AI Agent for Project Management: What Does the Work in 2026

Search for an AI agent for project management and you get three different products wearing one label. One drafts the status update. One decides when you will write it. One goes and collects it.

The board is not the work. A project tool is a place to record what the work is, who owns it, and when it slipped. Almost all of the AI shipped into those tools operates on that record: it summarises the record, rewrites the record, predicts what the record will look like in three weeks. The chasing, the collecting, the reporting and the reminding still land on a person.

Usually you.

I run an agent platform, so this is the complaint that reaches me most often from people who manage projects. They already pay for a tool with AI in it. Monday morning has not changed.

So this post sorts the category by what the AI is actually pointed at, and says plainly where the line sits between a feature and an agent.

Three tiers of AI in project management

This is the distinction the category will not make for you, because the word "agent" sells all three. Sorting a product into a tier takes about a minute and saves a procurement cycle.

The assistant tier

An assistant reads and writes against records you maintain. Ask it what happened last sprint and it reads the sprint. Ask it to write the update and it drafts one from whatever was recorded. What it never does is leave the tool, so the update still needs someone to gather the facts that were never typed into a task in the first place. It operates on the record, not on the work.

Where a scheduler earns its seat

A scheduler takes a known pile of work and decides the order and the timing. That is a real job, and for one person with sixty tasks and a fragmented calendar it is the whole job. The value is arithmetic you were doing badly in your head. The limit is that it cannot invent capacity, so a plan that does not fit stays a plan that does not fit, now with better graphics.

What a doer has to prove

A doer is given an outcome and comes back with it finished. Collect the three missing updates. Build the client status pack. Remind the vendor, again, and tell me if they answer. Three things make that safe: it must reach the systems where the task actually lives, it must stop and ask before anything touches a customer, and someone other than you has to keep it working when an API changes. Miss any of the three and you have bought yourself a new maintenance job.

Most people typing "AI agent for project management" into Google want the third tier. Most of what they find is the first.

How I sorted them

Four questions, asked of every product below. They are deliberately not feature questions, because feature lists in this category are written to blur exactly the line I am drawing.

• Does the work leave the tool? If the output is text inside the same product, it is an assistant, however it is marketed.
• Who is on the hook when it breaks? On a suite, you are. On a platform where someone published the agent, the publisher is.
• What is the unit of value? A paragraph, a reordered calendar, or a completed task. These are not close to each other.
• What do you still own afterwards? Board hygiene, the data model, approvals, the conversation with the person who is late. Something always stays, and knowing which part stays is the actual buying decision.

A note on where the numbers come from, because it is the reason this post looks different from the other results on this page. I read the pricing pages for Asana, monday.com, ClickUp, Notion, Motion and Dust on 21 September 2026, and every figure below is what those pages rendered that day, including the billing toggle each one was sitting on. I have not run a bake-off, and nothing here describes what these products feel like to use.

Atlassian's Jira pricing page is the seventh I asked for, and it returned its title and no body content. It renders its plans in JavaScript. So Jira gets no price, no plan names and no capability claims in this post, and its row in the table below says exactly that. A page that will not render is a real result, and filling the hole from a review site is how a post like this goes quietly wrong. The checklist further down tells you what to look at when you open a pricing page yourself.

Quick comparison

Prices and allowances are what each vendor's own pricing page rendered on 21 September 2026, quoted in their words where the words matter. Billing states differed: Asana and monday.com showed annual-billed rates on the toggle I read, ClickUp rendered its monthly and yearly rates together, Dust had a yearly toggle, and neither Notion nor Motion made a billing period determinable, so I have assumed neither. monday.com was running a "Special monthly discount" of $10 off Basic that day, which is a promotion and will end. Dust prices in euro and the page states "Excluding VAT"; no conversion applied. Jira's page returned no readable content. The tier column is my own classification, and "Not assessed" means exactly that.

