If we are going to redesign organizations for an AI era, we need to be clear about one thing up front: Tools have almost always pulled people apart.
In organizatiomns, the adoption of tools increase specialization, enable separation, and make it easier for work to be done at a distance. Over time, that distance stretches the gap between action and judgment. It reduces alignment between individuals and within teams. It hides understanding, and also the underlying logic of how we got there. People produce artifacts without seeing how decisions are actually made.
But judgment in an organization needs to be transparent, both to ensure organizational alignment, but more importantly to grow the organization's ability to judge and assess. AI accelerates the separation of people within the organization, and that is precisely why the correct response to AI is not better tooling discipline, but tighter human coupling.
One of the most important functions in an organization, one that is hidden for most leaders and managers because of the way we have operated for the last several decades, is how judgment gets passed from senior to junior people.
What AI Breaks
Before AI, a junior had to balance time, rhetorical quality, and content quality. For example, a junior had to balance: not taking too long in the task, creating enough content quality to be comfortable in reviewing with a senior, and also developing a rhetorical narrative around it as part of the presentation. If it was rhetorical content that was being presented, then some of the Scott combined and mixed together. Improving one usually cost the others. That friction mattered. It forced reflection. Juniors spent time revising, discarding, and judging their own work because producing it consumed effort.
AI removes that friction.
Now juniors can generate polished rhetoric instantly and in volume. “Good sounding” content appears faster than the junior can judge it. The rhetorical wrapper becomes convincing enough to masquerade as quality.
Two things happen at once.
Juniors spend less time in critical judgment because production is too fast.
And they escalate more material to seniors, earlier, and with less internal grooming.
As Bret Starr, agency AI guru and methods pioneer, puts it: this is no longer garbage in, garbage out. It is garbage in, landfill out.
The challenge them to become, how do seniors keep up with reviewing all this work, and that the volume of content for review can actually slow down project progress. Yes, the iterations come faster, but that begs a question.
Why Not Just Have Seniors Do the Work?
If AI makes execution easy, why hire juniors at all? Why not let seniors do the work faster with better tools?
Because seniors do not scale. Senior judgment is the most constrained resource in any organization. If seniors do all the work, they become the bottleneck and the future disappears. If juniors are not exposed to real judgment today, they do not become seniors tomorrow. You cannot hire judgment later. It has to be grown.
The problem is not junior capability. It is junior isolation.
To see why, we need to rewind how judgment has actually been learned.
The Three Ways Humans Learn Judgment
There are three distinct process structures humans have used to transfer judgment. They differ mainly in how tightly judgment stays coupled to action.
The oldest, and most fundamental, is pairing.
Long before formal roles or professions existed, humans learned by working side by side. Parents and children. Hunters and novices. Farmers and helpers. Cooking, gathering, building, repairing. Decisions were visible. Timing mattered. Judgment was absorbed in motion, not explained afterward.
Pairing is not a technique. It is the human default.
The second structure is apprenticeship.
Apprenticeship emerges later, when production becomes repeatable and reputation matters. A saddle maker, a blacksmith, a needle maker. Juniors work independently, but remain near masters. Judgment is still visible, but less continuously. Intervention is selective. Learning is slower, but scalable.
The third structure is supervised specialization.
This is the modern organizational norm. Juniors work alone. Seniors review later. Judgment is explained after the fact, if at all. This model optimizes throughput, not judgment formation. It only works when work moves slowly and errors are cheap.
Most organizations today live almost entirely in this third model.
The Structural Inversion AI Forces
As production velocity increases, judgment loops must tighten. Supervised specialization collapses first. Apprenticeship often becomes insufficient. Organizations are pushed toward much tighter forms of working together.
Those of you who are familiar with my work over the decades, helping over 100 agencies and other delivery organizations boost their delivery throughput by 30 to 50% regularly, you can see where the velocity is enhanced by the group activities that we would introduce, like the check-in meeting, the collaborative briefing model, team, driven, scoping, etc. All of these brought juniors and seniors together for longer periods of time, and expose the judgment layer within the work so is to better enable the juniors, and also, more importantly grow them.
In those techniques still work and are valuable in this AI enabled world, but extraordinary times call for some extraordinary techniques, and I believe that one of the next adaptations will be something I learned almost 25 years ago.
The Pairs Method
In the early 2000s, a small group of software teams experimented with something that seemed inefficient at the time. Instead of assigning work to individuals, two people worked together on the same task, at the same time. One was usually more experienced. The other less so. They sat together, talked through decisions, and adjusted the work in real time. This practice later became known as pairs programming and was associated with a broader movement called Extreme Programming, or XP.
In the early 2000s, when I was CTO of a SaaS intellectual property management company, we experimented with pairs programming and our development team. What's interesting about this moment is that we have been using agile for a few months, and we're doing a good job of driving velocity, but we were doing so at the cost of quality. Traditional agile, which is based on the more exploratory model, tolerates quality problems that deadline driven organizations can't afford… And we were on a deadline.
It seemed crazy and counterintuitive, but by that point in my career, I had already learned that counterintuitive can often win the race. It looked inefficient, and people assumed productivity would drop...how could it not?
It didn’t. We went a bit faster. One of the big changes was the quality went up dramatically and most importantly, which I only realize now was that the density of judgment per unit of work had changed. Decisions were spoken instead of hidden. Reasoning was externalized instead of compressed. Juniors were not left alone at decision boundaries.
More recently, we saw the same effect in a pod structure I designed for an agency in London. Daily discussion of real work (and other close-couplings) pulled judgment back into the loop. Drift slowed. Learning accelerated. Their AI enabled pod boosted productivity by more than 50%, and slashed span time while achieving greater quality.
Different forms. Same principle.
When velocity outpaces judgment, the only stable response is to bring people closer together while the work is being done. Pairing is simply the most compressed form of that response.
Working Better Together Makes AI Work Better
AI makes it possible to work farther apart than ever before.
It also makes it more dangerous. The correct response is not more tools, more governance, or better prompts, but rather, working closer together.
The good news is that your organization is better at this than you realize. It is one of the oldest thing humans know how to do... and in many ways it defines us.
