Trello AI Agent: What It Does and What It Costs in 2026

Open the board. Scroll to the third list. There are cards in there from June.

Nobody archived them, because archiving eleven cards is a four-minute job and four-minute jobs never get scheduled. So the board grows. By September the standup is a scrolling exercise.

That is the chore people mean when they search for a Trello AI agent. Not a roadmap. A board that tidies itself, chases its own due dates, and tells you on Monday what actually moved.

Two routes get you there. Trello sells AI inside its own paid plans. Other companies sell an agent that sits on top of the board through the API and reaches what Trello cannot see. Almost every page comparing the two skips the part that decides it, which is the price of the first one.

I read Trello's pricing page on 16 September 2026. The AI starts at $5 a user a month, and it is priced per user, which matters more than it sounds. Every other number below came off a vendor's own page the same day.

What Trello's own AI costs, and where it sits

Trello's AI is a paid feature. That is the sentence missing from every other page on this topic, and it is the first thing to settle before you go shopping.

On 16 September 2026 Trello's pricing page put it plainly: upgrade to unlock the new AI features, starting at $5 a month. Here is where that $5 sits.

• Free. A Free plan exists, and the AI features are what the page asks you to upgrade for.
• Standard, $5 per user a month billed annually, $6 billed monthly. The page sells the tier on AI features, unlimited boards and card mirroring, in that order.
• Premium, $10 per user a month billed annually, $12.50 billed monthly. Summarised on the page as adding AI to your boards along with admin controls.
• Enterprise, $17.50 per user a month billed annually. The page states the same figure as $210.00 a year per user, which is the one number on the table you can check with a calculator.

Read both billing columns before anything else. Standard is $5 or $6 for the identical feature set, depending on whether you commit to a year. Premium is $10 or $12.50. Paying monthly costs a quarter more on Premium and a fifth more on Standard.

What the AI does on a card is the part I will not embroider. The pricing table names the tiers and tells you AI arrives at Standard. It does not itemise the features there, and I am not going to invent a list to fill the gap. What matters for this decision is where the AI sits: inside Trello, on the plan you already pay for.

That is a real advantage. It is also the boundary.

What five seats actually cost

Per user is the phrase that changes the answer, because a board is almost never one person.

Five people on Standard billed annually: $25 a month, $300 a year. The same five on Premium: $50 a month, $600 a year. Switch to monthly billing and those become $30 and $62.50. Twelve people on Premium billed annually is $120 a month, and nothing about the board has changed except the headcount.

Seat pricing does exactly what it says on the tin. It is also the arithmetic that decides whether the native route or the agent route is cheaper for you. A solo Trello user pays $5 a month for AI features. A team of twelve pays $60 for the same features on the same board, because the price follows the headcount rather than the work.

Atlassian Guard is a separate line on the invoice

Trello is an Atlassian product, and Atlassian Guard is its own subscription, starting at $4 a month per user. It is not included in the Standard, Premium or Enterprise prices above.

If your admin asks for it, add $4 per person to whatever you worked out a paragraph ago. For the five-person board on Standard, that is $25 plus $20, which is a different conversation from $25.

Quick comparison

Five ways to get AI onto a Trello board, with what each one actually charges. Prices are what the vendor's own page showed on 16 September 2026.

Prices read from trello.com/pricing, zapier.com/pricing, lindy.ai/pricing and relevanceai.com/pricing on 16 September 2026. The true total on the Trello row depends on headcount: a five-person board on Standard billed annually is $25 a month, on Premium $50 a month. Atlassian Guard, if your admin adds it, is a further $4 a month per user. Zapier's Team plan is $800 billed annually, which works out at $66.67 a month.

One column deserves more attention than the price column, and it is the third. Trello's AI is bundled into the seat you already pay for. Zapier hands you a canvas. Relevance AI hands you a canvas and an implementation partner. Only Lindy and Gravity sell somebody else's finished work, and they split the territory: Lindy takes the inbox, Gravity takes the repeated business task with an outcome attached.

The cards that have not moved

This is the job everyone actually wants. Find the cards nobody has touched in three weeks, work out which are dead and which are quietly blocked, and do something about both.

An agent is good at the first half. Last activity date is a fact it can read. Whether a stall means abandoned or waiting on a client is a judgement, and it depends on things the board never recorded.

So give it the rule instead of the judgement. No movement in 14 days and no due date means a nudge in the comments. No movement in 30 days in a list called Backlog means a candidate for archiving, surfaced to you rather than archived on the spot.

The output that makes this useful is boring. A short list, once a week, of cards that have gone quiet and who they belong to. Something you can read on a phone while the kettle boils.

The Trello board automation walkthrough covers the routing rules in more detail, including what happens when a card matches two of them at once.

The board that needs archiving on a schedule

Archiving is the chore that proves the whole idea, because it is trivial, recoverable, and nobody ever does it.

The shape is always the same. Pick the completed list, pick an age, run it monthly, keep a record. Trello archiving is reversible, which makes it the safest first task to hand over.

It is also where people get burned, and the reason is never the AI. It is that somebody pointed a rule at a list called Done without noticing the team also drops reference cards in there.

