What AI changes about running an ERP program

Most of the AI-in-ERP writing is about the software. This one is about the program, the thing I actually run: the plan, the testing, the change management, the governance, the contract, and the shape of the team. Agents change all six. Almost nobody is writing that from the client-side seat, so here it is.

Start with the honest headline. Agents take real work off the program and add different work back. The teams that win are the ones that see the new work coming and staff it. The teams that get surprised treat the agent as a feature and discover, in testing, that they bought a new class of thing to manage.

What agents take off the plate

The busywork is the first to go, and good riddance. Status compilation, the weekly ritual of chasing eight leads for eight slides, is a task an agent does well because it is reading structured data and summarizing it. Report generation, first-pass test data setup, meeting notes and action logging: these are assistant tasks, and handing them over is the easy, safe win.

Here is the quiet consequence. Once the agent compiles the status, the program manager loses the last excuse for spending the week on status. The job that remains is the job that always mattered more: closing decisions. Decision velocity, the number of real decisions a program resolves per week, is the metric I watch above all others, because configuration does not stall programs. Unmade decisions do. Agents do not make your decisions. They remove the noise that let you pretend chasing status was the work.

What agents add: testing changes shape

A report is deterministic. Same input, same output, every run. You test it once against an expected value and it stays tested.

An agent is probabilistic. Same input can produce a different action, because the model weighs context and picks. Workday's returning CEO put it plainly this year: you cannot have probabilistic outcomes in running a payroll. That single sentence is the whole testing problem. You are no longer testing whether a number matches. You are testing a distribution of behavior across edge cases, adversarial inputs, and drift over time as the model updates underneath you.

That is a larger test scope, not a smaller one. It needs new artifacts: a library of scenarios the agent must handle, a defined acceptable failure rate, a monitored path for the cases it declines, and a re-test trigger every time the vendor ships a model change you did not schedule. Programs that budget for agent testing as if it were report testing will find the gap in UAT, which is the most expensive place to find anything.

What agents add: change management for a half-human workforce

Change management used to mean getting people to adopt a system. Now it means getting people to supervise a coworker that is partly software. The manager who approves an agent's work needs to know when to trust it and when to stop it. The employee whose task the agent now does needs a new definition of the job and an honest answer about what that means for them. That anxiety is real and it does not respond to a training deck.

The organizational readiness data says most companies are not staffed for this. Deloitte's 2025 study found only 21% of organizations have a mature governance model for agentic AI, while adoption runs well ahead of the guardrails. Change management is where that gap becomes a program risk, because an unsupervised agent and an unsupported team fail in the same quarter.

What agents add: the governance seat every program now needs

This is the seat programs are missing. Someone has to own agent behavior the way a manager owns an employee: what it is allowed to do, what data it can see, when it acts alone, who answers when it acts wrong, and what it costs to run. Access and security for agents is not a go-live task any more than it was for people, and I have watched one unowned security decision stall an entire test cycle for a week.

The platform vendors are validating the seat by building for it. Workday shipped an Agent System of Record this year, generally available in February, whose whole premise is that agents get managed like employees, with an identity, a role, and an audit trail. When the vendor builds a system of record for the agents, that is the vendor telling you the agents need governing. The buyer who does not name a human to sit in that seat has outsourced a control function to a slide.

The 2027 program org chart

Draw your current program org chart. Now add and subtract.

The status-reporting layer thins, because the compilation is automated. The testing team grows and changes skill, because someone has to design and monitor probabilistic test suites. A new seat appears near the top, next to the PMO: the agent governance owner, accountable for what the agents do, see, and cost. And the program manager's own box changes color. Less status chasing, more decision governance, more exception handling on the calls that agents escalate.

Picture a program week. The hours that used to go to compiling status and preparing to talk about status collapse toward zero. They do not vanish from the calendar. They move to closing decisions, governing agents, and handling the exceptions the agents kick up. The total hours are similar. The work inside them is more valuable and harder to fake, which is exactly why some people will resist the change.

The contract changes too, and that is the next article

One more thing moves, and it moves money: the contract. Consumption-priced agents mean the bill follows the work the agents do, and the program that does not write telemetry, caps, and pilot fences into the deal will meet the meter the hard way. That is its own piece, and it is the one that sounds most like the work I do. It is next in the series.

Do this week: add one box to your program org chart before you run a single agent pilot. Name the human who owns agent behavior, access, and cost. Not the platform, not the integrator, a person on your side with the authority to switch an agent off. If you cannot fill that box, you are not ready to give an agent the keys.