Atlassian starts metering Rovo, automation and agents on 3 December

What Atlassian announced

Atlassian published the change on 1 September 2026. Three meters start billing on 3 December.

• Rovo credits, covering deep AI interactions: Rovo Chat, Confluence AI Slides, the Jira Coding Agent and Teamwork Graph queries.
• Automation steps, meaning the execution steps inside an automation rule rather than the rule itself.
• AI agent resolutions, meaning an autonomous completion that finished without escalating to a human.

The framing in the announcement is that eligible paid cloud plans include substantial allowances and that organisations only see charges when usage exceeds them. That is accurate, and it is also the part worth checking rather than accepting, because whether an allowance is substantial depends entirely on what a single interaction costs against it.

What is not metered

Rovo Search inside Atlassian apps, definitions and summaries remain free, as do in-line rewrites. This matters more than it sounds. The features most people touch daily are the ones that stay unmetered, which means the meter is aimed at the deeper and more expensive interactions rather than at ordinary assistance. A team that uses Rovo to find things is largely unaffected. A team that uses it to do things is not.

Bitbucket allowances are also moving to organisation-level pooling, which is the same direction of travel as the Rovo allowances described below.

Why the notice period is the story

Thirteen weeks between announcement and effect is more than most vendors give, and it is the part of this that a buyer can actually use. A price rise announced with thirty days notice leaves time to complain and nothing else. A meter announced with a quarter's notice leaves time to measure consumption, decide a policy and adjust plans before the first billed month.

It also means the comparison most people will make in January is the wrong one. The question is not what December cost against November, because November is unmetered and free. It is what December cost against the allowance, which is the only number that recurs.

The allowances, per user per month

Rovo credit allowances are granted per user per month and pooled at the organisation level. Atlassian publishes them by product and plan.

Two properties of the allowance decide how it behaves in practice, and both are stated plainly by Atlassian. Allowances refresh monthly and do not roll over, so an underused January does nothing for a heavy February. And they pool at the organisation level, which is the saving grace: the allowance is not per person in the sense that a light user wastes theirs, it is a shared pot that a light user contributes to.

Pooling is the single most important detail for anyone estimating exposure. A 200-seat Jira Standard organisation holds 5,000 credits a month between everyone, not 25 each. Whether that is comfortable depends on the concentration of use, and in most organisations AI features are used intensely by a minority, which pooling handles well.

What 25 credits actually buys

An allowance is meaningless without the price of an interaction beside it. Atlassian publishes those too.

Ten credits for a basic event against a 25 credit Standard allowance is the number that should shape planning. Divided out, a Jira Standard seat funds two and a half basic interactions a month. Premium funds seven. Enterprise funds fifteen.

Stated per seat that sounds severe, and pooling softens it considerably. Stated across an organisation it is more useful: 200 Standard seats fund 500 basic events a month between them. If forty people use a basic Rovo feature twice a week, that is roughly 320 events and the pot holds. If the same forty use it daily, it is about 800 events and the organisation is into overage in the third week.

The variable tier is the harder one to model

Premium intelligence is quoted as variable credits per billable event rather than a fixed figure, which means advanced agents, Think Deeper and Confluence slides cannot be budgeted from the price list alone. This is not unusual for an AI meter and it is not a criticism, but it does mean the only reliable estimate comes from watching actual consumption rather than from arithmetic done in advance.

That is an argument for turning the visibility on early rather than waiting for December.

Three organisations, same plan

Pooling means the only figure that matters is total consumption against a total allowance, so the useful question is what proportion of seats use the features and how often. Three shapes of the same 200-seat Jira Standard organisation, which holds 5,000 credits a month.

The last row is a 200-seat organisation where half the staff use a basic AI feature every working day, and it costs $150 a month over the allowance. That is worth stating plainly because the per-seat arithmetic makes the allowance sound tighter than the pooled reality usually is. The exposure is real but it is not, at these rates, dramatic.

What changes the picture is the variable tier. Every figure above assumes basic events at a known 10 credits. Advanced agents, Think Deeper and Confluence slides are quoted as variable, so an organisation whose use skews to those features cannot produce this table for itself and has to measure instead.

What overage costs, and what happens if you refuse it

Extra usage lists at $0.01 per Rovo credit, or $10 per 1,000 credits.

The behaviour at the limit depends on a setting, and the two branches are genuinely different operationally.

With extra usage enabled, metered features carry on working without interruption and the charges appear on the next invoice. Nothing breaks, and the first signal is financial rather than functional.

With extra usage disabled, billable interactions stop until the allowance resets on the first of the following month. Nothing is billed, and a feature people had been using simply stops mid-month.

Neither is the safe choice in the abstract. A support desk running AI agent resolutions would rather pay than have resolutions stop halfway through a month. A team using Rovo for convenience would rather it stopped than accrue an unbudgeted invoice. The decision belongs to whoever owns the outcome, which is the argument for making it deliberately before December rather than discovering the default in January.

There is a third position worth naming, which is to enable extra usage and set a limit rather than leave it open. Atlassian gives admins the ability to set and adjust limits and to receive alerts before one is approached, which turns the binary into a ceiling. That is the arrangement most finance functions would recognise as normal, and it is the one least likely to produce either a stopped feature or a surprising invoice.

