Teemu Birkstedt
Board Member, Rework
Many large software organizations have given hundreds of developers AI tools, and many have already set targets for AI’s impact on EBIT. When I ask what the investment has delivered, the answer is usually that code gets written faster.
Why doesn’t faster coding shorten lead time?
After analyzing our customers’ ticket data with our analytics tool Rethink, we learned that coding takes up only a small share of the time a feature needs to get from idea to production. Most of the time goes to waiting for decisions, people or testing. If coding gets faster but the queues stay the same, lead time barely changes.
Four levels of AI use in software development
In the study we took part in, Laato, Birkstedt and Mäntymäki (ECIS 2026) interviewed 15 organizations and identified four levels of AI use. Assistants and AI integrated into the workflow bring moderate benefits at best. When agents complete whole tasks, output grows, but review load and friction grow with it. Only when automation validates the result do teams shrink and code stop being the bottleneck. Big gains therefore require redesigning the work.
How to link the token bill to EBIT targets
Redesign is expensive, though, and uses the most tokens, so you can’t jump straight to it. Rework first establishes from the customer’s data which level different teams operate at and where work waits. Then the most important changes are made, metrics are set for them, and the new way of working is brought into one value stream. Throughout, both the improvement and the token cost are measured every four weeks. This links the EBIT target to the token bill in a way that is grounded in numbers.
This is change management
It’s important to remember that this is change management. We have learned that when the numbers come from hard data, from the teams’ own work, and are presented at team level, the conversation moves from opinions to what to do next.
Software development is only the first function
Software development is only the first function where this shows up, because that is where tokens have been invested first. The same question will come next in product management, testing, customer service and finance. There too, it’s worth finding out where work waits before buying more tokens.
If your token costs are growing and the impact is unclear, I’m happy to talk.