Two checks before you schedule anything

First, run it in report mode for one cycle. The agent produces the list of what it would archive, and you read it. One card you would have missed is a rule bug found before it cost anything.

Second, insist on a log. Date, card name, list it came from, who ran it. Archived cards are recoverable in Trello, but only if you know what to look for, and a month later you will not.

The full Trello board archiving guide covers the review window, safe list naming, and what to do with cards carrying attachments somebody else still links to.

Turning an inbox or a form into cards

The work that arrives outside Trello is where the agent route earns its money. A shared support address, a Google Form, a WhatsApp message somebody screenshots into Slack.

Somebody reads it and makes a card, which means doing three things at once: deciding whether this is a real request, writing a title that still means something in a fortnight, and picking the list. Only the third is mechanical.

An agent handles the mechanical part well and the title surprisingly well, since turning a rambling email into a clear line is what language models are for. Judging whether the request is real is where you keep a hand on the wheel. Route the uncertain ones to a triage list.

Then watch for duplicates. The same client emails twice, once to support and once to someone they met at a conference. Intake needs a check against open cards first, and the check has to look at content rather than subject line.

Monday morning: what actually changed

The artefact people ask for is the same one almost no board tool produces: a short written account of what changed last week, delivered where they read things.

A count of cards moved is worthless. What moved, what did not move that should have, and what is now overdue: that is the part people read.

The good version is three paragraphs and a list, in Slack or email, before the standup rather than after it. The bad version is a wall of card titles nobody reads twice. The difference is entirely in what you tell the agent to leave out.

A summary is also the safest thing an agent can do, because it only reads. Ask for it first and read four weeks of them before you let anything touch a card.

For the same job across a portfolio of boards, the guide to AI agents for project managers covers status reporting, and the Jira agent guide is the sharper contrast: same Atlassian account, very different board.

Chasing due dates

Somebody has to be the person who asks. On most teams it is the same person every time, and they hate it.

Internal chasing is safe to automate almost completely: a message to the card member two days before, a nudge on the day, an escalation to the board owner when it goes past. Nobody minds being reminded by a robot, and several people prefer it.

Client-facing chasing is a different matter, and this is the line I would not cross. The agent drafts, you send. A reminder that reaches a customer in the wrong tone on the wrong day costs more than the card was worth.

The practical trap is due dates that were never real. Half the overdue cards on most boards carry a date somebody typed to make a filter work.

Where a Trello agent breaks

Three failure modes, in the order you will meet them.

The card does not contain the situation. A card holds a title, a description, a list, labels, dates, members, checklists and comments. It does not hold the phone call where the client changed the scope, or the fact that the person assigned left in July. An agent reasoning over cards is reasoning over the shadow of the work, and it will be confidently wrong about anything that lives in the gap. Whatever you want it to decide has to be decidable from the board.

Permissions are blunt. An agent acts as whoever authorised it, with that person's access to every board they can see. Archiving is recoverable. A card dragged into a client-visible list at the wrong moment is recoverable too, technically, but not socially.

Nobody owns the seam. This is the one that actually kills these setups. A typical assembly runs Trello for the board, a workflow tool for the trigger, an AI vendor for the reasoning, and Slack for the delivery. Four companies, four bills, four status pages. When the Monday summary does not arrive there is no single place to look, and the person who built it has usually moved on. The chain has no owner, and a chain without an owner fails quietly.

Why the answers get vaguer as the board grows

Ask an agent to read a 40-card board and you get specifics. Ask it to read a 400-card board and you get generalities, because it is now summarising a summary. The instruction did not change. The ratio of input to output did.

The fix is not a smarter model. It is a narrower question: one list, one label, one date range, one output. An agent pointed at everything tells you nothing, at length.

Which should you start with?

Start with what you already own, because most people buy a second thing before finding out what the first one does. If your team is on Standard or Premium, you have already bought AI features. Spend an hour finding out what they cover before you add a vendor. At $5 or $10 per user a month, that spend is happening whether or not anyone uses it.

If the chore crosses Trello and exactly one other app you already pay for, Zapier is the boring right answer at $19.99 a month, or $0 on the free forever tier. If the work you want back is really your inbox and your calendar, and the board is a side effect, Lindy is built for that at $29.99 per user a month.

Procurement in the room changes the shape: Relevance AI fits, and you will learn the price on the call, because the pricing page does not publish one.

If the board is small and the chore is genuinely four minutes a week, do it yourself. I say that as someone selling the alternative.

The comparison people get wrong is treating $25 of Trello seats and a $20 plan as an either/or. They are not the same purchase. The Trello seats buy Trello. The honest question is marginal: what does the next tier up cost you, against what does handing the chore over cost you.

My own bias, stated plainly. When the chore repeats every week and nobody wants to own a workflow, I would pick the option where nobody builds anything. That is Gravity: first agent free with no card, then Pro at $5 a month during the alpha. You can buy more usage if you run out. It is priced as a plan, so the number holds when the sixth person joins the board.

Gravity is in private alpha, so here is the caveat. The answer only exists if an agent for your task has been published, and the profession-by-profession index is the fastest way to check.

FAQ

Sources