Four of six publish a meter, and the meters are small

Here is the count none of the other pages ranking for this term will give you. Of the six pricing pages I read on 21 September 2026, four publish an explicit AI meter, one publishes an automations cap and no AI price, and one publishes nothing at all. That is the whole finding, and it applies to those six pages and no further.

• Asana, metered. The pricing page headline is "Agentic Work Management". Starter, $10.99 per user a month billed annually, includes "AI Teammates and Asana Dash, includes 5 requests per user per month, up to 50 requests per account". The same plan separately includes "AI Studio, includes 50K credits per billing account per month", which is a different meter for a different product.
• monday.com, metered. The page headline is "Plans for every team, people and agents included." Basic, $9 per seat a month billed annually, carries "3 tools, 1,000 AI credits / month". In the feature list, "AI agent workforce" is marked Limited on this plan. Standard, at $12 a seat billed annually, moves to "5 tools, 2,000 AI credits / month", itemised as "2,000 credits = $20".
• Motion, metered. Pro AI at $19 per seat carries 7,500 credits per seat per month, priced on the page at "25 cents/100 credits". Business AI at $29 per seat carries 15,000 credits, at "19 cents/100 credits".
• Dust, metered. Free is EUR 0 with 500 credits, which the page labels lifetime. Pro is EUR 24 per seat a month for 8,000 credits, Max is EUR 120 for 40,000. The page states "Excluding VAT".
• ClickUp, an automations cap and no AI price. Business, $12 per user a month billed yearly or $19 billed monthly, publishes 5,000 automations a month; Enterprise publishes 250,000. That is an automations cap. ClickUp Brain, the AI product, was not priced on the page, so there is no AI number here to report.
• Notion, nothing published. Business, $20 per member a month, is where Notion Agent appears, described on the page as "Does work for you. Completes complex, multi-step tasks using context from Notion, your connected apps, and the web." No request or credit cap is published against it.

Three things fall out of those quotes.

Asana's allowance stops being per user sooner than it reads. Five AI Teammates and Asana Dash requests each, capped at fifty per account, means a ten-person team reaches the account ceiling exactly and the eleventh person adds none. So a ten-person team on a plan sold under the words "Agentic Work Management" has fifty of those requests for the month, which is around two a working day for the entire company. AI Studio's 50K credits are a separate allowance for a separate product and do not top that number up.

Two of the four metered pages also print what a unit costs, which is the most useful disclosure any of them make. Motion prints "25 cents/100 credits" on Pro AI and "19 cents/100 credits" on Business AI. monday.com itemises its allowance on the bill: on the page's default selection of 10 seats, 1,000 AI credits were listed at $10 against $90 of seats, with a "Special monthly discount" of -$10 applied underneath. Roughly a ninth of that bill is the AI line. Promotions end; the itemisation is what makes the shape of the deal visible.

The sixth case is the one to ask about. Notion makes the boldest autonomy claim on any of these pages, "Does work for you", and publishes no cap at all beside it. No published cap is not the same as no cap, so ask before you buy.

None of this makes them bad products. Four of them are project management systems, and a metered assistant inside a system you already run is a reasonable thing to sell. What it does mean is that the words "AI agent" on a pricing page carry much less weight than they appear to. The question worth asking is how many times a month you are allowed to use it, and what the vendor charges after that.

Gravity is priced the other way round, as a subscription with usage included, and you buy more usage if you run out. That is a different bargain with its own trade-offs, and it is the honest reason the last row of that table does not look like the others.

1. Asana: the assistant inside a process you already run

Asana is a work management suite. Its pricing page sells timeline and Gantt views, reporting dashboards, forms, custom fields and unlimited automations, which tells you where its centre of gravity sits: the shared, recorded process. That record is also the boundary of what its AI can reach. A model reading a well-kept Asana instance has an unusually good dataset to work from. A model reading a half-maintained one does not.

The trap is buying AI to fix a reporting problem that is really a data problem. If half the work never gets entered, no assistant can report on it, and the status update stays a Slack archaeology exercise every Thursday afternoon.

