AI Assistants for Project Managers: Ten Tasks Worth Delegating, and Four You Should Not

A project manager once told an AI assistant to "make the schedule shorter." It did what any eager intern would do: it removed the permit approval and the inspection. The new schedule was beautiful, 40 percent shorter and quite impossible. The assistant was not stupid. It was obedient, which is a different problem and a more common one.

I tell this story, which is an illustration and not a report about a real person, because it contains everything useful about AI assistants in project work. They are fast, eager and fluent. They do not know your project unless you tell them, they do not know what they do not know, and they will say wrong things in the same confident voice they use for right ones. Used with that in mind, they can give a project manager back several hours a week. Used without it, they produce beautiful, impossible schedules.

This guide is practical. It lists ten tasks that are worth handing to an assistant, shows how to ask well, names the four jobs that should stay with a human and offers a simple test, in the middle of the page, for any task you are tempted to delegate. It is not about one product. Tools change every few months, and the habits below do not.

The intern model

The most helpful way I have found to think about an AI assistant is as an intern with unusual traits. The intern has read an enormous amount. It writes quickly and fluently. It will try anything you ask. It has no memory of your project, your client or last week's meeting unless you hand it the material. It cannot tell when it is guessing, and it would rather give you a confident answer than say "I do not know."

Now think about how you manage a good intern. You give a clear task, with context and an example of what good looks like. You do not ask them to make decisions that belong to you. You review their work before it goes to the client, more closely at first and more lightly as trust builds. And you never hand them the company's most sensitive files on their first day.

That is the whole method. Everything else in this guide is detail.

The wider picture agrees with the caution. In June 2026 PMI published a standard on the use of AI in portfolio, program and project management, and its Global Summit in October put the topic first on the agenda, as I summarized in the PMI Global Summit news post. A Gartner survey, covered in my Gartner news post, found that only about 13 percent of CIOs report significant value from AI, and my reading is that the missing ingredient is project discipline. Tools matter less than the way people run them.

Ten tasks worth delegating

These are the jobs where, in my experience, an assistant saves the most time with the least risk. For each one, I note the human step, because there is always one.

1. Meeting notes into summaries and actions. Give it the transcript or your rough notes and ask for decisions, action items with owners and dates, and open questions. The human step: check every owner and date against what you remember, since this is the part assistants get wrong most often.
2. Status report first drafts. Paste your bullet points and the report format and ask for a one-page draft. The human step: check the colors against your written rules and make sure the decision box says what you actually need. My status report guide has the template.
3. Stakeholder messages. Turn rough thoughts into a clear email, or the same message into three versions for the sponsor, the team and the client. The human step: read it aloud. If it does not sound like you, rewrite the opening line.
4. Risk brainstorming. Describe the project and ask for twenty things that could go wrong, grouped by category. It is a cheap way to run a pre-mortem. The human step: discard what does not apply, then score and own what remains in your risk register.
5. First drafts of charters, scope statements and plans. Give it your notes and a template and ask for a draft. The human step: everything in a charter is a decision. The assistant drafts, you decide, and the sponsor signs.
6. Schedule sanity checks. Paste a task list and ask what is missing, which dependencies look wrong and which durations look unrealistic. The human step: it can only comment on what you show it. Think of it as a second pair of eyes, never the author of the critical path.
7. Agendas and meeting prep. Describe the goal and the attendees and ask for a sixty-minute agenda with time boxes and decisions needed. The human step: remove half of it. Meetings with a full agenda run late.
8. Summaries of long documents. Drop in a long report or a thread of 80 emails and ask for the five points that matter and the questions it leaves open. The human step: for anything with money or legal weight, read the source. A summary tells you where to look, not what to sign.
9. Spreadsheet cleanup and formulas. Messy data into a clean table, a formula you cannot remember, a pivot explained step by step. The human step: spot-check ten rows against the original before you trust the whole column.
10. Lessons-learned patterns. Paste the comments from a retrospective and ask it to cluster them into themes and propose three changes. The human step: the team decides which changes to adopt. The method is in lessons learned from a closeout.

