The 'agent boss': why governing AI agents matters more than deploying them

By Zendoric · October 4, 2026.

The consultancy Setesca has put a name to a role many companies already need without knowing it: the "agent boss", or manager of intelligent agents. As its chief executive, Jordi Damià, explains in remarks reported by EFE, this is the person responsible for deciding which AI agents get built, what problem they solve, how they connect to the rest of the business and who answers for it when something goes wrong.

The figure Setesca uses to justify the position is striking: a well-integrated AI agent can free up to 150,000 work hours a year at a large organization by taking on repetitive tasks and automating operational decisions. At SMEs, the same source speaks of a 40% improvement in time spent on meetings and office tasks. These are the consultancy's own figures —and its chief executive also runs LiceoTIC Training, a training firm— so they are best read as an interested estimate rather than an independent measurement.

The problem that gives the role its meaning does have backing outside Setesca, and a name of its own: "agent sprawl", the proliferation of agents deployed in separate silos, with no governance model or shared metrics. "Many organizations are deploying intelligent agents in isolation, without integrating them into business processes or establishing a governance model that guarantees their effectiveness, security and scalability", Damià sums up. In September 2026, vendor WSO2 launched Agent Manager, an open-source control plane that gives each agent a verifiable identity, role-based access control, permission revocation and isolated execution in Kubernetes sandboxes (the standard system for orchestrating containers in production). According to InfoQ's analysis, the move confirms that governing agents is now a platform problem, not just a matter of good development practice.

Damià describes the agent boss as an in-house profile —never outsourced— who defines each agent's objectives and limits, measures its performance and cost, ensures regulatory compliance, coordinates collaboration between agents and people, and spots new processes to automate. The requirement is twofold: knowing how to code and understanding the business's processes in depth. "It is a profile geared towards generating value for the business, more than a technical specialist", he says. A company can lean on outside consultants to design the initial strategy, but as soon as agents are running critical processes, their governance has to stay inside IT, under the CIO (the executive ultimately responsible for information systems).

The market is already building the infrastructure for that governance. In June 2026, KPMG and Microsoft expanded their global alliance: the consultancy will roll out Microsoft 365 Copilot across its more than 276,000 professionals and adopt Microsoft Agent 365 to monitor both its own agents and those of its clients, with centralized security controls, according to Microsoft's statement. The trade magazine CRN identified at least ten platforms this year devoted to agent observability and governance —among them Arize AX, Datadog Agent Observability, New Relic AI Observability and Braintrust, which closed an $80 million Series B round in February, according to CRN—. All of them track the same things: latency, hallucinations, data drift, token consumption and permissions in systems where several agents work together.

The emergence of a role like this is not a semantic whim. It fits something we have been observing in our sector-by-sector review of AI's impact on employment: automation does not erase work, it shifts value towards judgment and coordination. The profile gaining weight is not the one who handles a particular tool best, but the one who decides when to use it, within what limits and under what human supervision. The "agent boss" is the corporate version of a pattern we have already seen in technology —where security and integration governance is on the rise— and in law —where expert judgment is on the rise—: the layer that manages AI gains value precisely when execution is automated.

Even so, it is worth separating two things the narrative easily blends. The problem is real: the investment by WSO2, Microsoft, Datadog, New Relic and at least half a dozen more platforms in governance tools is a verifiable market signal, not an interested story. The productivity figures —150,000 hours, 40% savings— by contrast come from the consultancy that sells the solution to the problem it describes, and have not been externally measured. The need to govern agents is confirmed by the market itself; the exact size of the benefit, not yet.

There is also a power angle that deserves attention. Whoever builds the governance layer —identity, permissions, auditing— ends up deciding which agents can act and which cannot, across the entire organization. Microsoft, with Agent 365 and its alliance with KPMG, and WSO2, betting on open source as an alternative, are already fighting over that position: the same pattern we saw in the struggle over interoperability standards between agents, where the winner is whoever controls the plumbing, not just whoever has the most capable model.

In the medium term, this fits our underlying thesis: AI frees up human time when someone takes responsibility for steering it well. The agent boss is not an HR curiosity; it is the piece that decides whether those freed-up hours translate into more capacity for people to devote themselves to what matters, or into a new technological chaos that nobody oversees. That difference will not be set by the agent, it will be set by whoever governs it.