What the page published on 21 September 2026: Personal is free forever for two users, with unlimited tasks and projects and list, board and calendar views. Starter is $10.99 per user a month billed annually, and it drops the seat limit while adding timeline and Gantt views, reporting dashboards, forms, custom fields and unlimited automations. Advanced and Enterprise sit above it. Their prices were below the fold of what I captured, so I am not stating them.

What the assistant cannot reach

It cannot reach anything that was never typed into Asana. The supplier who replies by email, the spreadsheet a client keeps, the approval sitting in someone's inbox: all of that is outside the boundary, and gathering it is the part of project management that eats the day.

2. monday.com: AI on top of the automations you configured

monday.com sells a configurable board that you shape to fit your process. Its pricing page counts entitlements in boards, docs and seats, and gates the AI behind the first paid tier: the free plan's line reads "AI features: Available from Basic". The AI arrives into a structure you built, which means what you get out of it is downstream of how well you built it.

Worth being clear-eyed about what that implies. You are the one who designed the board, so you are the one who maintains it, and you remain the one who notices when an automation quietly stopped firing three weeks ago. If your objection to project admin is that you have become the unpaid operations engineer for it, more AI on the same board is a lateral move.

Credit to monday.com for one thing most of this list does not do: it shows its working. The pricing page itemises the AI allowance as a line on the bill. On the default selection of 10 seats, Basic's "1,000 AI credits / month" was listed at $10 against $90 of seats, with a "Special monthly discount" of -$10 applied below it. So the AI line was about a ninth of that bill before the promotion and nothing after it, and the promotion will end. Seeing the split at all is the useful part.

Plans read on 21 September 2026

Free forever, up to 2 seats, with "AI features: Available from Basic". Basic $9 per seat a month billed annually. Standard $12 per seat a month billed annually. The seat selector defaulted to 10 seats, which is why the page prints a per-seat figure and a total.

AI allowance

Basic: "3 tools, 1,000 AI credits / month", with "AI agent workforce" marked Limited. Standard: "5 tools, 2,000 AI credits / month", itemised as "2,000 credits = $20".

Promotion on the day

"Special offer" of $10 off a month on Basic and $20 off on Standard, read 21 September 2026.

Verdict

The most itemised pricing page in this set. The itemisation is what lets you see how small the allowance is.

Before you add anything, open your own plan and find which AI actions are included and which sit behind an upgrade.

3. ClickUp publishes an automations cap and no AI price

ClickUp's pitch is consolidation. Its free plan alone lists unlimited tasks, collaborative docs, Kanban and sprint management, and the paid tiers add unlimited storage, dashboards and portfolio management. For an assistant tier product that matters, because the more of a company's context sits inside one boundary, the more a model can answer without being told.

It is also the cheapest paid entry of the six pages I read. Unlimited is $7 per user a month billed yearly, or $10 billed monthly. There is a toggle, and unusually the page renders both numbers at once, which saved me the guesswork the other pages cost. Business is $12 billed yearly or $19 billed monthly, and the page advertises a "100% Money-back Guarantee".

Here is the part that matters for this post. ClickUp publishes no AI allowance at all. What Business publishes is 5,000 automations a month, against 250,000 on Enterprise, and an automation is not an AI request. ClickUp Brain, the AI product itself, was not priced on the page I read, so I have no AI number to give you and neither will you until you ask them.

The cost of consolidation is the migration, and the second cost is that a workspace holding everything also holds a lot of noise. Three things decide whether it pays off:

• Whether your team will actually stop using the tools you are replacing
• Whether the docs get maintained, since a stale doc answers confidently and wrongly
• Whether the work you want automated lives inside the workspace at all, or out in email where most chasing happens

That last one is the same boundary problem every suite on this page has.

4. Notion makes the boldest claim and publishes no number

A lot of small teams run projects in Notion because they were already writing everything there and a database was one click away. The free plan alone lists databases with subtasks and dependencies, which is enough structure to hold a small tracker.

