Two routes for work with AI

Where it starts

The companies that invest most in AI grow about 10% faster in headcount than comparable companies. That comes from a study by Ramp and Revelio Labs of 21,559 American companies (June 2026). Companies that only took subscriptions or ran a few pilots did not grow faster.

According to the researchers, the advantage requires complementary investments, changes in the organisation and learning inside the company. Buying a tool is therefore not enough. The work itself has to change.

Who is going to organise that work differently? Transition research has a simple answer: a new company can, or an existing company can. Those are the two routes on this page.

Two words first

The transition researcher Frank Geels uses two words that help here.

• The regime is the way work is organised now: the habits, rules, roles and tools that make it happen the same way every day. An AI subscription fits in neatly. The work stays the same, a tool is added.
• A niche is a small, protected place where something new is tried. It is allowed to fail there, and people learn how it can be done differently.

A large change starts in niches and in the end changes the regime. The question is who does that.

Route 1: new companies take the work over

What happens. A new company sets up its work differently from the start, for example with agents doing part of the work and people doing the rest. It has nothing to take apart, because there is nothing yet. If that way of working proves better, the new company takes the work over from the established ones. Geels and Schot call this substitution.

What it asks of a company. Of the newcomer: a way of working that is different from day one, and the time to prove it works. Of the existing company this route asks nothing, and that is exactly the risk. The work and the customers go to someone else, and there is no moment at which the existing company has to decide anything.

Route 2: existing companies transform

What happens. An existing company reorganises its own work, with the people who already work there. According to Geels' research this often happens in steps:

1. First hedge carefully: watch, try, change little.
2. Then develop the new next to the old.
3. Sometimes a complete turnaround, down to the mission and the earning model.

What it asks of a company. Money, knowledge, time and above all nerve. Concretely:

• Choose one process and reorganise it with the people who do it now.
• Record for each step what an agent does, who judges the result and where a person decides.
• Treat that process as a niche: a place where it is allowed to fail and where the company learns.
• Measure what changes in the work, not only what the tools cost.

This is harder than buying a tool, because it touches habits, roles and decisions. It is also the route on which a company keeps its people and its knowledge.

There are more routes

Transition research knows more than two. Geels and Schot describe four, and a change can switch route. Under strong pressure from outside, it often starts with transformation according to them, and can still end in substitution later. Whoever buys tools today and leaves the work as it is, is in the first step of route 2: hedging. The next step is the work itself.

What I do not know yet

• Which route AI and work will follow. Geels' examples span decades, and we are only at the start.
• Whether AI causes the growth at those companies. The Ramp study finds an association and calls the mechanism unresolved.
• How the total number of jobs develops. The Ramp study does not measure that, nor where the new people come from.