Notice the pattern: in every item the assistant produces a first draft and a person owns the decision. That is not a limitation to work around. It is the design.

How to ask: five parts

Most disappointing AI results come from thin requests. "Write a risk register" will get you a generic one. A request with five parts gets you something you can use.

1. Role. Who should the assistant act as? "You are an experienced construction project manager helping a small business owner."
2. Context. What is the situation? Two or three sentences with the real facts: the project, the date, the team, the constraint that matters most.
3. Task. What exactly do you want? "List the twelve most likely schedule risks for this build-out and explain each in one sentence."
4. Format. What should the answer look like? "A table with risk, cause, early warning sign and a first action."
5. Limits. What should it avoid? "Do not invent numbers. If you are not sure, say so. Ask me questions if you need more information."

That last part is the secret ingredient. Asking the assistant to question you first changes the exchange from a vending machine into a conversation, and the questions it asks often show you what you have not thought about.

A small example, written as I would type it: "You are an experienced project manager. I am running a small restaurant build-out opening on March 1. Here are my notes from today's meeting. Produce a summary with decisions, action items (owner and date) and open questions, in under 200 words. If an owner or a date is not in my notes, write 'unknown' instead of guessing." That last sentence prevents the most common failure of the first task on the list.

Keep what works

When a prompt works, save it. A small library of five or six requests, kept in a document and refined each time, is worth more than any clever trick. It turns a skill into a team habit, and it makes the work consistent when someone else takes over. If you like systems, treat your prompts like templates: a name, a purpose, the text and a note on what to check afterward.

Four jobs that stay human

An assistant can help you think about these. It should not do them for you.

• Deciding priorities. Which project comes first, which risk matters most and which scope item goes are choices about values and consequences, and they belong to the people who own the outcome.
• Delivering bad news. A delay, a cost overrun or a difficult piece of feedback should come from a person, face to face when possible. An assistant can help you rehearse and draft; it should never be the messenger.
• Judging people. Performance, fit and trust are human judgments with human consequences. Be especially careful about anything that sounds like an automated assessment of a person.
• Anything where an error is dangerous or illegal. Safety, contracts, compliance, regulated advice and anything that could harm someone needs a qualified human to own it. The assistant can prepare questions for the expert. It cannot replace the expert.

These are not rules from a textbook, they are the places where the cost of being wrong is not a bad paragraph but a real problem for a real person.

Confidentiality, accuracy and ownership

Three cautions deserve a section of their own, because they are where most real trouble starts.

Confidentiality. Before you paste anything into a tool, ask where it goes. Does your company allow this tool for this kind of information? Does the tool keep what you type, or use it to train models? When in doubt, remove names, amounts and anything a competitor would enjoy reading, or use a tool your company has approved. Write the answer into a one-page policy: a template for one is on the way in the AI series, and the first part, where AI helps and where it does not, covers the basics.

Accuracy. Assistants can state invented facts, dates, citations and numbers in a calm voice. This is called hallucination, and it is not a rare event. Treat every fact as unverified until you have checked it against a source you trust. Ask the assistant to say where each claim comes from, and then check that source exists.

Ownership. You are accountable for what you send, whoever or whatever wrote the first draft. "The AI said so" is not a defense in front of a sponsor, a client or a court. Read everything. Sign what you can defend.

Are you ready? A test for any task

Before you delegate a task, run it through the checklist below. Tick every statement that is true. The more boxes you tick, the safer the task is to hand over. Use it on the ten tasks above and on your own list, and be strict about the confidentiality and review boxes.

Interactive tool

Is this task safe to hand to an AI assistant?

