PMI Just Published a Standard on AI in Project Management. Here’s What Actually Matters for Practitioners.
PMI released its first edition of The Standard for Artificial Intelligence in Portfolio, Program, and Project Management in 2026, and the timing is not incidental. Organizations are deploying AI into programs and portfolios faster than most governance structures can absorb it. The standard gives practitioners a principled framework, rather than a tool-specific playbook, for navigating that reality.
I spent time working through it with my own practice in mind. What follows are the ideas that I think actually change how a thoughtful project manager or TPM should be operating right now.
AI has to start with strategic value, not tool adoption.
Many organizations are approaching AI from the wrong direction. They start with a capability, a vendor demo, or an executive mandate, then work backward to justify the business case. A serious AI initiative should begin with the value it is expected to produce, the organizational objective it supports, and the measurable outcomes that will determine whether the effort was worth doing.
For project managers and TPMs, this means the business case for AI cannot be reduced to efficiency language. “Faster,” “automated,” and “AI-enabled” are not value statements. The value has to be tied to something the organization actually cares about: better decision quality, lower operational risk, more resilient delivery, stronger compliance posture, or some other durable business outcome.
This also means AI initiatives need active value reassessment throughout delivery. A use case that seemed compelling at initiation may become less so after discovery, data assessment, or stakeholder review. A project manager who treats the original charter as fixed in that environment may preserve the plan while losing the value.
AI is both a tool and a deliverable, and conflating the two causes real damage.
The standard draws a sharp distinction between AI used to manage projects and AI that is the thing being delivered. AI used for scheduling, resource allocation, or reporting is one category. AI delivered as a product or capability, such as a fraud detection model, recommendation engine, or NLP system, is another. These categories require different governance.
A TPM standing up a program that both uses AI to manage delivery and delivers AI as an output needs to hold two governance tracks simultaneously: one governing AI as part of the delivery operating model, and another governing AI as the product being shipped. A team may have a policy for using generative AI internally but no mature operating model for validating, monitoring, or decommissioning an AI system once it becomes part of a business workflow. That gap is where accountability gets blurry and where programs become overconfident in controls designed for the wrong use case.
The Human-in-the-Loop concept is a governance mechanism, not a cultural posture.
The standard treats Human-in-the-Loop as a structural requirement rather than a mindset reminder. A project manager needs to define, during planning, which decisions are AI-assisted versus AI-determined, which require human review, who owns the review, what authority that person has, and what happens when the human reviewer disagrees with the output. That accountability should be visible in the RACI, the escalation path, and the operating cadence.
A weak implementation says, “A person will review it.” A strong implementation says, “This category of AI output cannot move forward without review by this role, using these criteria, within this time window, with this override authority, and with this documentation requirement.” That level of specificity matters because AI has a way of quietly shifting responsibility away from people. I have seen programs where nobody asked the accountability question until something went wrong. By then, “the model recommended it” had become the answer to every escalation.
Data quality is a first-class project management concern, and treating it as a technical prerequisite is a governance failure.
Historically, many project managers have treated data quality as a technical concern to be resolved upstream by data engineers or platform teams. For AI initiatives, that posture is not sufficient. Poor data can distort outputs, reinforce bias, produce unreliable recommendations, create compliance exposure, and erode stakeholder trust. If an AI initiative begins with incomplete, stale, or inconsistent data, the project may already be failing before delivery execution begins.
A PM or TPM leading an AI initiative needs to ensure that data governance is operating with enough rigor to support the intended use case. That means asking early questions about data ownership, lineage, access control, privacy, and fitness for purpose, and making data quality visible in status reporting rather than treating it as a quiet technical workstream.
AI changes the definition of done.
A deployed AI system can drift, degrade, or behave differently as data changes, usage patterns shift, or the system is applied in new contexts. A project plan that ends at deployment is incomplete. The work breakdown structure should include monitoring and evaluation. The risk register should include model drift, output degradation, data quality deterioration, and escalation failure. The transition plan should include operational ownership, and the governance model should survive the go-live date.
Governance and compliance are not the same thing.
Governance is directional: it sets the objectives, roles, decision rights, policies, and oversight mechanisms that define how AI will be used and who is responsible. Compliance verifies that the governance structure is actually operating as intended. Organizations commonly draft an AI use policy, run it through legal review, and treat that as governance. But a policy does not answer who owns the AI system across its life cycle, who validates outputs, who investigates anomalies, or who decides when the system should be retrained or retired. Compliance becomes meaningful only after those questions have real answers. A static compliance checklist cannot manage a dynamic AI risk profile.
Stakeholder engagement for AI initiatives is expectation management, not just communication.
Stakeholders confronting an AI initiative may question whether their roles are being automated, whether the system’s decisions can be trusted, whether the data being used is appropriate, and whether accountability will remain human. Those concerns are not resolved with a standard communication plan. AI stakeholder engagement is expectation management around trust, job impact, data rights, decision rights, and acceptable risk.
The audience is not monolithic. People skeptical of AI in general require a different conversation than people supportive of AI but concerned about a specific implementation, who in turn require a different conversation than people directly affected by AI-driven decisions. When people raise concerns about bias, job displacement, or overreliance on automation, those should not be dismissed as change resistance. In many cases, they are legitimate project risks. The project manager’s job is to make those concerns discussable early enough that they can be governed or escalated before they become adoption failures.
The practitioner takeaway
What makes this standard more useful than many formal guidance documents is that it resists the temptation to chase tools. The tooling is moving too quickly for a standard to remain current for long. The structural failures are more durable, and addressing them is where practitioners can build lasting competency.
Organizations will continue to confuse AI as a tool with AI as a deliverable, launch pilots without clear strategic value, treat human oversight as a slogan instead of a control, bury data quality inside technical workstreams, and declare victory at deployment even though AI systems require ongoing monitoring, evaluation, and eventual retirement.
The project manager of the AI era cannot manage only scope, schedule, budget, and status. They have to help manage value, accountability, data quality, human oversight, stakeholder trust, life-cycle governance, quality, and operational readiness. The project plan may still live on the wall, but the real work now reaches into the data, the model life cycle, the governance process, and the human decision chain.
The full guide is available through membership in PMI.
Nabeil Sarhan, MBA, is a dynamic technology delivery manager with over 15 years of experience in tech, cybersecurity, and computing scalability. He excels in leading diverse teams and delivering enterprise-class systems across industries such as healthcare, finance, and retail. Nabeil’s passion for solution design, systems architecture, and performance optimization makes him a sought-after consultant. He holds degrees from Harvard, MIT, and Bryant University. Connect with Nabeil on LinkedIn