Notion's AI works over pages and databases you designed, which makes the schema the real variable. A properly typed database with dates, owners and statuses is queryable. A page of bullet points pretending to be a tracker gives a model nothing to query, and no amount of model quality fixes that.

The page gave three tiers on 21 September 2026: Free at $0 per member a month with a trial of Notion AI, Plus at $10 per member a month also with a trial, and Business at $20 per member a month. Business is where Notion Agent appears, described in the most confident words on any page in this post: "Does work for you. Completes complex, multi-step tasks using context from Notion, your connected apps, and the web." Nothing beside it says how many times. The section I read also showed no monthly or annual billing state, so treat both figures as the published rate and check the toggle yourself.

I wrote up the narrower version of this job in AI agent for Notion project tracking, including where an outside agent fits better than the built-in feature.

5. Motion: priced per seat, with the billing period still unreadable

Motion is the product I have now failed to pin down on five separate reads, and at this point the failure is the finding. Two paid plans rendered on 21 September 2026: Pro AI at $19 per seat, carrying 7,500 credits per seat per month at a printed "25 cents/100 credits", and Business AI at $29 per seat, carrying 15,000 credits at "19 cents/100 credits". It belongs in the metered column, and it is the only one of the six that prints a unit price against both paid tiers.

What the page would not tell me is whether $19 and $29 are monthly or annual rates. It carries Individuals and Teams tabs and a "Pay monthly" or "Pay annually (Save 33%)" control, with no default state I could determine. A 33% swing is not a rounding error. Five reads, no answer, so I am not going to print one, and I would treat any page that quotes a Motion price without saying which state it read as a page that did not check.

On fit, a judgement of mine, which the page makes no claim about either way: I have Motion in the scheduler tier, the middle of the three. If you know what needs doing and lose the week deciding when, that tier is aimed at you. If you know exactly when everything should happen and lose the week chasing other people, the calendar was never the bottleneck.

6. Dust: an agent layer, priced in euro and metered like the rest

Dust sits one level up from the suites above it. Its pricing page describes what it sells in its own words: "20+ frontier models", "Custom agents with your skills, knowledge & tools", "Multi-agent workflows on schedules & triggers", plus connectors to Slack, Notion, GitHub and Drive "or any tool via MCP". It lists SOC 2 Type II and both US and EU data residency. What you are buying is an agent you configure yourself.

The prices are in euro and the page states "Excluding VAT", so they stay unconverted here. Free is EUR 0 with 500 credits, and the page calls that allowance a lifetime one, so it never refills. Pro is EUR 24 per seat a month for 8,000 credits. Max is EUR 120 for 40,000. Enterprise is a sales conversation, offering "Workspace-pooled credits & volume pricing". A yearly toggle advertising 20% was on the page.

The comparison table further down prints hard limits worth reading before a company-wide rollout: the Business tier is described as "For teams up to 100 people" and carries 5 remote MCP servers, up to 3 connectors and 5 Spaces.

The condition attached to all of it is capacity. Somebody has to build the agents, connect the sources and decide what they are allowed to touch. In a company with an ops or platform team, that is a good trade. For a five-person business, it is a new project on top of the projects. Gravity vs Dust sets out the build-versus-buy version of that choice.

7. Gravity: best when you want the project task finished, not summarised

This is my product, so discount me accordingly. I have put it last, and the table above is the honest version of where it fits: one tier, one job, and a clear set of things it does not do.

You describe the task in plain words. Gravity matches it to an agent an independent builder already made and tested, you sign in to the accounts it needs, answer a few setup questions, and it runs on a schedule. The project work people hand over first is the repetitive administrative half: chasing updates that are three days late, assembling the weekly status pack, collecting documents that never arrive on time, sending the reminder nobody wants to send twice.

Gmail, Sheets, Calendar, Drive, Slack and Notion connect through an ordinary login. You never paste an API key. Anything that messages a customer waits for you to approve it, every agent passes automated checks before it is listed, and a run that fails on a platform error, with no usable output, is refunded under the refund policy.

