
What to Ask Instead
A piece of advice has been circulating: when you take a job offer, ask what your token budget will be. It's not a bad question — access matters and a company that gives nobody room to breathe is telling you something. But it's the wrong first question. It measures how much intelligence the company bought, not whether anyone there knows what to do with it.The better questions are quieter. Who reads the output? Who corrects it? Who owns the work when it ships? When the AI produces confident garbage, does catching it count as productivity, or does the company only reward speed? And the one that matters most — what happened the last time someone, human or machine, told the place something true and unwelcome?A mature organization knows how to receive unfinished intelligence without immediately defending against it. That's the bar. Most companies aren't there yet and most of us aren't either. That's fine — we can learn. We just have to stop pretending this is another software rollout.We built artificial intelligence and have been adopting it like artificial efficiency. The actual work is older and slower than prompt engineering: it's the work of getting something smart to do something useful before it's quite ready — which is what good managers and teachers and editors have always done. We just never had a tool that asked it of us. Now we do.
Your AI coding spend isn't underdelivering because the tools are bad. The day you handed your developers an agent, you promoted them from writing code to managing it — and forgot to tell them. It's a different job and most of them are stalling in the same four places.17 June 2026
Every institution we've built — academia, funding, startups, even our AI tools — actively punishes anyone who tries to work between fields. The most valuable work of the next decade will happen in exactly those seams. Nobody has a home there.20 May 2026