The RPA centre of excellence was the right answer to a specific question: how do we bring hundreds of bots under control that are springing up all over the company? It set standards, ran an orchestrator, managed licences and kept the bot farm running. Agents pose a different question — and the old organisation no longer quite fits it.

Why the bot farm is not enough

A bot is a script with a schedule. An agent is an actor with a goal, an identity and discretion. That changes operations in three places. Exceptions are no longer crashes a developer fixes but decisions a business expert makes — they need a queue with an SLA, not an error list. Costs no longer arise per licence but per call and per outcome — they need budgets per agent, not per server. And changes no longer concern only code but models, prompts and policies — each has to be versioned, evaluated and approved before it goes live.

Hub and tenant

The operating model that answers this we call hub-and-tenant: one central operations hub providing the platform, monitoring, evaluations and governance tooling — and per business unit (or per client, when a service provider operates) an isolated tenant with its own agents, identities, exception queues and budget. Data never crosses tenant lines; responsibility sits where the domain knowledge is; the platform supplies the discipline that has to be the same everywhere.

The difference from the bot farm is not the technology but the question of who owns what: the hub is responsible for agents running safely. The tenant is responsible for what they do.

What the automation team does next

  • From bot builder to agent engineer. Anyone who could decompose processes into RPA can decompose them into goals, tools and guardrails. The skill stays valuable — the building blocks change.
  • From orchestrator to operations hub. The team that ran the bot farm is the natural team for the hub: monitoring, exception queues, evaluations before every release.
  • From process expert to policy author. Whoever knows how a process really runs is the right person to define what an agent may do — and when it must ask.
  • From licence administrator to cost governor. Steering spend per agent and per outcome is the new core competence.

The pragmatic path

Nobody has to abolish the CoE. The most honest path is to stand up a first tenant next to the bot farm — one business unit, one agent, one exception queue, one budget — and build the hub out of what operating that first tenant demands. The bots keep running in the meantime; they become tools the agents call. This is not an overnight replacement but a shift of emphasis: from how many bots are running to which outcomes were demonstrably delivered.