BlogsSep 25, 2026

Work is a performance and there is a new arrangement in town.

AI Agents change what individual performers can do. That is not the same as changing the work system.

AI Agents change what individual performers can do. That is not the same as changing the work system. A faster or better step can leave throughput unchanged if the constraint sits downstream. Substituting an agent can cut labour and leave routing the same. Combining roles can be invalid when the tasks sit on mutually exclusive branches. An agent can be able to do the work while the operating model still requires review, authority, escalation, or deterministic control.

Quality, customer experience, and throughput that supports revenue are ordinary reasons to use agents. So is cost. None of those outcomes, by themselves, tell you whether the current way of working should stay in place.

The obvious first move is to put agents on the existing path and watch the metric the function already owns. That can be the right move. It can also improve the scoreboard while the constraint is still coordination, routing, capacity, authority, or control. Then quality needs more checkers, the customer still waits on the same acceptance step, and added volume dies in a queue you did not redesign.

Bot0 asks whether a change in what performers can do should change how the work system itself is designed.

What to do now?

Before you scale agents on a process, separate two decisions. Keep “change who performs the step” separate from “change how the work is governed.”

You can start from evidence you already have: process mining, case traces, SOPs, operating write-ups. The product models the current system, tests performer changes against that system, shows the consequences, compares alternative arrangements, and produces an implementation-ready specification for the agent stack you already use.

An arrangement here is not only governance. It is the architecture of the work system: performers, routing, coordination, authority, capacity, and controls. That includes when an output stops being a proposal and becomes the authoritative business result.

Tell me how I am measured, I will tell you how I perform

Get a dynamic picture of how the process runs across people, agents, software, and machines, including what the source did not specify.
Performer substitution and augmentation tested without pretending the rest of the system moved.
System consequences made visible: unchanged throughput, invalid role merges, review and authority that still bind, controls that still apply.
Alternative arrangements compared against the current design as the baseline, one coordination boundary at a time.
Implementation-ready specifications that deployment and engineering can build from, for the agent stack you already use.

What about cost control?

Before you deploy — you need a complete cost of running an arrangement. Coordination, review, control, audit, and the cost of being wrong are only partly in the model; a missing cost is not treated as zero.

On any workflow or system, are agents here to perform work under the current design, or is the current design what will cap quality, experience, or growth? If you cannot name that boundary, scaling agents will move the metric you already watch and leave the constraint in place.