AI has entered the building. Time to rethink how we work.

AI has made its way onto the agenda of almost every company. Organizations are looking to it for productivity, better decision-making and more intelligence in how they operate. Yet the results vary dramatically. Some report enormous leaps in productivity, while others are left cursing experiments that have fallen completely flat.

We believe the difference often lies in the operating models and workflows that AI is being plugged into. If we want AI to succeed, those are what we need to put under the microscope.

There is a common misconception around AI adoption: give the existing organization a good enough tool, and productivity will improve.

Here’s an example. At one growth-stage gaming company, testing had become a bottleneck in product development. An AI-based solution reduced the time needed to test games across different phone models to a fraction of what it had been. The company breathed a sigh of relief. Finally, the bottleneck had been solved.

Except it hadn’t. It had simply moved.

Production could now move so quickly that marketing could no longer keep up. Technology alone wasn’t enough to solve the problem. The operating model had to change too. People in different roles were brought more closely into product development, and ways of working were redesigned to become less linear. Together, the new operating model and AI genuinely accelerated the product development cycle.

When one stage of a process becomes radically more efficient, the balance of the entire system changes. If AI is introduced without questioning how work gets done, we may simply end up automating existing friction. We can make a seven-step process run faster when the real solution would be to eliminate four of those steps. We can produce a report in five minutes when the more important question is whether anyone needs the report in the first place.

So perhaps the right question isn’t “Which task could AI do faster?”

It is: “How would we design this work if we were building it today, with the technologies now available to us?”

Operational work contains an enormous amount of organizational busywork: documenting information, reporting, monitoring and coordinating. The time spent on this is time away from core work. Yet much of it is necessary to maintain a shared understanding of what is happening and keep the organization moving in the same direction.

Imagine a construction site.

Instead of site management spending time at the end of each day gathering information and writing reports, an AI agent calls the people responsible for different parts of the site and talks to them about what happened during the day. It connects those observations with schedules, plans and previous progress, creating a shared picture of the situation for everyone involved.

Things become even more interesting when AI doesn’t merely gather information but uses that shared picture to help direct the work itself. The system could identify a deviation from the schedule, assess its impact on the next stages of work and bring the issue to the right people for a decision. At that point, AI isn’t simply automating reporting. It is changing the workflow of information, coordination and decision-making.

Automation, however, comes with an uncomfortable paradox, one that Lisanne Bainbridge described as early as 1983 in her classic paper Ironies of Automation: the more a system automates, the more demanding the tasks left to humans can become.

If AI handles 80% of the process and humans only step in when something goes wrong, humans are a pretty poor failsafe. It is difficult to see the bigger picture or make sharp observations about a process you are no longer routinely part of.

This is why, for example, a construction site manager might be better off using the time freed up by AI to spend more time walking the site and talking to people. Then, when the agent flags a potential problem, the manager has enough context to judge whether the problem is real and what may be causing it.

The goal, then, cannot simply be to transfer as many human tasks as possible to AI. Instead, we need to find the most effective combination of people, technology, processes and decision-making.

When workflows change, roles, responsibilities, capabilities and decision points have to be reconsidered too. Successful AI adoption is therefore as much about redesigning the organization’s operating model as it is about developing the technology.