Developer Tools Under Agents: What Atlassian, GitLab and JetBrains Disclose

Atlassian's cloud revenue grew 31% in its June 2026 quarter. GitLab grew 21% and guided its next quarter below its last. The developer tools market is splitting. Tools that coding agents write into are gaining, and tools that sell a seat for AI help are moving to credits.

Short answer

Agents help the developer tools that act as a system of record and pressure those that sell a seat for assistance. Atlassian's filings show cloud seat expansion alongside rising hosting costs. GitLab's show a move from seats to credits. JetBrains publishes no financials, only price pages and a survey.

What the filings disclose about developer tools

Three companies make the products a typical engineering team touches between a ticket and a shipped change. Atlassian runs the issue tracker. GitLab runs the repository and the pipeline. JetBrains runs the editor. Coding agents now operate in all 3 places, so these 3 sets of disclosures are the cleanest available read on the developer tools market.

Atlassian's figures come from its fourth quarter fiscal 2026 shareholder letter and press release, both filed with the SEC on 6 August 2026. GitLab's come from its second quarter fiscal 2027 release of 1 September 2026 and its published pricing pages. I could not open the original Business Wire copy of the GitLab release, so I read a syndicated copy and say so in the references.

JetBrains does not file. The company describes itself as fully bootstrapped, and I found no financial statements for it. For JetBrains the evidence is a price page and a survey the company runs itself. Those are weaker grades of evidence than an SEC filing, and the comparison table below keeps them visibly apart.

The rule for this post is simple. Every number is a figure the company or its survey published. Where a vendor has not disclosed something, the table says "not disclosed" rather than filling the gap with an estimate. Adoption share for individual coding assistants is covered elsewhere on this site in the Claude Code, Cursor and Copilot verdict, so this post stays on the vendors' own numbers.

Atlassian: cloud growth with agents routed through Jira

Atlassian reported $1,766 million of total revenue for the quarter ended 30 June 2026, up 28%. Cloud revenue was $1,213 million, up 31%. Subscription ARR reached $6,606 million, up 23% from $5,382 million a year earlier.

Cloud seats are growing, Data Center seats are not

The shareholder letter credits cloud growth to "strong seat expansion in core Jira and Confluence", along with cross-sell of its Service Collection and Teamwork Collection bundles. That is a statement about direction. The letter gives no seat count, no seat growth rate and no net revenue retention figure.

The self-hosted product tells a different story. Atlassian announced in September 2025 that it will end of life Data Center in March 2029. For FY27 it guides Data Center revenue down about 17%, and one of its stated reasons is "muted seat expansion" among customers preparing to move.

That guidance also sets a ceiling on how much of the cloud number is new demand. Management expects cloud migrations to contribute mid-to-high single digits of cloud revenue growth in FY27. Some of the cloud growth is Data Center customers changing address, not new seats created by the AI era.

The MCP server turns Jira into something agents write to

The most useful disclosure in the letter is about usage, not revenue. MAU of the MCP server and Teamwork Graph CLI more than doubled in the quarter to pass 1 million. Overall MCP calls rose more than 400% quarter over quarter. Work items and Confluence pages generated through MCP rose nearly 4x.

One figure matters more than the rest. 98% of MCP users were also active in the Jira interface in the same month, per the letter. Agents are not replacing the people who use Jira. They are adding a second kind of user to the same system.

Atlassian describes the workflow directly: "Assign a Jira issue to Claude or Cursor and the agent reads it, accesses the repo, and opens a draft PR." It calls Jira the control plane. I think that framing is correct and commercially smart, because it makes the issue tracker the place where agent work is assigned, audited and measured. The wider protocol story is in our piece on the MCP interoperability standard.

The gaps matter too. Atlassian does not disclose Rovo monthly active users, the share of AI usage that is paid versus included, or any AI revenue line. It also does not say whether an agent assigned a ticket counts as a billable user. That last question will decide how much of agent activity becomes revenue.