Loading the tool…

Measuring whether it helped

Delegating is only worth it if you save time or raise quality. Measure it for two weeks. Before you start, write down how long the task takes you today, using a timer, not a guess. After each use, note the time you spent writing the request, waiting and checking. Compare. For the first task on the list, many people find the saving is real but smaller than they hoped, because checking takes longer than expected. That is fine. The target is a net gain with no loss in quality.

Watch for the opposite trap too. If you spend 40 minutes tuning a request to save 15 minutes of writing, you have invented a hobby. Delegate tasks that repeat, and invest in the request once. If you want a method for calculating payback on automation, see what to automate first, and the weekly rhythm in the weekly review template is a good place to log what you tried.

A starter plan for the first month

You do not need a grand rollout. A month is enough to build a habit.

1. Week 1: one task. Pick meeting summaries. Use the five-part request. Check every owner and date.
2. Week 2: add the status report. Draft it from your notes, apply your color rules yourself and compare the saving with your timer.
3. Week 3: add a risk brainstorm. Run one before your next planning session and bring the results to the team.
4. Week 4: write the rules. One page: what the team may paste, what it may not, who reviews outputs and where prompts are kept.

At the end of the month, tell your team what worked and what did not. People trust a tool more when they have seen its failures.

Common mistakes

• Delegating a decision. Drafting is delegable. Deciding is not.
• Skipping the review. The summary looked fine, so it went out, and it said the sponsor approved a budget she never saw.
• Pasting confidential material into an unapproved tool. One careless paste can become a contract problem.
• Thin requests. No context, no format, no limits, and then disappointment.
• Believing numbers and citations. Check them.
• Automating a messy process. If your meetings have no owners, an assistant will summarize chaos faster.
• Never saving what works. Each person reinvents the same prompt every Monday.

What to do this week

Choose one task from the ten, probably meeting summaries. Write the five-part request once and save it. Time the task by hand before and with the assistant after, for the same type of meeting, five times. Check every owner and date. At the end of the week, decide whether it saved time and whether the quality held, and write down the rule you will keep.

If you would like help choosing the first workflows to automate, writing a one-page AI policy or running a project with AI and discipline together, that is what I do with Project Focus and the AI and automation service, and you can tell me about your situation on the contact page. The publishing calendar shows the next parts of the AI series, and the newsletter sends a monthly roundup.

Which task would you hand to an assistant first, and which one would you never hand over? Tell me in the comments.

More tools. The weekly review, the risk register and the decision log are free in the toolkit.

This guide is general information, not legal or professional advice. Check your organization's rules on data and AI before you use any tool with company or client information.

Frequently asked questions

What can an AI assistant do for a project manager?

It can draft meeting summaries and action lists, first versions of status reports, stakeholder emails, risk lists, charters and agendas, check a task list for missing steps, summarize long documents, clean spreadsheets and cluster lessons-learned comments. A person should review all of it before it is used.

Is it safe to put project information into an AI assistant?

Only if the tool is approved for that kind of information. Check whether your organization allows it, whether the tool stores or trains on your inputs and whether your contracts restrict sharing. When in doubt, remove names, amounts and anything confidential, or use an approved tool.

Will AI replace project managers?

It is changing the routine part of the job, such as drafting, summarizing and checking, while the human parts remain: setting priorities, negotiating, delivering bad news, judging people and taking responsibility. Project managers who learn to supervise these tools well will have an advantage.

How do I write a good prompt for project work?

Use five parts: the role, the context, the task, the format and the limits. End by asking the assistant to question you if something is unclear, and to write "unknown" instead of guessing. Save the prompts that work so you can reuse them.

Join the conversation

Did this help? A quick reaction tells me what to write next.

Useful Made me smile Go deeper on this

Be the first to react.

Comments

Loading comments…

Leave a comment

NameEmail (required, never shown)

Which task would you hand to an AI assistant first, and which one would you never hand over? Post comment

Your name and email create a small profile that shows you are a real person. The email is never shown. The first time, I send you a confirmation link, and every comment is reviewed before it appears. Please keep it respectful and leave links out.