When the software is the workforce
The digital workforce is becoming a standard rather than an experiment. The Fieldbook tracks six organisational design patterns in production, from the agentic team to the diamond org, and six roles that did not exist as jobs two years ago: agent supervisor, eval owner, exception handler, human-in-the-loop reviewer, agent owner and orchestrator. These are not exhaustive, but they show how businesses are starting to shape their workforces and the leaders within them.
The Agentic Team
2–5 multidisciplinary humans supervise an "agent factory" of 50–100 specialized agents running an end-to-end outcome (onboard a customer, launch a product, close the books).
The Diamond Org
The pyramid collapses. Mid-level managers become high-value Orchestrators; hierarchy is organized by cognition, not seniority. (a.k.a. "no more pyramids".)
Above the Loop
Humans move above the loop to steer and direct outcomes, dropping within the loop only where human contact genuinely matters. Work shifts from doing tasks to orchestrating outcomes.
A human AE owns the relationship; SDR agents prospect and qualify at scale, enrichment operators feed them, a HITL gate guards outreach.
Agent-does / Human-supervises
The operating-model default: agents do the work, humans supervise. Every deployed agent has a named owner accountable for its behaviour and output.
One human pod (2–4) supervises a fleet of tier-1 resolution agents; exceptions escalate to humans, an eval owner holds the quality bar.
A controller supervises reconciliation agents that close the books end-to-end; a validator and approval gate keep it auditable.
The Five Pillars
An agentic organization rewires five pillars: business model · operating model · governance · workforce/people/culture · technology & data.
Workforce Unit Abstraction
Governs hybrid human + AI operations by treating each worker, human or agent, as a uniform "workforce unit" with capabilities, permissions and accountability, so org charts can mix both.
Agent Supervisor
Oversees a fleet of agents, sets goals, monitors behaviour, intervenes on drift.
Eval Owner
Owns the quality bar, defines and maintains the evaluation criteria every agent is graded against.
Exception Handler
Handles the off-spec cases agents escalate: the human fallback when the agent hits ambiguity.
Agent Owner
The named accountable owner every deployed agent must have: owns its remit, cost and outcomes.
Orchestrator
The evolved mid-manager, coordinates a hybrid team of humans and agents toward an outcome. Hybrid orchestration becomes the core managerial skill.
Two to five humans supervising a fleet, and the roles that emerge
Agents are becoming an everyday part of the workforce, and the next phase is structural: departments built around them, piloted by subject-matter experts and strong operators rather than by engineers, with a named lead accountable for what they do. Each of the six roles is the same shift, from doing the work to specifying and checking it, and the organisations getting value restructure first and deploy second.
Show more
The people who once played the instruments are becoming conductors. The agents work the tools, find the notes and build the harmonies; the humans decide what is played, when, and who plays with whom. Done well, that is how a business turns a fleet into a symphony rather than a noise: more output, less friction, and a tempo it could not hold before.
What is an agentic team?
A small group, roughly two to five people, supervising a large fleet of specialised agents that run an end-to-end outcome such as onboarding a customer or closing the books. The humans set goals, watch behaviour and handle exceptions. The agents do the task work.
What does "above the loop" mean?
Humans move up from doing the work to directing it, dropping back into the loop only where human contact or judgement genuinely matters. The agent loop runs below them. In practice it turns a manager's job from execution into orchestration.
What is the diamond org?
What happens to the pyramid when agents absorb the bottom layer. The junior operational base shrinks, the expert middle widens, and mid-level managers become orchestrators of hybrid teams. The org sorts by what someone knows rather than how long they have been there.
What new jobs does AI create inside a company?
The ones that only exist once agents do the work: agent supervisor, eval owner, exception handler, human-in-the-loop reviewer, agent owner and orchestrator. None was a job title two years ago, and they are all the same shift: from doing the work to specifying and checking it.
Who is responsible when an AI agent gets it wrong?
A named person, decided before deployment rather than after the incident. The working default is that every deployed agent has one human owner accountable for its remit, cost and outcomes. Accountability cannot be handed to the model, which is why the org chart matters as much as the technical controls.