task-ops
The pitch behind every AI task manager is the same sentence, stated more or less directly: stop deciding what to do next. Hand over the list, let the software work out the order, and start the day by opening whatever has been placed in front of you.
It is an appealing promise because deciding what to do next is genuinely tiring, and most people are bad at it in a specific way. Urgency wins over importance, the visible task beats the important one, and the thing that has been avoided for three weeks keeps getting avoided.
The promise is also partly true. Some of the work of running a list does hand over cleanly and stays handed over. Other parts break in ways that take about a month to notice, by which point the habit has formed and the list has quietly stopped reflecting reality. Knowing which is which before starting is worth more than a feature comparison.
Products in this category do up to three separate jobs, and most reviews blur them together.
Capture and triage. Something arrives, by email, in a message, spoken aloud, or in a page of notes, and it becomes a properly formed item with a title, a date and an owner. Todoist ships this as assistance for turning rambling input and email into tasks. Most workspace tools do it from meeting notes.
Ordering and prioritising. The list exists; the software decides what matters. Some products do this with rules, some with a model reading the descriptions, and the quality varies enormously by how much context the tool has.
Auto scheduling. The list is placed into actual calendar time. Each item gets a block, and when something slips or a meeting appears, everything downstream shuffles. Motion and Reclaim.ai are built around this. The scheduling is the product, not a feature bolted to a board.
These three fail for different reasons and suit different people. Treating them as one purchase is how teams end up disappointed by a product that was doing exactly what it said.
Capture and triage is the strongest of the three by a distance, and it is the least talked about because it makes a poor demo.
Consider what actually happens to a task between existing and being on a list. Someone says it out loud in a meeting. It gets written in shorthand in a notebook or a chat thread. Later, if the person remembers, it gets retyped into the list with a real title and a date. That retyping step is where most tasks die. Not because anyone decided they did not matter, but because there were eleven of them and it was six o'clock.
Removing that step changes the completeness of the list, and the completeness of the list is what everything else depends on. A brilliant prioritisation engine running on a list missing a third of the work is worse than useless, because it produces confident answers that are wrong.
The same applies to shared work. Pasting a page of meeting notes and getting a dozen draft items back, then deleting the three that are mistaken, takes a couple of minutes. Typing them by hand takes twenty and usually does not happen. This one mechanism explains most of the real productivity difference people report, and it has very little to do with intelligence.
Auto scheduling breaks on input quality, and the failure is subtle enough to be worth spelling out.
To place work into calendar time, the software needs two numbers from each item: how long it will take and when it is due. Both are supplied by the person, and both are routinely wrong. Estimates are optimistic by a large and fairly consistent margin. Due dates are often aspirational, set to create urgency rather than to record a commitment.
Feed those two numbers into a scheduler and it will produce a beautifully arranged week that cannot happen. Day one goes fine. By day three the blocks are shifting constantly because each task ran over, and the calendar becomes a rolling apology. The usual response is to stop trusting the blocks, at which point the tool has become an expensive list with colours.
There is a second failure mode in team and client work. A scheduler assumes the deadlines are yours to keep. In client work they frequently are not. The deadline moves because the file did not arrive, the approver was away, or the brief changed. Rescheduling an entire week off a deadline that is really a guess about somebody else's behaviour produces churn rather than clarity.
None of this means auto scheduling does not work. It works well for people whose work is largely internal, reasonably predictable, and estimated honestly, which describes a real and significant group. It works badly for everyone else, and the difference is about the shape of the work rather than the quality of the software.
Between capture and scheduling sits prioritisation, and it is the one that sounds easiest and turns out to be hardest.
The reason is that priority is not a property of a task. It is a property of the situation around the task, and almost none of that situation is written down anywhere the software can read. A two hour job is urgent because the client mentioned it twice on a call. A small item outranks a large one because the person waiting for it is about to go on leave. A task marked high is actually dead, because the project it belonged to was quietly cancelled last Thursday and nobody updated the board.
A model reading titles, labels and due dates sees none of that. What it can do reliably is apply consistency: surface what is overdue, flag what has no date at all, point out that four items are marked highest priority and therefore none of them are, and show what has not moved in a fortnight. That is genuinely valuable and it is a different claim from deciding what matters.
This distinction is worth holding onto when reading feature descriptions. A tool that says it prioritises the list is usually applying rules to the fields present. The useful question is not how clever the ordering is, but how much context the tool actually has to work from, and the honest answer is almost always less than the person has.
