PMI Wrote 295 Pages on AI in Projects. Here Is What a Small Team Should Steal.
The project management profession just had a busy few weeks. In June, PMI published The Standard for Artificial Intelligence in Portfolio, Program, and Project Management, the first global standard for applying AI in project work. It runs 295 pages, sets out eight guiding principles and five performance domains, and even maps to the EU AI Act and ISO 42001. This month, PMI is also refreshing the PMP exam to treat AI fluency as core content, not an elective.
Meanwhile the tooling market is moving even faster. Atlassian Ventures just invested in Rocketlane, a platform whose pitch is agentic delivery: AI agents that do implementation work like migrations and configurations, not just track it. Atlassian is not only an investor, it uses the product internally.
If you run a 10 to 50 person company, none of this was written for you. But some of it is worth stealing.
What the standard gets right
Strip away the framework language and PMI’s standard makes one point that applies at any size: AI in project work needs a human who stays accountable. An AI can draft the status report, flag the schedule risk, and summarize the sprint. It cannot own the outcome. Someone still has to decide what the flagged risk means and what to do about it.
That sounds obvious until you watch a small team skip it. The failure pattern we see is not AI doing bad work. It is nobody checking the work because it looks polished. A confident, well formatted, wrong project summary is more dangerous than a messy honest one.
So borrow the principle, skip the ceremony. You do not need five performance domains. You need one rule: every AI generated artifact that leaves your team (client update, estimate, risk log) gets a named human reviewer before it ships.
What the Rocketlane signal means
The Atlassian investment is a signal about where delivery work is heading. The last generation of PM tools tracked work. The next generation claims to do some of it. For services firms and product teams alike, that means the billable or buildable hours are shifting from execution toward judgment, client communication, and scoping.
For an SME the practical question is not “should we buy an agentic platform.” Most of you should not, yet. The question is which parts of your delivery are pure mechanical repetition (data migration, environment setup, templated onboarding), because that is what gets automated first, whether by you or by a competitor quoting half your price.
Make that list this quarter. Even if you automate none of it, you will know where your margin is exposed.
What to actually do
Three moves, in order of effort.
First, write the one page version of an AI policy for project work. Who reviews AI output, what data can go into which tools, and what never gets delegated (estimates and commitments, in our view). PMI members can download the full standard free and skim it for ideas, but your version should fit on a page.
Second, if anyone on your team is PMP certified or working toward it, know that the exam changes this month. AI, sustainability, and stakeholder engagement are moving into core content. Budget a little study time.
Third, audit your delivery process for the repetitive third. That is your automation roadmap, and your pricing risk, for the next two years.
The standard is a milestone for the profession. But small teams have an advantage here: you can adopt the useful 5 percent of it in a week, while larger organizations spend a year forming a committee to evaluate it.