GitLab: from seat add-on to metered credits

GitLab reported $286.3 million of revenue for the quarter ended 31 July 2026, up 21%. Dollar-based net retention was 117%. Customers above $100,000 in ARR rose 17% to 1,571. The CEO described record gross bookings and net ARR growth above 40%, though the release gives no dollar value for either.

The guidance is the sharper signal. GitLab guides Q3 revenue to $281 million to $283 million. That sits below the $286.3 million just reported, and the midpoint is roughly 1.5% lower. Full-year FY27 revenue is guided to $1,129 million to $1,133 million.

What GitLab Credits actually price

GitLab announced general availability of Duo Agent Platform on 15 January 2026. Premium and Ultimate subscribers automatically receive $12 and $24 of included credits per user each month. GitLab says these are available for a limited time and subject to change.

Beyond the included amount, each credit has an on-demand list price of $1, billed monthly with no commitment. Credits are pooled across the organisation, so heavy users draw on the allowance of light ones. Enterprise customers on annual commitments get volume discounts, though GitLab publishes no percentages.

The older seat add-ons are being folded in. GitLab says that on upgrade it will "roll forward your investment in seats" for Duo Pro and Duo Enterprise into credits. The AI product is no longer a per-seat line item. It is a balance that gets spent.

The arithmetic behind a credit, and the margin cost

GitLab's credits post states that one credit buys 2 model requests for Claude Sonnet 4.5, or 20 requests for small models such as GPT-5 mini. By simple arithmetic, the $12 Premium allowance buys about 24 Sonnet-class requests per user per month. That is my calculation from list terms, not a measure of typical usage. It shows how quickly agent loops, which make many requests per task, can consume an allowance. The billing logic is explored in our comparison of bundling and metering.

The margin line shows the cost of running these features. GitLab's GAAP gross margin was 84% against 88% a year ago, and non-GAAP was 86% against 90%. Adjusted free cash flow was $9.8 million against $46.5 million. A restructuring charge of $19.4 million sat inside the quarter.

GitLab also cites a Forrester Total Economic Impact study showing a 400% return on Duo Agent Platform. Treat it as marketing evidence. The study is commissioned by the vendor, and the release gives no sample size or customer mix. The release does not report Duo Agent Platform ARR, credit consumption or customer counts.

JetBrains: no filings, but a price page and a survey

JetBrains sits outside the filing comparison, yet it is the vendor closest to the developer's hands. Its two public artefacts are an AI pricing page and a company-run survey. Both are useful, and neither is audited.

The quota is priced at the sticker

JetBrains moved in August 2025 to a model where "each subscription's quota now matches its price." One AI credit is worth $1. Its AI Assistant help page, last dated 22 April 2026, lists AI Free at 3 credits per 30 days, AI Pro at $10 with 10 credits and AI Ultimate at $30 with 35 credits. Top-ups last 12 months.

The page prints a second table with $20 for Pro, $60 for Ultimate and a $60 Enterprise tier, and it does not label either table as monthly, annual, personal or organisational. I treat the first table as the one I can read cleanly and flag the second as unresolved. Junie, JetBrains' coding agent, shares the same credit system as AI Assistant, per the August 2025 announcement. So an agent session spends the same balance as a chat question.

What the developer survey says

JetBrains' Developer Ecosystem Survey 2026 covers more than 15,000 professional developers. It reports that 90% use AI coding agents at work at least weekly and 68% use them daily. Claude Code reaches around 39% of developers, GitHub Copilot 21%, Codex 16% and Cursor 12%. JetBrains AI and Junie together reach around 9%.

Read those figures with the sponsor in mind. JetBrains runs the survey, sells competing AI products and reweights its sample by familiarity with its own products. It is useful for direction and weak as a market-share audit. Note also that 39% of Copilot users say they use it in JetBrains IDEs, which suggests the editor is still a surface other vendors' agents want to reach. For the broader picture on how developers say they feel about these tools, see the gap between developer AI usage and trust.

Side-by-side: disclosed metrics for three developer tools

The table below lines up what each vendor has actually published. Cells that say "not disclosed" are findings, not omissions. They mark where a reader would need an estimate to say anything more.

