Agentic Layer Check
One of the cleanest outcome based AI meters we've seen from a major platform vendor sits right next to a packaging story that's starting to fragment into three separate mechanics. Here's how it reads against the nine layers.
What they sell
A connected work platform (Jira, Confluence, and the Rovo AI layer) sold to software and IT teams, then expanding into every function that plans and tracks work.
Who they sell to
Per seat licensing across teams of every size, with the enterprise motion and the "platform company" pitch increasingly aimed at whole organisations, not just engineering.
How they price
Per user tiers (Free, Standard, Premium, Enterprise) layered with two separate AI usage meters: Rovo credits, and a new per resolution charge for autonomous agent outcomes, effective 3 December 2026.
Two moves here are genuinely worth other vendors studying, not just noting.
The new AI agent resolutions charge only bills when an agent autonomously closes a request without a human stepping in: that's an outcome unit, not a proxy like seats or API calls. It's a concrete reference point for anyone still metering AI by token or by seat: price what the agent actually delivers, not what it consumes along the way.
Inline summaries, rewrites and Rovo Search stay free while the higher stakes capabilities (the coding agent, autonomous resolution) sit behind the meter. That's a sensible adoption sequence: let people build trust in the low stakes AI before asking them to pay for the high stakes AI, rather than metering everything from day one and slowing adoption before it starts.
The opportunity isn't in any one mechanic: it's in what happens when a buyer has to hold all three at once.
A buyer on a paid plan is now reconciling three separate mechanics to understand their bill: the per seat tier that markets "AI agents built in," Rovo credits metering the deep AI interactions, and the new per resolution charge for autonomous outcomes. Each is defensible on its own. Together, they ask the buyer to hold three different mental models of what they're actually paying for: right as "one number, one outcome" is becoming the standard other agentic vendors are starting to compete on.
That packaging fragmentation doesn't stay contained to Layer 3. It shows up two layers over. Layer 8 (Revenue Architecture) inherits three separate lines to recognise and forecast instead of one clean usage number: harder to report as a single "AI driven revenue" figure on an earnings call, and harder for a customer's own finance team to reconcile against a budget. Layer 4 (Sales Strategy & Enablement) inherits the harder conversation: a rep now has to walk a buyer through seat tier, Rovo credits, and resolution billing in the same deal, instead of one value metric a customer can hold in their head. Neither cost shows up on the pricing page itself: both show up in the deal cycle and the invoice, which is exactly the kind of cross layer effect that's easy to miss if you're only looking at the layer where the change was announced.
Part of the Agentic Layer Check series: named companies, read against the nine layers. Want your own read? Request a teardown →
This is a public signal read: Atlassian's own pricing pages and its own announcement of expanded usage based pricing, dated September 2026, not a private briefing or company input. It reflects what's publicly documented as of the date above: pricing pages change, and this piece will note it if Atlassian's does. Atlassian was not consulted on or paid for inclusion; see the disclosure approach this work is run under.
Sources: Atlassian: expanded usage based pricing → · Atlassian: Jira pricing →