Every few years something arrives that's supposed to make project managers redundant. Gantt software was going to do it. Then Agile. Then a wave of collaboration tools. Each time, the profession survived, mostly by absorbing the thing that was meant to replace it. AI is the current candidate, and the anxiety is real enough that I want to take it seriously rather than wave it away.
Projects have been run for as long as humans have built things. The pyramids. The Apollo programme. The Burj Khalifa. The scale and technology vary wildly. A constant runs through all of them: somebody has to work out what's needed, when and in what order, unblock what is stuck, and hold their nerve when the plan meets reality.
The Apollo missions are the example I keep coming back to. The coordination alone is staggering - every component arriving at the right place at the right moment, or you've got no rocket to put the astronauts in. But the part that gets less attention is the people. Those teams were doing things nobody had ever done, on a timeline a politician had picked out of the air, under a level of scrutiny most of us will never experience. The technical challenge was enormous. The human one was arguably harder.
What AI is genuinely good at
The practical case begins with information-heavy work. AI tools can summarise project records, compare updates, propose a first draft of a report and help identify patterns for a person to investigate. They may flag that a workstream is drifting or suggest ways to schedule scarce capacity across competing demands. Their value depends on the completeness of the records, the quality of the method and a user's ability to challenge the result.
That's not nothing. That's a meaningful chunk of the administrative load that eats a project manager's week.
Where the project manager still carries the work
Breathless predictions tend to treat the project plan as the project. The harder part is often relational work.
Reading a team means noticing when someone who is normally sharp has become withdrawn in stand-ups. It means testing whether the developer saying "it's fine" has confidence in the plan or needs help they do not want to request in front of the client. A dashboard may contribute a signal. The project manager has to create the conditions for a useful conversation and act on what they learn.
Then comes doing something useful with what you've learned. That requires empathy, judgement and a read on the politics of the situation that the firm should not assume a model can supply.
AI can help a team interpret project information. The project manager remains responsible for the relationships, decisions and interventions that move the work.
So: friend or foe?
Friend, I think, when you're clear about what you're using it for.
The useful promise is capacity. If reporting, resource modelling and early-warning analysis become faster without reducing their quality, the project manager gains hours to spend with the people doing the work.
And that matters more than it sounds. Nothing slows a project down like a project manager who's too busy producing status reports to notice that half the team is stuck. If AI takes that admin away and gives those hours back, it isn't a threat to the role. It's a promotion.
Firms get this wrong when they treat AI only as a headcount question: how many project managers can we avoid hiring? The capacity question is more revealing: which work should improve when administrative effort falls?
Choose tasks by consequence and evidence
Do not approve “AI for project management” as one use case. Break the role into tasks. Drafting a weekly status summary, comparing risks across reports, forecasting resource demand and recommending a response to a distressed stakeholder have different inputs and consequences.
For each task, ask four questions:
- What information does the system receive, and how complete is it?
- What action or decision can its output influence?
- How would the team detect a plausible but wrong result?
- Who reviews the output and remains accountable for acting on it?
A low-consequence drafting task may need sample review and a clear owner. A recommendation that changes budget, staffing or a client commitment deserves stronger testing and human approval. The same tool can sit in both categories; governance should follow the task rather than the product name.
Test against historic project data before relying on a live forecast. Compare the system's flags with what happened, look for important events it missed and identify whether gaps in the source records made the result unreliable. Then run it alongside the current process for a defined period. Record false alarms as well as missed risks, because a stream of weak warnings will eventually be ignored.
Decide where the saved time goes
An efficiency claim is incomplete until someone names the work that replaces the administration. If a weekly report takes two hours less, decide whether that time goes to dependency management, stakeholder conversations, coaching the team or another project. Measure whether the chosen work happens.
This is also a quality question. A faster first draft can create more time for challenge, or it can simply bring the deadline forward. Agree the expected improvement before deployment: earlier risk intervention, fewer unresolved decisions, more reliable forecasting or better sponsor understanding. Hours saved are useful; project performance is the reason to care.
The data caveat
None of this works reliably if the underlying project data is a mess. AI applied to inconsistent, half-maintained records can produce confident nonsense. If plans live in six different tools and nobody trusts the numbers, define the source of truth, ownership and update rhythm first. The clever work depends on the boring work.
Put one question to your delivery leads: if AI returned meaningful time to project managers, what higher-value work would they do with it, and how would you know it happened? If the answer is “more reporting”, the operating model needs attention before the tool.
Do not fund the pilot until the team can name that higher-value work. The answer defines the benefit, the measures and whether AI belongs in this part of project delivery at all.
