Rabobank launched its Agentic Hub this month: a single, central way for roughly 1,100 engineering teams to put AI agents to work across software development — backlog refinement, code generation and review, test execution. The number that gets the headline is 1,100. The number that matters is one. One hub, instead of eleven hundred teams each improvising their own answer.
That distinction is the whole story. Most enterprises responding to the agentic moment do the obvious thing: hand every team a coding assistant and let a thousand experiments bloom. It feels like progress. It produces fragmentation — every team reinventing the prompt, the guardrails, the definition of done. Myrna Vonk, who leads the initiative, named the trap directly: AI doesn't just change how you build software, it changes how engineers collaborate and organise the work — and without conscious organisation, adoption stays scattered. So Rabobank organised it.
What the Hub actually bundles is the tell: technical products, reusable building blocks, and validated use cases. Teams build on pre-approved solutions instead of starting from a blank prompt. We've written before that an agent factory without a factory floor is just a workshop with good tools and no floor. This is the floor. A bank built it.
Map the Hub onto the operating model and the pieces line up exactly. Reusable building blocks and validated use cases are what we call ground truth and skills — standards encoded once, referenced by the agents on every build, so output is correct by default rather than by accident. A central set of technical products is the platform — the secure, reproducible foundation that lets agents do anything more than a demo. Without the building blocks, agents drift. Without the platform, the building blocks have nowhere to run.
Then the part that the breathless coverage usually buries: the humans stay in charge. Rabobank is explicit that engineers set the direction, provide the context, make the final calls, and validate what the agents produce. That is the 20% — not a caveat bolted onto an automation story, but the load-bearing half of it. Agents execute at scale in the middle; judgment and accountability sit at the edges. The bank didn't automate the engineers away. It moved them up the value stack — toward architecture, quality, and customer value — which is the only version of this that actually compounds.
And because this is a regulated bank, compliance is built into the Hub from the start, not inspected in at the end. That is a gate in the pipeline by another name: governance and innovation running together instead of against each other. It is also the quiet proof point. If a financial institution can put policy-enforced, agent-driven delivery in front of 1,100 teams and still satisfy its regulators, the "we can't, we're too governed" excuse just lost its last hiding place.
That's what makes this story land differently from the McKinsey scenes and the Silicon Valley software factories. This isn't a frontier lab or a venture-funded demo. It's a Dutch bank, in our market, operating the model at scale under real regulatory weight. When the proof shows up that close to home, the conversation in every neighbouring boardroom changes — from whether agentic engineering is real to why their own teams are still each improvising.
The honest read: a hub is the start line, not the finish. Rabobank is piloting with selected teams in the coming months, and a platform of building blocks only pays off if the operating model around it compounds — if the validated use cases multiply, the skills deepen, and human review stays sharp as more of the codebase is marked safe-to-automate. Don't mistake the hub for the transformation. The hub proves the teams can build on shared rails. The transformation is whether the organisation gets faster every quarter because of it.
Our read at Pareto is the same one we've held all along: platform first, ground truth first, human judgment at the edges, gates in between. We built our delivery model around exactly this shape — agents owning the 80% of execution, humans owning the 20% that carries the responsibility. What's new this month isn't the thesis. It's that one of the Netherlands' largest banks just stood it up for 1,100 teams and called it a Hub. If you're still running scattered pilots, that's the benchmark now — and the gap to it compounds. The question isn't whether to build the floor. It's how long you can afford to keep working without one.