Published prices, checked on the vendors' own pricing pages, look like this.
| Product | Free tier | Paid |
|---|---|---|
| Motion | No free tier, trial only | $19 per seat per month, or $29 for the business tier |
| Reclaim.ai | Free forever, 1 user team, 1 week scheduling range, 5 AI agents | $10 per seat per month, up to 10 seats; $15 for the next tier, up to 100 seats |
| Todoist | Free, 5 personal projects, 3 filter views, 1 week of activity history, limited assisted capture sessions | Paid tiers add assisted task and email capture with no session limit |
| ClickUp | Free Forever with trial AI access | Base seat from $7 per user per month annually, plus an AI add-on at $9 or $28 per user per month |
Two patterns are worth pulling out.
Where the AI is the product, there is no free tier. Motion does not offer one, because there is nothing left after removing the scheduling. Reclaim's free tier limits the scheduling range to a week, which is the same restriction expressed differently.
Where the AI sits on top of a tool, it is a separate line. ClickUp's add-on pricing at $9 per user per month is higher than the base seat price on the entry plan, which is the clearest illustration in the market of how the cost structure has changed.
Both patterns push in the same direction: AI in task management is not a free feature, and any evaluation that skips the pricing page is incomplete.
Everything above assumes one person's list. Once the list belongs to a team, a different constraint takes over and it is not an AI constraint at all.
A shared list only works if everyone can see the same thing and talk about it in the same place. The most common reason a team's task list decays is that the decisions live in chat, the tasks live in a tool, and the two drift apart within a fortnight. Nothing about better prioritisation fixes that. The conversation has to sit next to the item it concerns.
For the same reason, automatic reordering is less useful on a shared list than on a personal one. When software silently changes the order, people stop being able to say "it is third on the list" and mean anything by it. Shared lists derive much of their value from being stable enough to refer to.
There is also an accountability effect that gets overlooked. When a person orders their own list, they have committed to that order and can be asked about it. When software orders it, the answer to why something did not get done becomes a shrug at the tool. That is a small cultural change with a large effect on a team of five, and it is not recovered by turning the feature off later.
The realistic shape for a team is therefore assisted capture plus a stable shared board, rather than automated ordering. Get things onto the board without typing them, then let people see the whole thing. A workspace that keeps boards and chat in one place removes the copy and paste step where a decision made in conversation has to be manually turned into an item.
Trials of these products tend to produce impressions rather than decisions, because the first week feels good regardless. A method that produces an answer:
Before starting, write down the three chores that consume the most time. For most people they are capturing things said out loud, writing the recurring update, and working out what has stalled.
For two weeks, record two numbers per chore. Minutes spent, and how often the output needed correcting versus rewriting from scratch. A draft that gets rewritten saved nothing, and it is easy to feel productive while rewriting.
For auto scheduling specifically, record one more thing: how many of the blocks placed in the calendar were actually worked at the time they were scheduled. If that number is below half by week two, the tool is not failing so much as revealing that the estimates were never accurate, and no scheduler can fix that.
At the end, compare against the published cost for the full team rather than for the person who ran the test. Uneven adoption is the norm, and per seat pricing does not care.
Pick the capture problem, not the prioritisation problem. For two weeks, stop retyping tasks: get them into the list by pasting notes, forwarding messages or dictating, whatever the current tool supports. A complete list is the prerequisite for every other claim in this category, and most people do not have one.
If the shared board underneath is the thing that needs replacing, Pinateca is free for up to five people and ten boards, with kanban, list, calendar and Gantt views and chat included from the start.
It can order a list against rules and deadlines, which is not the same thing. The ordering is only as good as the estimates and due dates supplied, and both are usually optimistic. For work with honest estimates and internal deadlines the result is often useful. For client work, where deadlines move for reasons outside the team, automatic reordering tends to produce churn rather than clarity.
Usually not, if the project tool already has assisted capture. The strongest benefit in this category is getting items onto the list without retyping them, and most workspace tools now do that. A dedicated auto scheduler is worth a second subscription only for someone whose specific problem is fitting committed work into a calendar, not for a team trying to see what everyone is doing.
The ones built entirely around scheduling either have no free tier or restrict the scheduling range severely, because the scheduling is the product. Tools where AI sits on top of a board tend to have a free plan with trial level AI access. Either way, free plans in this category are for evaluation rather than for running a team on.
Because the two inputs it needs are guesses. Time estimates are consistently optimistic and many due dates are set for motivation rather than recorded as commitments. The scheduler propagates both faithfully, so by the third day the blocks are shifting constantly. Tracking how many scheduled blocks were actually worked at the scheduled time shows within two weeks whether the underlying numbers are good enough.
A stable shared board that everyone can see, with the conversation attached to the items rather than in a separate chat tool. Stability is a feature on a shared list, because people need to be able to refer to positions and have that mean something. Use the AI for capture and for summarising the state of the board, and leave the ordering to the people who own the work.