In one Reddit discussion, a commenter said AI saved time on some administrative work while the critical thinking and judgment calls remained theirs (“Learning AI as a Project Manager,” date unknown). This is a personal report, not a measured before-and-after result. To assess AI in project management workload, measure the complete recurring task, not just the speed of generating a draft. Include context, review, corrections, and coordination with what happens next.
Start with a routine PM task
Choose one recurring activity with a clear start and finish. Meeting summaries or a first-pass status report are useful candidates because the output can be reviewed before it is shared. Keep the test narrow: one task, one AI workflow, and a defined period or number of runs. Avoid changing the task, tool, and review process all at once, or it will be hard to tell which change affected the result.
Before adding AI, write down what counts as finished. For a summary, that might mean a draft is ready for the intended audience. For a report, it might mean the update has been checked and is ready to send. The exact endpoint depends on the task. What matters is using the same endpoint in the baseline and AI-assisted versions.
Also note the information needed to do the work. If the AI requires project history, decisions, or status details, gathering and preparing that context is part of the workflow. Do not treat it as free simply because it happens before the prompt is sent.
Map the work before and after AI
Record the baseline for the task as it works today. Then use the same task definition to map the AI-assisted version. A simple log can have one row per run and separate columns for each kind of effort:
| Work stage | What to record |
|---|---|
| Prepare | Time spent collecting and supplying project context |
| Generate | Time spent prompting, waiting, and refining the request |
| Review | Time spent checking accuracy, completeness, tone, and relevance |
| Correct | Time spent fixing omissions, errors, or unsuitable wording |
| Coordinate | Time spent clarifying ownership, getting input, and following up |
| Finish | Total time until the agreed output is ready for use |
Use the same units and definitions for both versions. If several people contribute, record whose time is included. If a task stops and resumes later, capture the effort across the full cycle rather than only the time spent actively typing.
The comparison is straightforward: baseline effort minus AI-assisted effort. For the AI-assisted total, add context preparation, prompting, review, corrections, and coordination. A positive difference means the measured task took less effort in that run; a negative difference means it took more. This is a measurement rule for your workflow, not a promised result.
Include the work that appears around the output
A useful way to spot hidden effort is to look at what the PM must do between generation and delivery. Ravi Kalluri’s October 2026 PM World Journal paper calls the integration burden a “coordination tax.” It describes four components: trust calibration, translation, mediation, and metacognitive work. The paper proposes a framework and says the burden’s precise magnitude is not yet established, so it should guide what to observe rather than serve as a ready-made estimate (Kalluri, The Coordination Tax: Who Pays When AI Joins the Team).
For a PM’s time log, those components suggest practical questions. Trust calibration: how long did it take to decide whether the output was reliable enough for this task? Translation: did you need to turn the output into language or a format the team could use? Mediation: did the recommendation conflict with team judgment or project context, and who resolved that conflict? Metacognitive work: how much effort went into deciding what to delegate, what to check, and when to stop relying on the output?
These are prompts for observation, not separate charges to add twice. If checking an output and resolving a disagreement happen in the same review window, record the time once and note both activities. The purpose is to see where the work goes, not to make the log more elaborate than the task.
Make review ownership explicit
Before using AI on a recurring task, name the person responsible for checking the result. Decide what that person must verify, what kinds of uncertainty require escalation, and who can approve the finished output. The review can be light or thorough depending on the task’s consequences, but it should not be left as an unspoken assumption that somebody else will catch a problem.
The practitioner discussion describes keeping judgment calls with the PM even when AI helps with administrative work (Reddit discussion). Turn that distinction into a working rule: AI may prepare an output, while a named person remains accountable for deciding whether it is fit for its intended use. If the reviewer cannot verify a material point from the available context, record that as unresolved and route it to someone who can.
Track the number of review rounds and the time spent on corrections. Also note the type of issue found, such as missing context or a detail that needed confirmation. These are local measures. They do not prove that one AI system is reliable in general, but they help a team decide whether its chosen workflow is working as intended.
Decide what to monitor before scaling
Look at the results across multiple runs rather than treating one unusually smooth or difficult task as the answer. Compare total effort, review time, correction time, and coordination time. Keep the task definition and completion point consistent. If the workflow changes, record what changed so the comparison remains interpretable.
Do not treat output volume as proof of time saved. A faster draft can still require more review or generate extra follow-up. Tempo’s commissioned 2026 survey of 300 senior project, portfolio, and PMO leaders in North America and Western Europe reported that 46% of teams were duplicating work with AI; 43% identified quality and oversight of AI output as a top challenge. The figures are self-reported and do not establish that AI caused the problems, but they underline why teams should measure duplication and review effort rather than assume the draft is the whole job (Tempo, The 2026 State of AI in Portfolio Management Report, fieldwork June 2026).
Set a decision rule before reviewing the log. For example, decide what amount of net time reduction would make the workflow worth keeping, and what kinds of unresolved issues would mean it needs redesign. A workflow can still be useful if it does not reduce total minutes, but be clear about what value you are trying to get. Do not call it a time saving unless the complete task takes less time under your chosen measure.
A small test is better than an assumption
The evidence available here does not establish a universal amount of PM time saved or added by AI. Kalluri’s paper presents coordination tax as a proposed framework, not a precise workload estimate. The Reddit account is anecdotal. Tempo’s survey describes self-reported challenges among its sampled leaders, rather than a before-and-after time study. Use those sources to shape the questions, not to predict the result for your team.
Start with one recurring task. Record its current end-to-end effort, then track context, prompting, review, correction, and follow-up when AI is used. Keep the review owner clear and inspect what the log shows before expanding the workflow.
The useful measure is not how quickly AI produced text. It is what the complete task demanded from the people responsible for using it. Choose one recurring PM task and track both the time AI saves and the review it requires.
Key takeaway
Measure the whole workflow, not just how quickly AI produces a draft. Include context preparation, review, corrections, and coordination, and keep a named person accountable for the final output.

Andrii founded It’s a Plan and leads its product and roadmap, from the open-source edition to Pro for teams. He works with the project’s contributors and the teams that run It’s a Plan.
All posts by Andrii Poluosmak