My good friend and AI adoption co-writer, Gerard Pietrykiewicz, asked me a question recently. Are you a sheep or a shepherd? Do you own the loops, or are you a piece of one?
My answer was neither.
Sheep and shepherd both assume you’re inside someone else’s structure. One takes direction, the other gives it.
Independent work doesn’t have that structure.
Nobody assigns my tasks. Nobody checks my drafts before they go out. Nobody tells me when the work is off, because nobody’s reading it before the client does. Run through what that actually removes: no performance review, no peer looking over your shoulder, no boss’s gut check before something ships. When you work for yourself, all of that lives in one person. You.
That’s not a complaint. I chose this. But it’s worth calling out because it changes what “adopting AI well” actually requires.
There are two things that can fill it. Your own judgment. Or whatever tool is fastest and easiest at 4pm on a deadline day.
AI is very good at quietly becoming the second one. Especially if you’re someone who’s used to trusting your own read on things. It’s easy to mistake “the AI didn’t flag a problem” for “there is no problem,” when what you actually needed was a second set of eyes (that you don’t have). It’s the same effectiveness-before-efficiency trap that shows up when teams optimize for output over judgment.
Teams have redundancy built in. A colleague catches what you miss, even by accident, even if nobody’s officially checking.
Independents don’t have that. If your own judgment slips, there’s no safety net until the client notices — and by then it’s not a quality problem anymore. It’s a relationship problem.
You’ve probably seen “human in the lead” this year. It’s been framed, repeatedly, as the shift enterprises need to make — humans directing AI instead of just supervising it, so frontline workers stop feeling like a checkpoint on their way out. Fair point, for an enterprise.
But that framing assumes there’s an enterprise to redefine. A team, a hierarchy to reassure. Strip that away and the phrase means something else entirely. There’s no one to lead but yourself. No org chart to redesign around the new relationship with the tool. It’s not human in the loop. It’s human in the lead, except there’s no one else in the loop. Just you, and whatever discipline you bring to checking your own work.
Here’s where it stops being just an independent’s problem.
A go-to-market team (Sales, Marketing, Product, CS) looks like it has the redundancy independents don’t. Someone signs off on the forecast. Someone reviews the campaign brief. Someone technically outranks the person who drafted it. That structure implies a check exists. That’s the same illusion of control that shows up whenever process gets mistaken for oversight.
Often it’s signing off on the output, not the judgment behind it. A VP approving a forecast built with AI assistance is rarely re-deriving the assumptions underneath it. They’re trusting that whoever built it did the thinking. If that person leaned on AI to fill a gap in their own judgment, the sign-off doesn’t catch it. It just adds a signature to it.
That’s not a hunch: a 2026 Grant Thornton survey of 950 executives found 78% lack strong confidence they could pass an independent AI governance audit within 90 days. The structure exists. The proof underneath it usually doesn’t.
Ask the same question, just inside a team: what's genuinely being checked here, and by whom, with the expertise to actually catch a bad call? “Someone outranks me” and “someone is checking this” are different claims. Most org charts only guarantee the first one.
It requires a few specific things, not more willpower.
Lovable used to be the same. Before its MCP server launch, it was cut-and-paste like everything else. Now Claude connects to Lovable directly and can build, iterate on, and deploy my prototypes without me relaying anything by hand. That upgrade also changed what kind of dependency it is. Without it, the prototype doesn’t exist. That’s binary.
Most people never sort their own work this way, and fewer still notice when a tool crosses from one category to the other. Workflow dependency means staying sharp enough to catch a bad output. Binary dependency means knowing exactly how exposed you are if the tool disappears — and that exposure can arrive quietly, the moment a tool gets more capable, not less.
Being your own shepherd is a specific job, independent or not. You have to build the checks a boss, or an org chart, would normally imply deliberately, or they don’t exist.
Start with the sort: go through what you actually use AI for and split it in two: what disappears completely without the tool, and what just gets slower. That should take less than ten minutes. Most people have never done it.
Then ask: who’s actually checking the slower stuff before it goes out? If the honest answer is nobody — whether you’re independent or three levels into an org chart — that’s not an AI problem. That’s the job nobody built.
Human in the lead. No one else in the loop. Nobody’s assigning you tasks, or nobody’s really checking them. Same job either way.
Co-authored by Gerard Pietrykiewicz and Achim Klor. Follow us on LinkedIn, schedule a call with Achim, or contact Gerard if you need help with AI adoption. Subscribe below for more.
This article is AC-A and published on LinkedIn. Join the conversation!
