When AI agents become abundant, new ways of working become possible

A purchase order does not only cost the work of raising it. It costs getting the work to happen, holding the request together, checking what must be checked before the company commits, and bearing the consequence if the result is wrong.

A useful way to express that is:

A group of agents can make doing it look cheap while making the overall work more expensive. More participants can mean more context, calls, waiting, reconciliation and review — and sometimes a more expensive mistake.

If you only count execution, you are not comparing arrangements. You are comparing one part of their bill. An arrangement-changing proposal should not be declared economically superior from execution savings alone; where it creates coordination, review, control or failure exposure the model cannot yet measure, the answer is not zero. It is that the economic comparison is incomplete.

Moving from one way of working to another also costs something once — integration, migration, training, redesigning policy and roles, rollout, adoption. That is change cost, and it sits beside the operating bill rather than inside it:

C = E + C_e + R + I + pL   what an arrangement costs to run
C_h   what changing arrangements costs, once
the case for the change = Δ(value − C) − C_h

A build-out you pay for once does not pre-pay the checking and controlling you will do on every case. Keep the one-time and the recurring apart, or the recurring quietly disappears into a number that was already spent.

Organizations have historically designed work around two scarce resources: human attention and human judgment. That naturally favors defined processes and management. If the next action can be encoded in a process, nobody has to continuously watch and decide what should happen next. Where it cannot, a manager can make the call.

Consider a shared operational board. A shipment becomes late, inventory falls below plan and a customer commitment becomes exposed. Logistics, procurement, production and customer service can see the same changing state and respond when something relevant to their role occurs. With people, that can require several people to remain attentive, interpret what changed and decide whether to act.

Or consider a peer network. One participant raises an issue, another contributes evidence, another challenges the conclusion and another revises it until an explicit convergence rule is satisfied. Doing that with people for every case is expensive.

Organizations have therefore often simplified richer forms of coordination into alerts, queues, assignments, meetings and handoffs.

Reasoning AI changes those constraints. AI agents can remain available, observe changing conditions, interpret them, reason within stated boundaries and coordinate with others at a very different cost.

But coordination does not become free. The constraints shift toward context, computation, latency, verification, reconciliation, control and trust. Ten agents can still generate more work than they absorb. The scarcity moves.

A shared board, market, ledger or peer network can therefore become a viable design choice where previously it may have been uneconomic. Changing who performs the work and changing how the work is organized are economically different decisions.

Review is not simply “being careful.” It follows from what must be true before the result can become official, including required evidence, agreement and reversibility.

Changing an arrangement can move several terms at once. A five-agent team may reduce human execution cost while increasing coordination, latency and review. One capable agent may cost less overall. Conventional software may cost less still. A deterministic route may win when the work is predictable or the consequences of failure are high.

The commonest mistake in an agent business case is a chain of inferences that feels like arithmetic:

30% of the work moves to an agent → 30% more capacity → 30% lower cost → $X benefit

None of the later steps follows from the first. Allocation is only allocation. It says who performs the work. It does not say how fast, how well, or what it costs.

Capacity comes from the performer, not from the share of work handed over. If an agent genuinely processes two thousand purchase requests a day and the arrangement lets it serve the station, then the station’s capacity really is two thousand a day — a real operational gain, and the number came from a fact about the agent, not from the percentage. But capacity is not throughput. If forty requests a day arrive, roughly forty a day are completed. The agent has fifty times the capacity; the work is not fifty times larger.

An operational improvement is already a benefit. Capacity up, defects down, exceptions down, cycle time down, first-pass yield up — those are real whether or not anyone has put a dollar on them. What they are not is automatically a dollar. Economic benefit is the economic consequence of an operational improvement, and the model should say whether that consequence is expected, simulated or observed. It needs a causal link: more demand served, a hire avoided, rework hours not worked, a loss not incurred. Where that link is not known, the economic value is unmeasured. Not zero — unmeasured.

So an agent added to sourcing has two sides, and an honest case states both: what the deployment costs, and which properties of the work system it changes or is expected to change — processing time eighteen minutes to eleven, capacity, first-pass yield ninety-two to ninety-seven percent, exceptions eleven to six, span of control, or another operational outcome. Stating the cost precisely while leaving the reason for the agent vague is how a cost becomes a business case.

And with twenty agents, benefits compose the way capacity composes; they do not add. Two agents that each add fifty a day on a sequential route do not add a hundred. Two agents that each save a hundred review hours may be saving the same hours. The Formation is where those effects are reconciled — which is one more reason the Formation, and not the agent, is the unit whose economics are compared. Participant-, task- and role-level effects may be authored or measured bottom-up, but they must reconcile through the Formation before they become Formation-level economics; the same Formation can then roll into the economics of the larger work system.