A small team usually runs its projects in a mix of places. A task list here, a spreadsheet there, decisions buried in chat, and a calendar that does not know about any of it. Someone spends part of every week stitching it together.
Project tools now sell AI as the fix. The pitch is that it writes the plan, assigns the work, spots the delay and updates the status for you. Some of that is real. Some of it is a nicer way to type into the same list.
This post looks at two tools we list, Notion and Motion, and says what the AI adds to each, what it does not, and how to test it before you commit.
Quick answer: If your team keeps notes, documents and tasks together and wants AI that fills in and summarises that material, look at Notion. If your problem is working out who does what and when, with deadlines and capacity, look at Motion, which plans the work onto calendars. Neither fixes a project with no owner or no clear next step. Run a real week of work through a trial before paying, and check what each plan gates.
There are three jobs, and they are not equally mature.
Filling in and summarising. Turning a meeting into a list of actions. Filling in fields on a task. Summarising where a project stands. This is writing work, which language models do well, and a person can check the result in seconds.
Drafting the plan. Giving a tool a description and getting back stages, tasks and dates. Useful as a first draft. The risk is that it looks finished, so people stop thinking about what is missing.
Scheduling and prediction. Placing tasks on a calendar, warning that a deadline is at risk, forecasting a finish date. This depends entirely on your inputs. If task estimates are guesses, the forecast is a guess with a chart.
The pattern is simple. The more the AI writes things you can read and correct, the safer it is. The more it decides things you will not look at again, the more you need to watch it.
Notion: the workspace with AI inside it
Notion is a workspace for notes, documents and databases. Projects are one use of it. The vendor says its AI can fill in database properties automatically, run multi-step tasks across your pages and databases through an agent, run recurring work on a trigger or schedule, and write up meetings without a bot joining the call. It also describes answers with citations across connected apps such as Slack and Google Drive.
On data, the vendor says it does not use your data to train models unless you opt in, and that its AI providers are bound by contract not to train on customer data. It says the strictest handling, zero retention with the model providers, is on the Enterprise plan, and that other plans keep data with providers for a limited period. Read that page yourself if client information goes into the workspace.
Who it fits. A team that already wants its documents, notes and tasks in one place, and would use the AI to keep that place tidy.
The catch. In our review, AI is no longer sold to new customers as a separate add-on, so full use means the Business plan, and older articles quoting a separate AI price are out of date. If you are on the free or a lower plan you get a trial of the AI, not the real thing.
What it is not. It is not a scheduler. It will not look at your team's week and tell you that Priya is overbooked. You build the structure, then the AI helps you fill it.
Motion: the planner that works from deadlines
Motion starts from the calendar. The vendor says its AI planner takes tasks with priorities, dependencies, deadlines and durations, builds a plan, and re-plans as things change. It says the plan warns you when a task looks at risk. For teams, it describes showing when someone is over or under capacity, and predicting a project's finish date from the team's other commitments.
It also describes an AI project manager that can generate a project from a description and documents, with tasks, deadlines, assignees and stages. The vendor describes the output as something you review and polish, and quotes an accuracy figure for it. Treat that figure as the vendor's claim. We have not tested it, and what it measures depends on how they define accurate.
Who it fits. A person or team whose work is mostly tasks with real deadlines, split across several people, where the hard part is fitting it all together. The vendor points to agencies, consultants and service businesses.
The catch. In our review, it has grown from a calendar app into a larger work suite, and reviewers report price rises and a learning curve. There is a trial rather than a free plan, so the test has to happen inside the trial. We cover its scheduling side in more depth in our post on scheduling assistants.
Where the AI adds noise
A few patterns to watch for in either tool.
- A generated plan nobody owns. A tool can create thirty tasks in a minute. If no one is named on each, you have thirty more things to ignore.
- Estimates that were never real. Scheduling and finish-date prediction rely on the durations you enter. Wrong inputs give confident-looking wrong outputs.
- Updates nobody reads. Automatic status summaries are only useful if someone acts on them. If the weekly summary goes unread, the saving is zero.
- A weak process, faster. AI applied to a project with unclear goals produces unclear work more quickly. Fix the brief first. Our post on briefing an AI agency covers what a clear brief contains, and the same questions apply to an internal project.
One independent note on evidence. A 2023 literature survey of AI in software project management, on arXiv, reviewed eleven publications and discusses the challenges of using AI there. It is about software projects and it is now a few years old, so it tells you the field was still being worked out, not how today's tools perform. We found no independent testing of either tool's planning accuracy, which is why we say to test it yourself.
A one-week test before you pay
- Pick one real project with a deadline and three or more people.
- Put the actual tasks and honest estimates in. Do not tidy them first.
- Let the AI do its part: draft the plan, fill in fields, summarise the meeting.
- At the end of the week, count the edits. How many generated items did you keep as they were, change, or delete?
- Ask the team one question: did this save you time, or did it give you something extra to maintain?
If the answer to the last question is the second one, stay on what you have. A shared task list and a weekly check-in cost nothing and work for many small teams.
Where to start