The arithmetic of an overage

At $10 per 1,000 credits, a hundred basic events beyond the allowance costs $10. A thousand costs $100. These are not alarming numbers at small scale, which is exactly why they are worth watching: a meter that costs little enough to ignore per incident is the kind that accumulates quietly across a year.

What to do before 3 December

Thirteen weeks is enough time to answer the questions cheaply rather than expensively.

Turn on the visibility now

Atlassian gives organisation and billing admins real-time usage visibility across teams, the ability to set and adjust limits, and alerts before a limit is approached. Usage between now and December is not billed, which makes this a free measurement period. An organisation that watches November consumption knows its December exposure. One that does not will learn it from an invoice.

Decide the extra-usage setting deliberately

Pick the branch that matches the consequence. Where an interruption costs more than an invoice, enable it. Where an unbudgeted invoice is the worse outcome, disable it and accept that features stop. Write down which was chosen and why, because the person who receives the January invoice is often not the person who set it.

Audit automation steps separately

Automation is the meter most likely to surprise, because steps accumulate from rules that were written once and forgotten. A rule that fires on every issue transition in a busy project generates steps continuously with nobody watching. Reviewing which rules fire most often is worth doing before the meter starts, and it is worth doing on its own merits regardless.

Check the plan against the allowance, not against the feature list

The gap between Standard at 25 credits and Premium at 70 is nearly threefold, and the gap to Enterprise at 150 is sixfold. If AI features are genuinely central to how a team works, the allowance is now a real input to the plan decision rather than a footnote to it. If they are not, the free tier of Rovo Search, definitions and summaries covers a great deal and the meter never engages.

Where this sits

This is the same movement visible across enterprise software this year rather than an Atlassian peculiarity. Per-seat pricing prices access to a tool. Metered pricing prices what the tool does, and AI features make the second far more expensive to supply than the first.

The pattern is consistent: an assistive layer arrives free, use becomes habitual, and a meter is introduced with an allowance that covers ordinary use and charges for concentrated use. Judged against that pattern Atlassian's version is comparatively transparent, with published allowances, a published credit price, a documented free tier and thirteen weeks of notice.

What it shares with the rest is that the cost becomes variable, and a variable cost needs an owner. The organisations that handle this well will be the ones that decided in October who watches the meter, rather than the ones that find out in January that nobody was.

Why per-seat stopped working for AI

A seat licence works when the marginal cost of one more action is close to nothing, which is true of loading a page and false of running a model. Under per-seat pricing the heaviest user of an AI feature costs the vendor many times what the lightest one does while paying the same, and the vendor absorbs the difference. That is sustainable while adoption is low and stops being sustainable at exactly the point the vendor wanted to reach.

Metering resolves it by moving the variable cost to the customer who generates it. The consequence for a buyer is that a line item which used to be predictable a year ahead now has a floor and no ceiling, and the ceiling has to be imposed by whoever holds the budget.

What to watch after December

Two things will tell you whether this settles well. The first is whether the free tier stays free: Rovo Search, definitions and summaries are the features most people actually touch, and their being unmetered is what keeps this a targeted meter rather than a general price rise. The second is whether allowances move. An allowance is a number in a table, not a contractual commitment, and the same table can be republished with smaller figures.

Neither is a prediction. Both are cheap to check periodically, and checking is the difference between noticing a change and being surprised by one.

Questions people ask

When does Atlassian usage-based billing start?

3 December 2026 for most meters. Atlassian announced the change on 1 September 2026. Usage before that date is not billed, which makes the intervening period a free measurement window.

How many Rovo credits do I get?

Per user per month, pooled across the organisation. Jira and Confluence give 25 on Standard, 70 on Premium and 150 on Enterprise. The Service and Teamwork Collections give 250, 700 and 1,500 on the same tiers. Allowances refresh monthly and do not roll over.

What does a Rovo interaction cost in credits?

A basic intelligence event, such as a quick answer or a basic agent, costs 10 credits. Premium intelligence features including advanced agents, Think Deeper and Confluence slides are quoted as variable. Enriched Teamwork Graph API calls typically fall between 1 and 10 credits. Rovo Search, definitions and summaries are free.

What happens when the allowance runs out?

It depends on your extra usage setting. Enabled, metered features keep running and the charges appear on your next invoice at $0.01 per credit. Disabled, billable interactions stop until the allowance resets on the first of the following month.

Which Atlassian AI features stay free?

Rovo Search inside Atlassian apps, definitions, summaries and in-line rewrites. The meter is aimed at deep interactions such as Rovo Chat, Confluence AI Slides, the Jira Coding Agent and Teamwork Graph queries rather than at everyday assistance.

Is automation billed per rule or per step?

Per step, meaning the execution steps inside a rule rather than the rule itself. A single rule with several actions consumes several steps every time it fires, which is why a frequently triggered rule is the most likely source of an unexpected figure.

Sources

Every price, limit and date above was checked against these pages on the day shown. Where a figure has since moved, the vendor’s own page is the authority and this one is a snapshot.

1. 1Atlassian's usage-based pricing: AI value with predictability and control · AtlassianThe 1 September announcement, the 3 December effective date, the three meters and what remains free.Checked 4 September 2026
2. 2How Rovo credits work · AtlassianAllowances by product and plan, the 10 credit basic event, the $0.01 per credit overage price and the behaviour at the limit.Checked 4 September 2026
3. 3Usage-based pricing: visibility and control at scale · AtlassianThe usage and pricing guide the announcement refers customers to for allowance figures.Checked 4 September 2026