task-ops
Search for AI project management tools and the results are ranked lists. Five best, six best, fourteen tested. They are useful for learning which products exist and close to useless for deciding, because every one of them is written for a reader with a budget line and a procurement process. A team of five or six people has neither, and the constraints that actually decide the question for that team appear in none of the rankings.
Those constraints are unglamorous. Whether the free tier fits, what the bill looks like at the sixth seat, whether the AI is billed per person or per use, and whether the product is built for coordinating a group or for scheduling one person's day. Get those four right and the feature comparison mostly stops mattering. Get them wrong and no amount of AI quality rescues the decision.
The phrase "free forever" carries very different meanings across the category, and for a team of five the differences are decisive rather than cosmetic.
| Tool | Free tier | Published cap |
|---|---|---|
| Asana | Personal, free forever | 2 users, files up to 100 MB each |
| monday.com | Free | Up to 2 seats |
| Reclaim.ai | Lite, free | 1 user, 1 week scheduling range |
| Todoist | Beginner, free | 5 personal projects, 3 filter views, 1 week of activity history |
| ClickUp | Free Forever | Unlimited tasks, 60MB storage, trial access to advanced AI |
| Motion | No free tier | Trial only |
A two seat cap is not a free plan for a team. It is a plan for an individual who occasionally shares something. Any team of four or more is choosing between paid plans from the first day, and should evaluate them on that basis rather than on the trial experience.
This is the single most common mistake in the decision. A small team trials three tools on their free plans, picks the one that felt best, and discovers at rollout that the plan they trialled cannot hold the team. The evaluation then restarts under time pressure, which is how teams end up on whatever was easiest rather than whatever was right.
The practical move is to decide the seat count first, look up what that seat count costs on every candidate, and only then look at features. It takes twenty minutes and removes about half the shortlist.
The rankings list these tools together, but there are two genuinely different products in the category and they solve different problems.
Auto schedulers. The product reads a list of tasks with estimates and deadlines, looks at the calendar, and places the work into the gaps. When something slips, everything downstream reshuffles. Motion and Reclaim.ai are built this way. The AI is the product rather than a feature on top of it, which is why Motion has no free tier: there is nothing to give away.
Coordination tools with AI added. The product is a board, a list and a schedule that a group shares. The AI drafts, summarises, answers questions about the workspace and runs agents. ClickUp, Asana, monday.com and Notion sit here.
These fit different problems. An auto scheduler is transformative for a person whose difficulty is fitting committed work into a week, and it does nothing for a team whose difficulty is that three people disagree about what is committed. A coordination tool is the reverse. Reading a comparison that ranks both against each other produces a ranking of apples against a Tuesday.
The honest test is one question: is the problem that the team does not know what everyone is doing, or that individuals cannot fit their day together. Small studios and agencies almost always have the first problem. Consultants and solo operators usually have the second.
There is a further wrinkle for teams that do client work. Auto schedulers assume the estimates are real and the deadlines are yours. In client work neither holds. A deadline moves because the client did not send the file, and an estimate is a negotiating position rather than a measurement. Rescheduling a whole week off numbers that unreliable produces confident nonsense, and teams abandon it after a month. Where auto scheduling does hold up is internal work with genuine estimates, which in an agency is a minority of the calendar.
None of this means the scheduling category is weak. It means the category is precise about who it serves, and the ranked lists blur that by placing it alongside coordination tools under one heading.
Published pricing now splits into three models, and they suit very different usage patterns.
| Model | Example | How it behaves |
|---|---|---|
| Separate per seat add-on | ClickUp Brain at $9 per user per month, or $28 for the higher tier | Predictable, paid for every seat including the ones that never use it |
| Included with a request cap | Asana Starter at $10.99 per user per month annually, with 5 requests per user per month capped at 50 per account | Cheap to start, hits a wall; extra requests run $0.50 prepaid or $0.60 pay as you go |
| Credit meter | Notion custom agents at $10 per 1,000 monthly credits; monday.com allowances of 1,000 to 3,000 credits a month by tier | Light use costs little, heavy use is hard to forecast |
For a team of six, the arithmetic matters more than it looks. A per seat AI add-on at $9 costs the same whether one person uses it constantly or all six do. Since adoption is almost never even, that usually means paying six times for value that lands with one or two people.
A request cap has the opposite failure. Fifty requests across an account is one busy afternoon for one person. The cap is not a budget control so much as a signal that heavier use is expected to be bought separately.
Neither model is wrong. The point is to know which one the team's usage shape fits before signing, rather than after the first overage notice.
Strip away the demos and a short list of chores are the ones teams still hand over six months later.
Pulling structured items out of unstructured input is the strongest. A page of meeting notes becomes a dozen draft items, three of which get deleted. Nothing else in the category removes as much mechanical typing.
Producing the recurring written update comes second. Whoever assembles the Monday summary by reading boards and rewriting item titles is doing work a model does acceptably, and the correction pass is far shorter than the writing pass.
