To plan a project with AI, you describe the goal, the deadline, the people and your constraints in plain language, let the AI draft the phases, tasks, owners and dependencies, and then check and correct that draft yourself before anything is assigned. AI is fast at turning a rough idea into a structured plan. It is not good at knowing your team's real capacity or your past mistakes unless you tell it, so the review step is where the plan becomes realistic.

What AI is good at in planning, and what it isn't

AI use at work is now common. In Microsoft's 2024 survey of 31,000 knowledge workers, 75% said they use generative AI at work, and 90% of those users said it saves them time (Microsoft Work Trend Index, 2024).

The best evidence on quality comes from a field experiment with 758 Boston Consulting Group consultants. On realistic consulting tasks within the AI's capabilities, consultants using GPT-4 completed 12.2% more tasks, worked 25.1% faster and produced results rated more than 40% higher in quality. On a task chosen to be outside those capabilities, consultants using AI were 19 percentage points less likely to reach the correct answer (Dell'Acqua et al., Harvard Business School, 2023).

For project planning, that splits roughly like this:

AI does well	You still need to do
Breaking a goal into phases and tasks	Deciding if the goal is the right one
Suggesting dependencies and an order	Checking who actually has time
Drafting task descriptions and checklists	Adding what went wrong last time
Spotting missing steps against a pattern	Setting priorities against other projects
Turning a long brief into a short plan	Owning the deadline

Step 1: Write a one-paragraph brief

The quality of the plan depends on the brief. Include:

Goal: what is true when the project is done. "Launch the new pricing page with three plans" beats "website work".
Deadline: a real date, and whether it is fixed or can move.
People: who is in, and roughly how much time each has.
Constraints and rules: budget, approvals, things that always apply ("every offer includes a price", "legal reviews all contracts").
What happened last time: a similar project that ran late, and why.

The last point matters more than it looks. People plan from the best case. In a classic study, students predicted they would finish their honors thesis in 33.9 days on average; they took 55.5 days, and fewer than a third finished by their own estimate (Buehler, Griffin and Ross, Journal of Personality and Social Psychology, 1994). An AI has the same blind spot if your brief only describes the plan and not the history.

Step 2: Let the AI draft the plan

Ask for a specific structure, for example:

Draft a project plan with phases, tasks, an owner for each task from this list of people, a due date for each, and which tasks depend on others. Flag any task that has no obvious owner.

A good draft has:

3 to 6 phases,
tasks small enough to finish in a few days,
one owner per task,
dependencies that make sense ("design approved" before "build page"),
the final deadline at the end, with some slack before it.

Step 3: Check the draft like a reviewer

This step is where your judgment matters most. Go through the plan with these questions:

Are the owners real and available? AI can assign tasks to people who are already full.
Are the dependencies right? Look for steps that could run in parallel, and for circular ones.
Are the estimates honest? Compare them with a similar past project, not with how long it should take. Buehler and colleagues found that prompting people to connect their prediction with relevant past experience removed the optimistic bias in one of their studies (1994).
Is anything missing? Approvals, reviews, waiting time for customers or suppliers.
Is there buffer? Hold back time before the deadline for what you didn't foresee.

Step 4: Cut the scope, not the quality

If the reviewed plan doesn't fit the deadline, ask the AI for options: what to drop, what to move to a second phase, which tasks could run in parallel. Then choose. Fewer, clearer goals usually help. In Atlassian's 2024 survey, 70% of knowledge workers said progress would be easier with fewer, more specific goals (Atlassian State of Teams, 2024).

Step 5: Move the plan into the place where work is tracked

A plan in a chat window is not a plan. Create the tasks in your shared work list with owners, dates and dependencies, so the plan is what people actually work from. If you use Claude or ChatGPT, tools that support MCP let the assistant create those tasks directly. See Project Management Tool That Works With Claude and ChatGPT.

Step 6: Keep the plan alive

Plans change in the first week. Keep the AI involved after the start: ask it what is late, what is blocked, which tasks have no owner, and what the change to one date means for the rest. How to Keep Projects From Missing Deadlines covers this part.

How Orakis does it

In Orakis, you plan a project in a conversation: you describe the goal, the deadline and who is in, and Orakis drafts the phases, tasks, owners and dependencies, taking your rules into account. The plan is checked in code for real assignees, valid dates and no circular dependencies, and nothing is created until you confirm. Everything stays editable. After the project starts, Orakis keeps coordinating it: it offers unassigned tasks to the right person, tells people when they can start, and checks finished work against the brief. Orakis is a young product in closed beta.

FAQ

Can ChatGPT or Claude plan a project? Yes, they can draft a solid structure from a good brief. They don't know your team's capacity or history unless you tell them, and they can't track the plan unless they are connected to where your tasks live.

How detailed should an AI-generated plan be? Detailed enough that each task has one owner and can be finished in a few days. Larger tasks hide risk; much smaller ones create noise.

Should I trust the AI's time estimates? Treat them as a first guess. Compare them with how long similar work took before, and add buffer before the deadline.

How to Plan a Project With AI (Step by Step)