AI in project management: what it can do today and what it cannot

AI in project management is most useful today as a drafting and pattern-spotting assistant. It can summarize status from tickets and updates, draft RAID log entries from meeting notes, produce minutes and action lists, and flag schedule slippage in plan data. It cannot be accountable for a decision, it is only as good as the project data it reads, and it can state wrong things confidently. Treat its output as a draft that a named project manager reviews, and govern it like any other tool that touches project records.

What can AI do in project management today?

Four uses are practical with current general-purpose assistants and the AI features built into many work-management tools:

1. Status summaries. Given ticket updates, a changelog and last week's report, a model can draft this week's status in your template. The saving is in assembly, not judgment: the project manager still sets the RAG status and decides what to escalate.
2. RAID log drafting. From meeting notes or a transcript, a model can propose candidate risks, assumptions, issues and dependencies with suggested owners. A person accepts, merges or rejects each one. RAID logs and steering decisions shows what a finished entry should contain.
3. Meeting notes and actions. Transcription plus summarization produces minutes, decisions and action items quickly. Errors tend to cluster around who agreed to what, which is the part that matters most.
4. Schedule risk detection. Pointed at plan data, AI can flag tasks whose remaining work exceeds the time left, dependencies whose predecessors slipped and people booked twice. It finds candidates; critical path analysis decides whether they move the date.

PMI takes a similarly practical line. Its hands-on generative AI course for project managers covers workflows such as drafting meeting summaries and linking requirements to test cases. PMI: Practical Application of Generative AI for Project Managers

Which project tasks suit AI, and which should stay human?

The column that matters is "Control". A task can move from "needs review" toward "good fit" once your own sample checks show that review rarely changes the output.

Where does AI fall short in project management?

Accountability. A model can recommend delaying a launch; it cannot own the consequences. Decision records should name a person, with any AI output attached as input rather than as the decision.

Data quality. AI summarizes whatever the tools contain. If tickets are stale, the plan has not been re-baselined or decisions live in private chats, the summary will be fluent and wrong. Fix the record before automating the report.

Confabulation. NIST's Generative AI Profile defines confabulation as the production of confidently stated but erroneous or false content. NIST AI 600-1: Generative AI Profile In project work it appears as an invented dependency, a misattributed action or a date nobody agreed. AI hallucinations in customer-facing products covers reduction techniques that apply here too.

Missing context. Much project risk is political or informal: a sponsor's shifting priorities, a partner's unstated capacity problem. None of it is in the data.

Confidentiality. Project records hold contract terms, partner roadmaps and sometimes personal data. Check what a tool retains and where it processes data before you connect it.

How should a PMO govern AI use?

The NIST AI Risk Management Framework gives a usable structure. NIST released it in January 2023 as voluntary guidance with four functions, govern, map, measure and manage, with govern designed to run across the other three. NIST: AI Risk Management Framework · NIST: AI RMF release A proportionate PMO version:

Regional rules add to this. In the EU, ask counsel how the AI Act applies before using AI to allocate work to individuals or assess their performance; in the UK and the Gulf, check sector regulator and data protection guidance for your organization. For enterprise-wide controls, see an enterprise AI operating model.

What changes when AI agents act in delivery tools?

An assistant drafts; an agent acts. Agents can update tickets, chase owners, schedule meetings and post updates. PMI now offers a course for project managers on understanding and overseeing AI agents. PMI: Agentic AI for Project Managers

Three controls matter before an agent gets write access:

• Scoped permissions. Read broadly, write narrowly. An agent that can reassign tickets can reshape a plan without anyone deciding to.
• A log of every action. Each change should show that the agent made it, on whose behalf and from which instruction.
• Human approval for external messages. Anything addressed to a partner, customer or regulator waits for a named approver.

How do you start using AI in a project team?

Pick one frequent, checkable task, usually the weekly status report. Measure how long it takes today, run the AI-assisted version alongside it for a few cycles and record every correction the project manager makes. Expand only when corrections are rare and minor. How to choose the first AI use case describes the same approach for operations teams.

FAQ

Will AI replace project managers?

Not the accountable part of the role. AI cuts the time spent assembling reports and notes, leaving more time for decisions, negotiation and risk, which still need a person.

What is the best first AI use case for a PMO?

Drafting the weekly status report from existing tools. It is frequent, low risk and easy to check against source records.

Can AI predict project delays?

It can flag warning signs in plan and ticket data, such as slipped predecessors or overbooked people. Whether a flag moves the delivery date still needs schedule analysis and a conversation with the team.

Is there a PMI certification for AI in project management?

PMI offers the PMI Certified Professional in Managing AI (PMI-CPMAI) and several free AI courses for project managers. PMI: AI courses

How do you stop AI from making up project facts?

Ground it in your project records, require a link to the source ticket or document for each claim, and have the owner review every output before it is shared.