Action logs: turning decisions into owned, dated work

By the end, you can

• Define what makes a meeting output a genuine action-log entry — a specific, checkable action, a named owner and a due date — and distinguish it from a decision or an unresolved issue that hasn't yet become someone's committed work (PM-3).
• Use AI to draft a first-pass action log from a set of meeting notes, and explain why a named owner must confirm the action and the date before it is treated as committed (PM-3).
• Identify an action-log entry missing an owner, a date or a clear description, and correct it (PM-3).

Before you start

This is Module 4, Lesson 2 of the AI-Powered Project Management course. It builds directly on Lesson 1's meeting record — specifically the decisions and unknowns a properly separated set of notes surfaces — and turns those into an action log. It assumes you have completed Lesson 1 and have used at least one AI chat assistant for drafting text.

An action log is not the same as a list of things that were discussed

A meeting can produce a genuine decision — Lesson 1's "we will add a penalty clause to the next contract" — without anyone yet turning it into work a specific person is doing by a specific date. An action log is what closes that gap: for each entry, a specific action, a named owner and a due date. "Look into the penalty clause" is not yet an action-log entry; it is closer to a decision waiting to become one.

This course has already met a close relative of the action log. APM's project management glossary defines an issue log as "a log of all issues raised during a project or programme, showing details of each issue, its evaluation, what decisions were made and its current status." An issue log tracks problems and how they were resolved. An action log is broader and more immediate: it tracks the small, committed next steps a meeting produces — some tied to issues, many not — and each entry is meant to close out once the owner completes it.

What makes an action genuinely tracked, not just noted

A vague action is barely more useful than no action at all. APM's glossary defines an activity, in part, as "the smallest self-contained unit of work in a project," and defines a task more precisely as "the smallest indivisible part of an activity when it is broken down to a level best understood and performed by a specific person or organisation." An action-log entry needs to meet that same bar: small and specific enough that one named person could actually pick it up and know what "done" looks like, not a broad area of responsibility dressed up as a next step.

"Sort out the client email" fails that bar — it names no specific outcome and could mean several different things to different readers. "Draft a reply to the client's delivery-delay question, for the account lead to review by Thursday" passes it: one person, one checkable outcome, one date. Without a named owner, an action belongs to whoever remembers it existed, which in practice means it often belongs to nobody. Without a date, "soon" quietly becomes "eventually," and an action with neither is not really being tracked — it is just something that got said out loud once.

AI can draft the log; the named owner confirms they've actually accepted it

Given a set of meeting notes — ideally already checked against Lesson 1's fact/interpretation/decision/unknown split — an AI assistant can draft a first-pass action log: scanning the notes for anything that sounds like a next step, and proposing an owner and a date for each. That is a real time-saver; combing through a page of notes for every implied commitment is easy to do badly under time pressure.

It is not the same as knowing who actually agreed to do the work, or by when. The US National Institute of Standards and Technology's (NIST) AI Risk Management Framework describes an AI system as something that "can, for a given set of objectives, generate outputs such as predictions, recommendations, or decisions" — a generator of a plausible next step and a plausible owner, not a party who was in the room when someone actually said "I'll take that." A drafted action log can also make the opposite mistake the Open Worldwide Application Security Project's (OWASP) guidance on misinformation warns about: it "occurs when users place excessive trust in LLM-generated content, failing to verify its accuracy," and here that risk shows up as a floated idea — "maybe someone should look at the storage costs at some point" — turning into a firm-sounding logged action with an invented owner and date, simply because the draft needed to put something in those fields.

Confirming that the named owner has actually accepted the action, and that the date is one they can genuinely meet, is itself the kind of call OWASP's guidance on excessive agency has in mind when it recommends that a human "approve high-impact actions before they are taken." An action log that goes out uncorrected commits a real person's time on their behalf without asking them; checking it belongs with the meeting's organiser or the named owner, not with the draft that first proposed it.

A worked example: a university alumni association's reunion committee

A university alumni association's five-person reunion committee meets to plan a 20-year reunion event. The committee chair types rough notes during the meeting, then asks an AI assistant to draft an action log from them.

The draft includes: "Book the venue — [owner not specified] — by next month," "Someone should look into catering options," and "Follow up with the caterer from last year's event — treasurer — by Friday." The chair checks each entry. The venue-booking action has no owner because two committee members were still negotiating who would take it when the meeting ended — she follows up separately and adds "venue coordinator" once he confirms he'll take it, with a firm date instead of "next month." "Someone should look into catering options" was one member thinking aloud, not a commitment anyone made — the chair leaves it out of the action log entirely and instead adds it to next meeting's agenda as an open question. The caterer follow-up is accurate: the treasurer did agree to this, though she privately knows "by Friday" is unrealistic given the treasurer's other commitments that week, so she checks with him directly and moves the date to the following Wednesday before the log goes out to the committee.

Accessibility notes

This lesson is text-first, with no images, audio, video or downloadable artifacts. The practice exercise's model answer sits behind a native disclosure control that is reachable and operable by keyboard and correctly announced by screen readers. The knowledge check uses native radio-button inputs with a visible question and options, and posts its result to a live status region so assistive technology announces the outcome without a page reload.

Sources and limits

This lesson synthesises the sources below into a practical learning model. It is not a security standard, legal advice or a guarantee that any particular agent design is safe.

1. Project management glossary — APM (Association for Project Management). Defines an activity as the smallest self-contained unit of work in a project, and a task as the smallest indivisible part of an activity, broken down to a level best performed by one specific person.
2. Project management glossary — APM (Association for Project Management). Defines an issue log as a log of all issues raised during a project or programme, showing details of each issue, its evaluation, what decisions were made and its current status.
3. AI Risk Management Framework 1.0 — NIST AI Resource Center. Frames an AI system as an engineered system that generates outputs such as predictions, recommendations or decisions — not a self-directing decision-maker.
4. LLM09:2025 Misinformation — OWASP Gen AI Security Project. Defines overreliance as placing excessive trust in AI-generated content without verifying its accuracy.
5. LLM06:2025 Excessive Agency — OWASP Gen AI Security Project. Recommends human-in-the-loop control requiring a person to approve high-impact actions before an LLM-connected system takes them.