Mike Cannon-Brookes on AI Valuations, Software Design and Building Atlassian to Last

Mike Cannon-Brookes, co-founder and CEO of Atlassian, joins Harry Stebbings to discuss building through technological change. He describes himself as “unreasonable”: when something does not work as he thinks it should, his instinct is to change it rather than accept the boundary.

Cannon-Brookes says people drawn to technology often see creative combinations that do not yet exist and try to build them. Some work and some do not. At Atlassian, he sees opportunities to make better technology amid changes in AI, energy and the climate.

Looking back on more than two decades as co-CEOs with Scott Farquhar, Cannon-Brookes says Atlassian would not be where it is without the partnership. They began at similar stages of life and with comparable experience, which made it easier to learn, share the pressure and balance family life.

He also thinks it helps when each partner believes the other is better at the job: both stretch to meet a high standard. Their friendship mattered too. Over a long journey with inevitable ups and downs, being able to laugh and enjoy each other’s company made the work sustainable.

For new co-CEOs, Cannon-Brookes recommends openness about real and perceived conflicts, a balance between partners and clear areas of responsibility. He estimates that his and Farquhar’s roles overlapped by 60–80%: enough shared context to agree on goals, but room for each to lead.

The responsibilities changed over time. Cannon-Brookes says he led sales and marketing for roughly 13 years before Farquhar took it on for four, and each at times covered the whole business when the other was away. Neither a complete overlap nor no overlap would have worked as well, in his view.

Their first shareholder agreement included a best-of-three rock-paper-scissors clause for unresolved disagreements, although they never needed it. Cannon-Brookes says disagreements about acquisitions or strategy usually came back to persuasion: if he could not convince Farquhar an idea was good, perhaps it was not worth doing.

He recalls an early security incident that required Farquhar to be notified while on honeymoon in a remote part of Africa. A radio message eventually reached him. For Cannon-Brookes, the episode also illustrates the value of having a second founder who can step in.

Stebbings asks whether Cannon-Brookes’s period as sole CEO represents an aggressive exercise in “founder mode.” Cannon-Brookes disagrees: there was no deliberate plan to make bold moves simply to signal a new era, and many decisions had been planned well before the leadership change.

Stebbings points to acquisitions and sports investment as examples of major decisions. Cannon-Brookes says some were long in the making, and rejects the idea that Farquhar’s departure freed him to pursue ideas previously blocked. He still calls his co-founder for advice.

The AI era, he says, demands activity, investment and choices from technology companies. He expects people to look back on this period as chaotic: some widely anticipated ideas may come to nothing, while unforeseen developments become large. That is his prediction rather than a claim to know the winners.

Cannon-Brookes describes Atlassian’s earlier acquisition of Loom as a substantial bet and says it has performed well. As Atlassian’s CEO, he has a direct commercial interest in Loom and in the acquisitions and products he discusses.

Asked how an investor should judge sustainability, he says many assets are probably vastly overvalued, while a few may be undervalued enough to outweigh the failures. He invokes the dot-com era and Amazon to explain why picking those few is difficult; he does not identify a dependable valuation formula.

He illustrates the uncertainty with hypothetical money flows: an AI company pays a model provider $100 million, the model provider pays a cloud provider $150 million, and the cloud provider spends $200 million on chips. Revenue appears at each stage while each loses money in the example. These are illustrative numbers, not reported transactions.

Cannon-Brookes believes AI is foundational but says its durable business models and ultimate distribution of economic value are unknown. Acknowledging uncertainty is compatible with believing the technology creates real value; it just makes investment decisions harder.

Atlassian responds by making explicit working assumptions about the next several years and revisiting them roughly every quarter. Cannon-Brookes argues that a business must have a view and be willing to abandon it when evidence changes, rather than waiting for complete certainty.

He also recommends talking deeply to enough of the right customers. In a noisy period, customers can help an organization develop conviction about what is actually changing, even as it continues to take risks and make strategic bets.

One Atlassian bet was that multiple foundation-model providers would compete. The company deliberately built for multiple models rather than training its own, and invested in testing, selecting and deploying new models quickly. Cannon-Brookes thinks the choice is working, while cautioning that it is early to declare any AI thesis proven.

A second bet is on design in the broad sense: how products work and feel, not merely their colors. Cannon-Brookes expects fundamental design to matter when technology changes. He cites pull-to-refresh on phones as an interaction that emerged with a new platform and spread because it worked.

