A few weeks ago, mid-call, someone asked what I actually do. What came out wasn’t my usual answer. Make the people already in the business better. Capable of more, instead of finding a way around them.
There was a pause on the line. “That sounds like change consulting,” he said. “Not coaching.”
Fair catch. I’d drifted off the script I usually run and said the thing I actually believe.
Most companies don’t ask that question. They ask a narrower one: what can this task do without a person in it? And when they find the answer, they call it an AI strategy.
I don’t think that’s the goal. I think it’s the easy version of the goal.
The fork every AI rollout hits
Every AI rollout lands on the same fork, whether anyone names it out loud or not.
One path is automation. Find the task, replace the person doing it, count the headcount you didn’t have to add. It’s the easier story to tell inside a company, because it’s a number. Fewer heads, same output, the math works on a slide.
The other path is augmentation. Keep the people, make them better at the job, grow the business because they can do more than they could six months ago. It’s the harder story to tell internally, because there’s no clean line item for it. You can’t put “my team got better judgment” in a budget row the way you can put “we cut two roles.”
Ask a founder which one they want, and augmentation wins almost every time. Watch what they actually purchase six months later, though, and it’s usually a tool built for the first path anyway, because a number on a slide is easier to defend to a board than a claim about better judgment.
That’s the fork founders default to without meaning to. Automation because it’s legible. Augmentation because it’s right. Nobody makes that trade on purpose. They just pick whichever version survives the room, and augmentation rarely does.
It’s worse for a smaller company than a big one. A 500-person org can absorb a bad automation bet and barely notice. A 15-person company feels every wrong call immediately, because there’s no layer of middle managers to quietly compensate for it. And the instinct that runs a small company under pressure is almost always the same one: the answer to a bottleneck is to hire someone, or to buy the tool that promises to make the bottleneck disappear without anyone having to get better at the job underneath it. Both moves dodge the harder question. Neither one answers it.
The catch: neither path fixes what’s actually broken
The part that gets skipped is also the expensive part.
Neither path fixes a company that isn’t aligned on what it’s actually trying to do. Automation without alignment doesn’t solve the confusion. It runs the confusion faster. I’ve said this before, to a leadership team stuck exactly here: if the leadership isn’t aligned on what the goals actually are, all you’re going to do is produce faster garbage.
Not slower garbage. Faster. A team that disagrees about the goal, handed a faster tool, doesn’t converge on the goal. It ships the disagreement at a higher volume, with more confidence, because the output looks more finished than it is.
There’s a second failure mode hiding in the same misalignment, and it might be the more dangerous one, because it never looks broken. The work ships on time, clean and competent, aimed at nothing the business actually needed, because nobody upstream ever agreed on what that was. Nothing catches that kind of output. It gets praised instead.
That’s the trap that swallows automation projects specifically, because automation promises a clean, bounded fix: replace this one task, get this one number back. It hides the fact that the task was never the real problem. The real problem sat upstream, in a room where nobody had agreed what “done” or “good” actually meant. Automating a disagreement doesn’t resolve it. It prints more copies of it, faster than a person ever could.
Augmentation doesn’t automatically dodge this either. Give a person a faster tool without alignment on what they’re actually optimizing for, and they’ll use it to produce more of whatever they already believed mattered, whether or not that’s what the business needs. The tool isn’t the fix. The alignment is the fix. The tool just decides how fast you find out whether you had one.
What augmentation looks like in practice
So what does augmentation actually look like, past the tidy phrase?
The only version I can speak to honestly is the one I run myself. My operation runs on a crew of named AI agents, not one tool bolted onto a task, each one taking a specific piece of the load off my plate so the actual judgment calls stay mine.
I won’t dress this up as a case study. There’s no control group here, no outside numbers proving it’s the right call for your company. It’s my own bet, made because I’d rather run the thing I sell than describe it from a deck. What changed for me is mechanical: typing and formatting shrank, and so did the time I spent chasing old threads. What didn’t shrink is the part that actually requires a judgment call, and if anything it gets more of my attention now that less of it is bleeding into the typing.
I could have bought a tool instead, one built to write the draft and catch the mistake before it ships. I didn’t want the version where the tool makes the call and I sign off after the fact. I wanted the version where I’m still the one deciding what’s true and what’s good, and everything else just moves faster around that decision.
That’s the test for whether something is augmentation, or automation wearing augmentation’s language. After you install it, is the human doing more of the judgment work, or less of it? Automation quietly moves judgment out of the room and into the machine, one convenient decision at a time, until nobody notices the room is empty of it. Augmentation keeps moving more judgment toward the human, because the human is finally free to carry it.
The test to run on whatever you’re building
There’s one test worth running, if you’re in the middle of an AI rollout right now, at any stage, in any part of the business.
Look at the people who are already there. Not the software. The people. Are they better at their jobs because of what you just installed? Not busier, and not just faster at some narrower task. Better, in a way you could actually point to: trusted with more judgment than they carried six months ago.
Or did you just take the judgment that used to live in their heads and hand it to something that runs without them?
That’s the whole question. The roadmap, the tool selection, the rollout plan, all of it sits downstream of how you answer it.