AI is changing project management skills, but are we focusing on the right ones?

As Artificial Intelligence (AI) begins to draft plans, summarise meetings and automate workflows, the project profession may be asking the wrong skills question. Technical fluency matters, but it is only the starting point. The real differentiators are the human capabilities that determine whether recommendations can be trusted, challenged, supported and responsibly owned.

AI literacy is only the starting point

By 2030 The World Economic Forum predicts that 39% of all workers' skills will evolve, and analytical thinking is still the most sought-after skill in the eyes of employers. That tension matters. AI can create plans, manage risks and generate initial drafts more quickly than most teams can. The more difficult questions are not answered by speed: Is the output relevant? What is the context that is not provided? Who will trust it? Who is responsible if it's incorrect? The question isn't whether project professionals can leverage AI, it's how they can make what AI generates decisions they can trust and act on.

Are we solving the wrong skills problem?

Much of the discussion about AI upskilling begins with prompting, copilots, automation platforms and data literacy. Those skills are important, but they don't make professionals ready for delivery. Projects do not often fail due to failure to create another document. They stall when stakeholders disagree, assumptions are not challenged, users lose faith or when the original plan doesn't match reality. PMI reports that 91% of executives consider power skills as critical success factors and that 63% fewer project failures are associated with a greater focus on power skills. Influencing, empathising and adapting are thus delivery skills and not a luxury. The danger is that organisations think they know how to use the technology, but they don't know how to lead the change that it brings.

What AI changes, and what it does not

AI is changing the pace of project work. It can turn meeting notes into actions, compare options and produce a first-cut plan in seconds. Microsoft reports that 81% of leaders expect AI agents to become moderately or extensively integrated into organisational strategy within 12 to 18 months. It is not a question of whether project professionals will work with these systems, but how responsibly they will do so with what they produce. A recommendation might seem complete, but it may be based on weak evidence, or lack local context or consideration of its impact on users. There could be various routes through technology but the organisation needs to determine what compromises are acceptable, who has the authority to question the recommendation, and who will have the responsibility for the outcomes. Adoption needs to have points of review, intervention and ownership of decision.

Where human capability becomes the differentiator

An answer that appears convincing and the situation is not clear, reveals the capabilities that are most important. Now, question the output by using judgement. What evidence is present to support it, which assumptions are weak, what is missing from the context? McKinsey found that just 27% of the respondents who are using generative AI reported that all the content generated is reviewed by their employees before use, giving a sense of how fast a well-crafted response can get through without anyone stopping to verify it.

Second, influence and empathise the people. Finance, operations, legal and user and sponsor teams may have different views on the same risk; moving forward requires discovering where those interests lie, articulating trade-offs, and understanding where change causes anxiety, exclusion, hidden effort or lost trust.

Third, adjust the approach when policies, requirements or user behaviour change, which is not what the original assumptions were. Lastly, take ownership of the consequence by having a named decision owner, a documented reason and a path to challenge and clear escalation. NIST centres trustworthy use, while the EU AI Act requires traceability and appropriate human oversight for high-risk systems. These capabilities are not alternatives to technical skill; they are what make it useful in practice.

What live delivery has taught me

The early indicators of success were obvious in one automation project with a large volume of cases: cases moved through the system quicker and there was less manual handling. The more difficult one was what would happen when the standard road was not available. Those cases were most often due to missing information, conflicting rules, or customer impact or a decision that no one was clearly responsible for. We changed our thinking from exceptions being a defect in the main process to a delivery problem. We established when work should stop, who was empowered to intervene and what evidence should be documented and how more frequent exceptions should enhance the design.

The lesson is that automation is not about taking away human capability, it's about redistributing it. Routine work is processed in less time, and more judgement is focused in unusual and consequential cases. A field study of 5,172 customer-support agents reached a similar conclusion: assistance increased productivity by 15% on average, with the greatest gains among less-experienced workers. The value came from strengthening people’s performance, not making professional responsibility disappear.

How organisations should redesign upskilling

Organisations need to cease and desist from using the number of courses completed or the number of tools accessed or the ability to create a finished product as indicators of readiness. A more robust method is that of realistic decision practice, where there is partial evidence, conflicting parties and a need to assign responsibility. When the normal path doesn't exist, project professionals should provide an explanation of what they would trust, what they would challenge, who has the authority to decide and where the ownership lies.

Upskilling should also include exception ownership (who sees unusual cases, who has the ability to overrule a recommendation, what evidence is needed to keep, how do recurring issues come back into the design). Challenge reviews should verify if assumptions are still valid, if responsibilities are still clear and if anyone has been overlooked who is affected by the decision. While technical knowledge can enable people to utilise new tools, it is judgement, influence and accountability that will ensure that decisions are comprehended, embraced and executed responsibly.

From AI-capable to judgement-ready

The project professional of the future will not be distinguished by how quickly they can produce an answer, but by how well they can question it, explain it, adapt it and take ownership of what follows. Technical capability will remain essential, but it will not be enough. The real measure of readiness is whether professionals can turn information into decisions that are understood, trusted and responsibly owned. The goal is not simply to become AI-capable, but judgement-ready.