SYDNEY, 9 AUG 2026 — Atlassian's shares rose by around a third after fourth-quarter results that beat expectations on every line. The market was not repricing a good quarter. It was repricing a thesis.
That thesis is the one hanging over every seat-based software company: that AI agents will do the work the seats were bought for, and the seats will go. Atlassian's numbers are the first substantial evidence against it — and they are one quarter, from one company.
The quarter
Revenue of $1.77 billion beat the $1.66 billion expected, and adjusted earnings of $1.87 a share beat $1.50. Cloud revenue, the line investors actually watch, accelerated to 31% growth.
The key word is acceleration. Growth rates at this scale normally decay; a cloud line speeding up in the quarter when the market had priced structural decline is why the reaction was violent rather than merely positive.
The number that is not the share price
Remaining performance obligations rose 44% to $4.8 billion.
RPO is contracted revenue not yet recognised — work customers have committed to pay for. It is the least fashionable figure in a software release and the most informative, because it is forward-looking and much harder to flatter than a quarter's revenue. A company can have one good quarter for reasons that do not repeat. It cannot book 44% more future obligation unless customers signed.
Growing RPO faster than revenue means the commitments being made now are larger than the ones being consumed. According to the "AI eats SaaS" thesis, RPO should have been the first number to deteriorate. Instead, it accelerated.
Where the growth came from matters more than that it happened
Customers spending more than $3 million a year grew by over 50%, to 164 accounts. Those above $5 million grew by more than 70%, to 69.
Read those two together and the expansion is concentrated at the top. This is not a story about many small new customers; it is large enterprises materially increasing what they spend with an existing vendor.
That specifically contradicts the mechanism the bear case relies on. The bear case was never about companies cancelling software outright. It was that AI would reduce headcount, causing seat-based licences to shrink in turn. Enterprises are instead buying more, and buying more at the accounts with the most sophisticated procurement and the most exposure to AI-driven efficiency.
What the bear case actually claimed
It is worth restating the thesis precisely, because a vague version of it is unfalsifiable and the precise one is now under test.
The argument was mechanical rather than mystical. Seat-based software is priced per person doing a job. If AI agents do a meaningful share of that job, the buyer needs fewer people, and a licence sold per person shrinks with the headcount — not because the software got worse, but because the unit it is priced in evaporates. On that logic the damage shows up first in net expansion at existing accounts, then in new-seat growth, and only much later in revenue.
This is why the customer-tier figures matter more than the revenue beat. The thesis predicts existing large accounts flattening or shrinking first. Accounts above $5 million grew more than 70%.
The thesis also has a timeline problem that cuts both ways. Enterprise licence agreements are multi-year, so a company reducing headcount today may not be able to reduce seats until renewal — which means a quarter of strong expansion could be measuring contracts signed before the question arose. RPO growing 44% works against that reading, since those are new commitments, but it does not eliminate it.
What the result does not settle
Three cautions, and they are not small.
It is one quarter from one company, in a category — developer and project tooling — where AI adoption arguably increases the amount of software work being tracked rather than reducing it. More code written faster is more tickets, more reviews, more coordination. A company selling the coordination layer may be the least exposed seat-based business in the market, not a representative one.
The company's own commentary attributed momentum to AI-driven products, which cuts both ways. If new AI features are driving growth, then this is a story about Atlassian successfully monetising the transition, not about the whole category surviving it.
And the share move itself is not evidence about the business. A stock rising a third in a session says the market was positioned for something considerably worse; the jump measures the surprise, not the quality of the outcome itself.
Reading it if you buy software rather than shares
The useful takeaway for a technology buyer is about your own renewal conversations rather than anyone's stock.
If your vendor's growth is coming from enterprises expanding, the pricing leverage in your next negotiation is worse than the AI narrative may have led you to expect. The assumption that seat-based vendors would be defensive going into 2027 is, on this evidence, not universally safe.
Check whether your own seat counts are behaving as the thesis predicts. Most organisations have not measured whether AI adoption reduced their tooling seats, and the answer is knowable from your own licence data rather than from an earnings call.
What to watch
Whether the cloud acceleration holds for a second quarter, since one datapoint is a datapoint. Whether other seat-based vendors report the same top-end expansion or whether Atlassian's is specific to development tooling. And whether RPO keeps outgrowing revenue, which is the earliest place the thesis would show up if it were true after all.
View author profile → · Editorial policy
