One of my observations from the past two years helping teams transform their AI organizations is that the lift provided is not even. This has become more of a conversation in recent weeks, and I wanted to spend a moment on it. In my opinion, the gain we see inside these organizations is largely k-shaped and favours the hackers and more senior workers.
This observation is a little uncomfortable, but naming it is not the same as withdrawing support. It helps us understand what we are asking of the team and how people fit into the equation.
I think teams also struggle to conceptualize this problem because there are really two separate accelerations that are largely conflated within an organization.
1. Tool acceleration. The clearest example is the auto-IDE sitting next to a developer. Same developer, same job, faster shovel. The pattern shows up across the team too - the marketer drafting a first pass, the analyst pulling a query they would have asked for help on a year ago. The lift is real and it is broad - but it is the tool getting faster, not the person doing the work. Some of this work is what deterministic flows will absorb over time, some of it won't, but either way the ceiling is the tool itself.
2. Orchestrator acceleration. A smaller group of builders is getting faster as builders. Where tool acceleration shrinks the time on a point task, orchestrator acceleration changes the size of the thing a single person can take on. Same person, larger solution, with auto-builders alongside them as collaborators rather than shortcuts. They are reaching for more ambitious work, with less support around them, and shipping it. This is the one most often missed in productivity conversations, and the one largely responsible for the k-shape.
Two years ago we used to describe AI workflows as a factory line. Picture ten expert woodworkers on the floor, each shaping a part by hand. Now picture the person maintaining the line itself - the one who has to both spot when something is off in the wood and know how to tune the machines that are increasingly doing the shaping. They are woodworker and engineer at once. The woodworkers on the line are the tool-accelerated: faster at the thing they were already doing. The woodworker-engineer keeping the line running is the orchestrator: a different job, with a different ceiling.
Orchestration is the same job in every lane: How will I break this down. How will I de-risk it. How will I attack it. It does not require being the deepest expert in any one lane. It requires enough understanding of every lane to direct work across them.
How we saw this at Isle of Code
I originally founded Isle of Code in 2012 as a dev shop servicing Nasdaq, Siemens, AdParlor, AssetMark, TD, and others. By last summer, a two-person pod was producing what would have historically been four to six FTEs of work, and that number has only accelerated since. For the most part, the continued lift has come from orchestrator acceleration and not tool acceleration. Observing this is what convinced me the k-shape was real.
About a year ago, my partners and I shifted our own workflows and got more directly back into the active lines. That worked because the three of us are generalists by trade and had a lot of skill in this orchestration - the work of standing at the top of a problem and asking How will I break this down. How will I de-risk it. How will I attack it. For us, AI was an amplification of skills we had been building on. From where we sit, this is what an agency leader has always done. The historical role - sizing the work, sequencing it, pulling the right people onto the right slice at the right time - is the orchestration job. The tools changed. The job did not.
Hackers, and a certain persona within an organization, land in a similar place from a different starting point. Anybody who is a self-taught builder or maker, who has learned by doing and is comfortable not being the deepest expert in any one lane, is probably already expanding their reach. These people were maybe already operating that way - or trying to - before AI showed up.
The bet on depth
Most of the work we have hired for, typically development work, over the last 15 years has rewarded depth in one lane. The best engineers were the ones who could write production-grade code at scale, and the best salespeople were the ones who could carry a complex enterprise deal from first call through procurement without losing the thread. Staying inside the box was heavily rewarded, and most highly skilled ICs were intentionally protected from orchestration work so they could keep compounding inside their craft. This was the correct call at the time, but it is no longer building the skill that supports us in the new world.
Tool acceleration is where most of the org lands by default. It is the bottom leg of the k - the part deterministic flows are coming for. Some of that work will be absorbed entirely. Some of it won't. Either way, it is not where the lift compounds.
The k-shape was always there in teams. AI turned the volume up on the one skill at the top, instead of turning the volume up on the number of people enabled to pursue what their persona has always strived for. The lift did not broaden access, it concentrated the top.
There is a popular fix of AI literacy training, workshops, and internal academies, which functionally bets the gap is tools. From where we sit, that is correct for the first (tool acceleration), but the real gap is orchestration, which was not generally on people's curriculum and is a much more challenging skill to build.
The asymmetry is the whole story. The bottom leg shrinks as deterministic flows absorb it, and the top compounds in the hands of the few who can orchestrate. The distance between the two is what grows.