Sources: Atlassian Q4 FY26 shareholder letter and press release, GitLab Q2 FY27 release and credits pages, JetBrains help page and survey. Periods differ by one month and one quarter-end, so growth rates are comparable in direction, not to the decimal.

Gross margin: agents cost money to host

Revenue growth tells you whether customers are buying. Gross margin tells you what it costs to deliver what they bought. For developer tools, that second number now carries the AI bill.

Atlassian guides FY27 non-GAAP gross margin to 86.5%, and its guidance language cites "the negative impacts of growing Rovo usage and hosting costs." Its non-GAAP gross margin in the June quarter was 89%. That comparison mixes one quarter with a full-year guide, so treat the 2.5 point gap as a direction, not a measurement.

GitLab offers a cleaner comparison. Non-GAAP gross margin was 86% in the July 2026 quarter against 90% a year earlier. That is 4 points of margin lost over 12 months at a company growing revenue 21%. The mechanics of that squeeze are covered in our analysis of real AI gross margin.

Which developer tools gain from agents and which get bypassed

Put the disclosures together and a pattern shows. Agents add value where they need a durable record to work against. They reduce value where the product's pitch is that a human gets help.

The record that agents write to gains

An issue tracker, a repository and a pipeline are all places where an agent's work has to land, be reviewed and be audited. Atlassian's 98% overlap between MCP users and Jira users says agents are pulling more activity into the tracker, not routing around it. I would bet on systems of record in this cycle, because they hold the context an agent cannot invent.

GitLab's position is plausible on the same logic, though its disclosures are thinner. It reports net retention of 117% and says the platform is attached to larger deals, but it does not report agent usage numbers in its release. The cleanest evidence for GitLab is its credits model, which treats agent activity as billable consumption.

The seat sold for assistance is the exposed product

The editor-side assistant is the product most exposed to bypass. JetBrains' own survey puts JetBrains AI and Junie at around 9% of developers, against around 39% for Claude Code. GitHub Copilot's reach fell from 29% to 21% over a year in the same survey. Standalone terminal and cloud agents do not need an IDE seat at all. The decision to build or buy that layer is analysed in our build-versus-buy guide for coding agents.

Pricing is converging on the same unit. GitLab and JetBrains both sell AI as $1 credits. Atlassian ships AI credits inside its collections but publishes no credit price. Each design makes the vendor's costs variable and the buyer's bill variable too, which is what our analysis of seat compression predicts when a seat stops being the unit of value.

Where this argument is weakest

Every post about vendor positioning is tempting to over-read. This one has 4 real weaknesses.

First, the missing seat counts. Neither Atlassian nor GitLab reports total seats, so "seat expansion" is a management phrase and not a measurement. I cannot show that seats are compressing or holding. I can only show what each vendor chose to publish, and that choice is itself informative.

Second, Atlassian's cloud growth includes migration. Management expects migrations to add mid-to-high single digits to FY27 cloud growth, so some of the 31% is existing customers moving, not agents creating demand. A skeptic can fairly say the agent story is a small share of that number. The filings do not let anyone size it.

Third, GitLab's soft Q3 guide has more than one possible cause. It may reflect a shift to consumption pricing, where revenue arrives later than a seat contract would. It may also reflect competition from standalone agents. The release does not separate the two, and I will not guess.

Fourth, the survey is vendor-run. JetBrains reweights by familiarity with its own products, and adoption is not spend. A tool used by 39% of developers can still produce less revenue than a tool used by 12% on enterprise contracts. The case against my position is that distribution and price will matter more than usage.

Frequently asked questions

Where to start this week

Open your own renewal list and find every developer tool that charges a seat for AI. For each, write down the unit the vendor will move to, whether it is a credit, a request or an outcome, and whether the price is published. Any vendor that cannot name the unit is a budgeting risk.

Then pick 1 agent workflow your team already runs, such as an agent opening pull requests from tickets. Count the model requests it uses in a week and multiply by the vendor's credit rate. That gives you a real monthly figure to set against the $12 and $24 allowances, instead of a vendor's average.

Related reading