Raiffeisenbank gave AI agents to 500 engineers, but speed didn't improve. Here's why

Raiffeisenbank has reported the results of the second part of its experiment on bringing AI into an engineering organization of 500 specialists: speeding up individual roles locally with coding agents did not change the key metrics Lead Time, Throughput and Defect Rate.

Why copilots didn't speed up the team

Giving coding agents to engineers does not, by itself, change the speed of the system. That is the conclusion of the first part of the experiment, which Raiffeisenbank shared in a blog post on Habr. According to the authors, you can grow MAU LLM (the monthly audience of a language model), then MAU API (the number of calls through the API), hand coding agents to people, and still see no change in three engineering metrics: Lead Time (time from idea to result), Throughput (the capacity of the development process) and Defect Rate (the share of defects in released code).

These three metrics are a standard indicator of the health of an engineering process, and the blog's authors call their lack of change after mass AI adoption the main result of the first part of the experiment.

• The experiment covers Raiffeisenbank's engineering organization of 500 specialists
• Metrics that did not change after AI adoption: Lead Time, Throughput, Defect Rate
• Local speedups run into requirements, review, testing, approvals, neighboring teams and business experts
• The company's next step is a move to Agentic Engineering

Where speed is actually lost

Speeding up one person or one role does not speed up the whole system, the authors of the Raiffeisenbank blog explain: the task still passes through all the other stages of the process. One employee may start doing their part faster, but the result still waits in line for review, testing, approval or a decision from a neighboring team. The authors describe this as a local optimization problem: speeding up one link in the pipeline does not change overall throughput if the bottleneck sits elsewhere in the process, with reviewers, testers or approving departments.

"One person or one role may start doing their part faster, but the task still waits for requirements, review, testing, approvals, neighboring teams or a business expert," says the

Raiffeisenbank blog on Habr.

What's next: Agentic Engineering

Raiffeisenbank did not stop at handing tools to individual engineers and decided to go further, into what the company calls Agentic Engineering. In the industry, this term usually describes a shift from point AI assistants for a single developer to agents embedded in the process at the level of teams and stages, rather than only at an individual's workstation. What exactly Raiffeisenbank means by the term, and how it will tackle the problem of approvals and queues between roles, the company promises to reveal in the continuation of its article series on Habr. The piece is explicitly titled the second part of the story of introducing AI to 500 engineers.

What this means

The Raiffeisenbank case shows that handing AI tools to engineers one by one does not speed up development if bottlenecks remain at the seams between roles and teams. A real effect appears only when the speedup touches the whole process, not the work of an individual.