The final strategy readout ran to ninety minutes and forty slides. Good work, on the whole: the ambition was sound and the diagnosis was fair. It all came apart neatly into six workstreams, each with a name, a rationale and a rough sequence.
Everyone in the room agreed with the map. Then someone asked who would run them.
The answer took a while to arrive, and nobody said it out loud when it did. It arrived as a list of function heads: supply chain, quality, operations, finance. Each of them now owned a workstream on top of the job they already had. None of them had asked for it. None of them was going to say no. Within a fortnight they were project managers, and the organisation had appointed its entire delivery capability by default.
That moment deserves more attention than it gets. A strategy always assumes a delivery capability, and the assumption is almost never tested during the strategy work. When the gap finally shows, few organisations admit it. They fill it with whoever is standing nearby.
The project managers you already have are the wrong project managers
Most industrial organisations do have project managers. They sit inside engineering, they often hold certifications, and many of them are genuinely good. Capital projects, plant extensions, equipment qualification, commissioning schedules — real discipline, earned on work where the deliverable is physical and the finish line is visible from a distance.
Operating-model change is not that work. Nothing gets installed. The deliverable is a way of working that exists only once people adopt it. Most of the effort therefore sits in stakeholder ground rather than in the schedule. Risks are political before they are technical. Nobody signs a factory acceptance test for a new planning process.
So the organisation holds project management competence with no domain match. Moving an engineering project manager onto an operating-model workstream is not obviously safer than appointing the function head. Both of them are learning the same unfamiliar half of the job.
A foundation course is no longer the proportionate answer
The textbook response is to run the accidental managers through project management onboarding. Book the course, issue the templates, appoint a mentor, wait a quarter.
It rarely works, for reasons that have nothing to do with the quality of the training. Method learned away from the work competes with a day job the moment the course ends, and the day job wins. By the time the cohort finishes, workstream three is already late and the templates are being filled in retrospectively.
There is also a proportionality argument, and it comes from the standard rather than from preference. The eighth edition of the PMBOK Guide sets out forty processes. It also states plainly that they are not prescriptive. They exist to be tailored to the approach, the life cycle and the environment. A demand for full-method compliance on a six-workstream transformation has no authority behind it. The standard does not ask for it.
The method layer has become the cheap part
This is where the arithmetic has genuinely changed. A well-configured GenAI project agent covers most of what the foundation course was for. It covers it at the moment of need rather than six months early.
It holds cadence. Updates that have not arrived get chased, and a vague first answer gets a second question. The organisation’s own procedures sit inside it, because those documents can be handed to it directly: which gate needs what evidence, which template is mandatory, which approval sequence applies. It drafts the status pack, the risk entries and the first version of the schedule. Method questions get answered in context, at eleven at night, without the accidental manager having to admit to a colleague that they do not know.
It also notices inconsistency. A plan saying one thing and an update saying another is exactly the pattern an agent catches reliably and a busy human misses.
None of that is trivial. It is most of the mechanical load. Removing it separates an accidental manager who copes from one who quietly stops updating anything.
An agent cannot be accountable for anything
What the agent will not do matters more than what it will, because the gaps are the ones that sink transformations.
- It cannot carry accountability. When a milestone slips, someone has to answer for it in a room full of people who outrank them. An agent has nothing at stake and cannot acquire any.
- Flagging a decision is not making one happen. The agent will report an overdue decision every week, indefinitely. Bounding how long a decision may take requires standing that no tool has.
- Resource negotiation is political work. The two engineers the workstream needs report to a director with other priorities. That conversation is won on relationship and leverage, not on evidence.
- It cannot refuse the sponsor. Saying no to added scope requires a position in the organisation. An agent will absorb whatever it is given and dutifully replan around it.
- Outputs and outcomes look identical to it. Training delivered, process published, tool deployed — all of it closes cleanly while behaviour stays exactly as it was. The agent marks the task complete and is not wrong to.
- It only knows what somebody wrote down. The corridor warning, the supplier who has stopped replying, the team member who has mentally resigned — none of it reaches the register, and none of it reaches the agent.
- Tailoring is a judgement, not a rule. Deciding which parts of the method are disproportionate here takes someone with context. An agent applies the rules it is given, thoroughly, whether or not they earn their cost.
- It cannot resolve a collision between two workstreams. Each agent optimises the project it serves. Two of them arguing over the same scarce resource produce a well-documented deadlock and nothing else.
Reporting throughput is not decision throughput
The pattern to watch for is an organisation that mistakes better records for better delivery.
Governance is the right to decide. Reporting only informs the decision, and the two are constantly confused because reporting is visible and deciding is not. An agent will make the records excellent. What it produces, if nothing else changes, is an impeccably maintained account of a decision nobody made.
This is the unstated assumption underneath every project management framework in use: that when a decision is required, someone with authority makes it within a reasonable time. Where that assumption fails, the standards go quiet. No chapter covers what a project manager does when the steering group declines to resolve a priority conflict for the third month running. The silence is not an oversight, since it sits outside what a consensus standard can address. It is also precisely where transformations live, and where the mandate question surfaces.
An accidental project manager is worse placed on this than anyone. A professional project manager knows an unmade decision is an escalation. A function head running a workstream part-time reads it as a delay and works around it.
Somebody has to watch what the agent cannot see
This is the case for a shadow project manager beside each accidental one, or a PMO holding that function across the set. Not a reporting line and not a compliance check, but an experienced person owning the part that cannot be automated.
- Who decides, and by when. Every open decision needs a named owner and a date. Unbounded latency is indistinguishable from no governance.
- Collisions between workstreams. Two workstreams needing the same scarce people is not a project issue, and escalating it as one guarantees it recycles. It is a portfolio decision somebody is not making.
- The benefit no single workstream produces. Advisers cut work into parallel streams because streams are legible. Value lands where they combine, and that joint outcome has no owner unless one is appointed.
- Dependencies visible only from above. Each accidental manager sees their own stream clearly and the seams between streams not at all.
- Scope drift arriving as clarification. Additions rarely announce themselves. They turn up as a refinement in a meeting nobody minuted.
- Risk that is managed by not being written down. A function head has a career here and a register that colleagues can read. The incentive to soften is entirely human.
- Whether “done” means the output or the outcome. Someone must own transition and sustainment, and that ownership has to be designed in before the workstream closes.
- Actual capacity. The day job wins every contest for time. A workstream owner’s availability is a forecast, and it needs re-testing monthly.
- How much method stays switched on. Someone has to make that call deliberately, or the answer defaults to everything or nothing.
- The point at which the work outgrows the person. Somebody has to notice when a workstream has passed what a part-time owner can hold, and say so early.
Accidental is a status, not a role
The arrangement has an expiry date, whether or not anyone sets one.
Three honest endings exist. The person becomes a project manager properly, with the title, the time and the development to match. The workstream gets a full-time professional and the function head returns to their function. Or the work finishes before either question becomes urgent.
What should not happen is the fourth ending, which is the common one: nobody decides, and accidental quietly becomes permanent. The organisation then holds a delivery capability made entirely of people who never chose the role and were never developed for it. They cannot be held to it. They will disappear back into their function the week that function needs them. Such an arrangement survives on goodwill until the first serious conflict.
Deciding which ending applies is a governance act. It belongs in writing at the start, in the same document that establishes what the delivery function may actually decide. The alternative is a conversation twelve months later about why nothing landed.
The agent buys the organisation time. It does not buy it a project manager.