Atlassian is treating AI as a company-design reset

Atlassian did not describe AI as a feature release. It described AI as a reason to redesign the company.

In its March 2026 team update, Atlassian announced a reduction of about 10 percent of its staff, roughly 1,600 employees. Mike Cannon-Brookes wrote that the company was making the move to self-fund more investment in AI and enterprise sales, strengthen its financial profile, and reorganize around its System of Work so the company can move faster.

The painful fact sits beside strong operating numbers. Atlassian cited cloud revenue growth above 25 percent, RPO growth above 40 percent, more than 600 customers at $1 million or more in annual recurring revenue, and Rovo above 5 million monthly active users. This was not framed as a collapse response. It was framed as adaptation under a higher bar for software companies: growth, profit, speed, and value creation at the same time.

This note matters because it treats strength and restructuring as the same story.

Public AI coverage is still full of product language: copilots, agents, chat surfaces, and better ways to summarize, draft, search, and automate. Atlassian’s memo points at the deeper change. Once AI becomes part of the work itself, the company has to decide which skills still matter, which roles are still shaped correctly, which teams own revenue outcomes, and which parts of the operating system have to be rebuilt.

Cannon-Brookes rejects the simple line that AI replaces people. Then he says the sharper part: AI changes the mix of skills needed and the number of roles required in some areas.

A smarter sidebar is not the reset. Company design is.

A big software company can respond with a top-down redesign. It can reallocate capital, change leadership structure, narrow the skill mix, add enterprise sales capacity, and absorb the cost of a large restructuring. It can look at AI and ask which roles disappear, which teams should be rebuilt, and which revenue lines deserve more investment.

Whether that is humane, wise, or well executed is a separate question. Layoffs are not an operating insight. They are a human cost. Atlassian also named that cost by offering separation support, continued health coverage for eligible employees, mobility help, outplacement, visa support, and other transition assistance.

Do not read the transferable lesson as “cut staff and buy AI.” That is a big-company move under public-market pressure. The real lesson is that AI changes the unit of management.

For years, software was sold to departments. Jira for delivery. Confluence for knowledge. Service tools for support. CRMs for sales. Spreadsheets for the gaps. Work lived across tools, meetings, habits, policies, and private memory. AI makes that fragmentation harder to ignore because agents need usable context, clear standards, and defined authority. A model cannot safely improve a process that the business itself cannot name.

Atlassian’s System of Work language matters for the same reason. The winning layer is not another isolated app. It is the substrate that connects work, data, accountability, and automation. If the work system is legible, AI can help people move faster and make better decisions. If it is not, AI mainly makes confusion cheaper to produce.

SMBs will face the same pressure, but they do not have the same playbook.

A local services company, clinic, contractor, franchise group, or professional firm usually is not carrying thousands of redundant roles. It is carrying missing infrastructure. The owner knows the exception. The office lead knows the customer promise. The senior technician knows the workaround. The estimator knows which photos matter. The bookkeeper knows which invoice pattern is risky. The business works because experienced people hold the operating system in their heads.

AI does not automatically turn that private memory into a company system. It can expose the gap faster.

A copilot can draft the customer reply, but it still needs the warranty rule. An agent can classify the service call, but it still needs the exception boundary. A quoting assistant can assemble an estimate, but it still needs the margin rule, the photo standard, and the handoff owner. A sales helper can prepare follow-up, but it still needs the buyer promise and the no-fit signal.

This is where the SMB response has to diverge from the enterprise response.

Enterprise responses often start with compression: fewer roles, more automation, tighter reporting, more capital behind the highest-return teams.

SMB responses should start with conversion: turn domain experts into owner-operators by giving them the operating substrate they were never handed.

Start with workflows with named owners. Make rules inspectable. Tie customer promises to evidence. Build apprenticeship paths that teach judgment, not only task execution. Keep scoreboards that show whether the business is improving. Put agents inside accountable boundaries instead of floating them above a messy process.

Hadto’s ontology-as-ops thesis sits here. An ontology is not only a data model. Used correctly, it is a business contract: what the company means by a job, a customer promise, a handoff, a defect, a ready estimate, a closed ticket, a qualified buyer, a billable exception. Once those meanings become explicit, people and agents can work inside the same operating language.

That is how AI can create owners instead of only compressing labor.

The Atlassian note is a signal from the public-company frontier. The pressure is moving from “which AI features did you ship?” to “how did AI change the way the company runs?”

SMB owners should hear the warning without copying the move. Do not start with headcount math. Start with the operating system. Name the work, the standard, and the owner. Make the private judgment teachable. Then use AI to expand the number of people who can run the business well.

Source: Mike Cannon-Brookes, An important update on our team, Atlassian, March 2026.