Who keeps the agent working

The builder who published it keeps it working. When an API changes or a format shifts, fixing it is their job and their revenue depends on it, which is the structural difference between this and a workflow you built yourself at eleven at night. If you want the longer version of that argument, what an agent operating system is explains why the maintenance question decides more of these purchases than features do.

Best for

The person who spends Monday morning asking other people for things, and Friday afternoon turning the answers into a document.

Pricing

First agent free, no card. Then Pro at $5 a month during the alpha, or Max at $20 a month for more agents and usage. Buy more usage if you run out. As of September 2026.

Free tier

Yes. One agent, free.

Verdict

Keep your tracker. Hand over the chasing and the assembling that happen outside it.

What to check before you pay for AI in a project tool

The prices above will age. The method will not, so here are the five lines to find on any pricing page in this category before the card comes out. Gravity's own answers, so you can hold me to the same test: the usage is included in the subscription, it is not priced per seat, and there is no second meter to find.

1. Is the AI included in your current plan, or is it an add-on with its own line? These are frequently different answers on the same page.
2. Is it priced per seat? Multiply by everyone who needs access, then ask how many of those seats will open it after the first month.
3. Is there a second meter underneath? Actions, credits, runs or tokens. Find the allowance first, then hunt for the unit price, which is usually a click further down. Asana prints its allowance on the plan card, monday.com itemises it on the bill, and Motion and Dust print theirs per seat, so this took one read on each. ClickUp and Notion published no AI allowance at all, which is its own answer.
4. Does the free plan or trial include the AI? monday.com's free tier settles it in one line, "AI features: Available from Basic", and Dust labels its free credits a lifetime allowance, so 500 is all you ever get. A trial that excludes the feature you are evaluating tells you nothing about the feature you are evaluating.
5. Is the headline number monthly or annual? Six pages gave me four different behaviours on 21 September 2026. ClickUp rendered both states at once. Asana and monday.com sat on a yearly toggle with the saving advertised beside it. Dust had a yearly toggle too. Notion and Motion gave me no determinable state, so I assumed nothing and said so.

Those five turn a pricing page into a comparable number in about four minutes. If you want the same logic applied across agent platforms specifically, AI agent pricing explained breaks down the billing units, and cheapest AI agent platforms is the low-budget end of the same market.

Which one should you choose?

Start with the tier. Nearly every bad purchase in this category is a tier error: someone with a chasing problem buys an assistant, or someone with a sequencing problem buys an execution platform.

If your team already lives in a suite, the cheapest possible experiment is the one you have already paid for. Turn on whatever AI your Asana, monday.com, ClickUp or Notion plan includes and use it hard for a fortnight before you evaluate anything else. You learn two things: whether the gap is reporting or gathering, and how far the allowance actually goes. On Asana Starter a ten-person team is working through fifty AI Teammates and Asana Dash requests for the whole month, so two weeks of real use finds the ceiling.

If the answer comes back "gathering", no assistant in any suite will close it, because the information is not in the suite.

If you do not have a project management system yet, buy one of these four before you buy an agent. Gravity is not a tracker and is not trying to become one. An agent is worth adding once there is a system for it to work around, and it is a bad substitute for the system itself.

Choose Motion if you are one person and the week collapses under sequencing, and settle the billing period with their sales team first, because the page would not settle it with me. Choose Dust if you have the internal capacity to build and maintain agents, a company-wide set of tools worth connecting, and no objection to a euro invoice.

Choose Notion or ClickUp on the strength of the workspace, because in both cases the assistant is only as good as the structure you keep underneath it, and that structure is work you were going to do anyway. Neither publishes an AI allowance, so ask for one in writing before you plan around it.

My own bias, stated plainly: for the repeating administrative half of project work, the chasing, the collecting, the weekly pack, I would pick the option where nobody builds anything and the task comes back finished. That is Gravity, and the first agent is free. For planning, prioritising and the conversation with the person who is late, you are still the project manager.

FAQ

Sources