Answering questions about the state of the work comes third, and it is the one that scales with team size. When colleagues can ask the workspace directly, the lookups that used to interrupt the coordinator stop arriving.
Surfacing what has gone quiet is fourth and the most underrated. Boards are bad at showing stalled items, because a card sitting still looks identical to a card being worked on. Asking which items have had no activity since a date is a query a model runs well.
Notice that all four are reading, restating and retrieving. None of them decides anything. Any claim that a tool will run the project is a claim about the fifth category, and it is worth treating as unproven until it is tested on real work.
Three checks change the outcome more often than the feature grids do.
What happens to the AI on a messy workspace. Summaries and workload answers read the same fields a person would read. Items with no owner and no date produce answers that say so. The teams who get the most from these features are the ones who spent an afternoon filling in fields first, and that afternoon is free.
Whether the data comes out. Export format, whether comments and attachments come with it, and whether history survives. This never matters until the day it matters entirely.
Whether the board types the work needs are present without an upgrade. A team that needs a schedule view, a calendar and a kanban should confirm all three exist on the plan being bought. Gantt in particular sits behind a paid tier on several products. The side by side comparison of the major tools is a quicker way to see where those lines fall than opening five feature pages.
Every ranked list compares subscription prices. None of them price the thing that actually dominates the total: the hours the team spends moving onto the tool and learning it.
For six people, a realistic migration is a day of setup by one person and a week of reduced output while everyone relearns where things are. At any plausible hourly rate that dwarfs a year of subscription differences between two candidates. Which means two things follow.
First, a difference of a few dollars per seat per month is not a reason to choose one tool over another. It is noise next to the switching cost. Choosing on price alone and then switching again in eight months is the expensive path, not the cheap one.
Second, the value of a tool that the team does not outgrow is much higher than the price difference suggests. A product where the schedule view, the calendar and the shared conversation are all present from the start avoids the migration that happens when a team hits a wall on a tool that only does one of those well. The same logic applies to seat growth: a plan that becomes uneconomic at the twelfth person is a migration scheduled for whenever the twelfth person is hired.
This is also the reason to be cautious about picking a tool because its AI is currently ahead. AI capability is the fastest moving part of the category and the easiest for a competitor to close. Board types, data portability and pricing structure move slowly, and those are what the team will live with.
There is a fourth option that the rankings do not cover, because it does not produce an entry in a list of tools.
Most small teams already pay for ChatGPT or Claude. If the project tool supports MCP, that subscription can be pointed at the boards directly. The assistant reads what is due this week, turns pasted notes into items, moves things between lists and fills in fields, using the same public API a person's clicks go through. The actions land in the activity log exactly as manual ones do.
The consequence is that the AI capability arrives without a second AI bill, and the project tool can be chosen on whether it holds the work well rather than on whose assistant writes better summaries. Tools that connect to ChatGPT and Claude this way are still a minority, so it is worth asking directly rather than assuming. On the ChatGPT side, adding a custom connector requires a paid plan, which is a real constraint for anyone on the free tier.
This route also ages better. Assistants are improving faster than project tools are, and a setup where the assistant can be swapped without migrating the work is worth something.
Write down the team size for the next twelve months, then price every candidate at that number including whatever the AI costs. That one step eliminates most of the shortlist and takes less time than reading a ranked list.
Then test the shortlist on one real project rather than a sample one, and judge it on the four chores above. If the base tool is what needs fixing rather than the AI on top of it, Pinateca is free for up to five people and ten boards, with kanban, Gantt, calendar and timetable views and chat all present from the start.
The AI features are worth it if the team has recurring written chores: status updates, turning notes into tasks, answering questions about what is in flight. They are not worth much to a team of five whose real problem is that the work is not written down anywhere. In that case the base tool matters and the AI on top of it does not, because there is nothing for it to read.
Several free plans cap at two users, including Asana Personal and monday.com Free, which makes them unsuitable for a team of four or more. Others cap on storage or project count instead. The practical approach is to look up the seat cap on each candidate before trialling, since a free plan that cannot hold the team will force a second evaluation later.
A scheduler takes one person's task list and places it into calendar gaps, reshuffling automatically when something slips. A project management tool with AI keeps shared boards and adds drafting, summarising and question answering on top. The first solves an individual time problem, the second solves a group coordination problem, and a team buying the wrong one will find the features irrelevant rather than merely weak.
It depends on the billing model rather than the tool. A per seat add-on is charged for every seat whether used or not, which on published pricing can exceed the base seat price itself. Credit based pricing costs little for light use and is harder to forecast for heavy use. Included allowances are the cheapest to start and the easiest to exhaust, with extra requests then billed individually.
For reading boards, creating and moving items and drafting descriptions, often yes, provided the project tool supports MCP and the AI plan allows custom connectors. The limits are real: billing, API tokens and account settings are deliberately not exposed to an assistant, and the assistant can only see what the account connecting it can see. For the common chores, it covers most of the same ground without a second subscription.