Over the past two years, most companies have given their developers AI tools. Coding assistants, chat and better search. It works, and developers get there faster.
But it's mostly like giving them an e-bike. You get there faster, but you're still doing the pedalling. A developer writes the prompt, reads the response and approves every single step. The context has to be rebuilt in every new conversation, and the quality depends on who's riding the bike.
That's why the gains level off. According to analyses from BCG and Bain, among others, they typically reach up to 30%. That's a lot, but it doesn't change how the business works.
The self-driving car is something else
The next level is handing the task over. A person describes the destination: what needs to be done, where the boundaries are, and when it's finished. A team of AI agents takes the journey from start to finish. They build, test, review each other's work and document it, and a person assesses the result at fixed approval points.
Here, delivery no longer scales with human attention. It scales with computing power and clear goals. That's a completely different equation.
The car is the easy part
Every AI model provider wants to sell you the car. But a self-driving car goes nowhere without roads, maps, traffic rules and a driving licence. That's where most companies are today.
The gap between the e-bike and the car isn't about tools. It's about infrastructure and trust. Is it defined what done means before the work starts? Are architecture and decisions written down, or kept in the heads of a few people? Can you see that a change is right without someone reading every line? And who vouches for something no human has written?
And then the bottleneck moves
Once the car is moving, something happens that matters more than speed. The bottleneck moves.
For years, leadership teams have prioritised digital projects based on how much the development team could get through. The rest ended up on the backlog. When building becomes significantly cheaper and faster, capacity is no longer the limit. Two other things are: deciding what to build, and judging whether the result is right.
That puts pressure on the business, product and leadership, not just the technical organisation. And it raises three questions many haven't been forced to answer clearly before.
- What should be built? When almost anything is possible, saying no becomes a strategic decision.
- In what order? The wrong order costs more than before, because you can end up building a lot of the wrong thing.
- What is it worth? Projects that were shelved for being too expensive need to be recalculated.
The budget moves too
It's tempting to think the gain lies in fewer developers. But the money moves rather than disappears. It shifts from headcount towards the foundation the car runs on: tests, environments, documented knowledge and reskilling the people who will steer the agents. The computing power itself is the smallest item.
This is also where the biggest risk lies. If you turn up the speed without building the roads, you just get faster delivery that no one can vouch for.
Part of the bigger journey
For us, this isn't a sideline. Digital transformation is about moving the business, and development has always been one of the most expensive and slowest parts of that journey. Agentic Delivery is our answer to exactly that part. We build the roads so you can focus on the destination.
Where do you start?
Don't start with the technology. Start with your portfolio and an honest look at the roads. Which projects are waiting? What would you do if capacity weren't the constraint? And how far are you from being able to let go of the wheel?
Those are exactly the questions we address in a discovery meeting. We give an honest assessment of where the new capacity creates the most value, and where it doesn't.