He says Atlassian has expanded its design capability to help customers use AI without understanding the technical distinctions between probabilistic and deterministic systems. Users want to accomplish something; making that experience coherent is part of the company’s product strategy.

Stebbings asks about the argument that business applications will collapse into agents. Cannon-Brookes is skeptical of a single universal assistant replacing everything. Email and spreadsheets can reproduce many processes in principle, he says, yet specialized HR, CRM and project-management tools exist because they serve specific needs better.

He expects AI to change applications fundamentally and says some may disappear while others benefit, but does not think all applications reduce to interchangeable databases and business logic. His position should be read in light of Atlassian’s own business selling enterprise applications.

On future interfaces, Cannon-Brookes thinks design will be essential but admits nobody yet knows the answer. AI could theoretically generate a different interface for every screen, though he doubts constantly changing controls would work well for users who value stability and familiarity.

The practical result may combine conventional interfaces with prompts. A button or slider can be more convenient than writing a long instruction, while a prompt might let a user customize a complicated interface to their role. He presents this as a possibility, not a settled product pattern.

If software becomes cheaper to create, Cannon-Brookes expects good design to become more valuable because it remains scarce. He is optimistic that more technology will be built and that lower creation costs could let teams attempt a solution several times instead of settling for the first adequate version.

He predicts Atlassian and AWS will employ more software developers in five years, not fewer, while those developers become more efficient. His reasoning is that businesses have more ideas than engineering capacity; the prediction is not evidence that every developer role or company will grow.

Asked whether Atlassian will create a vibe-coding environment for enterprises, he says the company is launching one. He believes more people will create technology without necessarily calling themselves software engineers, provided their output meets the relevant design, security and compliance standards.

A finance professional might build a small application or write analysis code instead of filing a request or wrestling with a spreadsheet. Cannon-Brookes expects more generalists to use such tools, while core technologists continue building platforms and more complex systems.

Atlassian wants developers, app vendors and customers to create applications on its platform at lower cost, he says. Some might use code for sophisticated products; others might generate a simple app with prompts and discard it after use. This is a description of Atlassian’s own commercial direction.

On entry-level engineers, Cannon-Brookes predicts there will be more, and says Atlassian is hiring more graduates than in the previous two years. He suggests AI tools may amplify both exceptional and average engineers, but the 10x and 100x comparisons he uses are hypothetical rather than measured productivity results.

He believes graduates can introduce new habits that change a workplace. He cites Loom, an Atlassian product, as an example: younger employees familiar with video and voice messaging may bring those tools into organizations and encourage colleagues to communicate differently.

Atlassian engineers use its own Rovo tools alongside Cursor, GitHub Copilot and others, he says. He would probably not pay ten times the price for the current tools unless productivity justified it, partly because he believes competing tools make switching possible.

Different tools have different strengths. Some work well for starting a new project but less well with a large existing codebase or changes across hundreds of repositories. Cannon-Brookes expects engineers to choose based on the task and their preferences.

Writing code is only one part of an engineer’s work, he says; meetings, finding information, understanding APIs and debugging production systems also consume time. He mentions Atlassian’s own operations product among tools intended to help run software and acknowledges that more software can also bring more operational complexity.

Stebbings asks whether low switching costs make AI companies fundamentally weak businesses. Cannon-Brookes argues that rapidly built businesses often have low switching costs at first; whether they develop stronger customer relationships and defensibility over time is a separate question.

He sees today’s AI wave as unusual because many technically capable people gained access to models through cloud services at roughly the same time. That makes experimentation widespread and copying faster. He suspects defensibility could eventually come from delivered value, data, workflows, familiarity and design, but says the answer is not yet clear.

For Atlassian, the race is whether an incumbent with existing customers and substantial R&D can integrate AI faster than startups acquire distribution. Cannon-Brookes cites the company’s hundreds of thousands of customers and millions of users; these are his company-scale claims and a competitive argument, not a guarantee.

On AI margins, he says it is too early to know the durable profile when pricing and monetization models keep changing. A model that costs a hypothetical $1 billion to train may face a newer competitor months later, leaving providers to work out how they recover costs.

Economic value could accrue at different layers, from chips and power to applications. Cannon-Brookes doubts that power producers will capture all of it: customers may pay for an application’s design and experience as well as calculable output. He warns that a simple model wrapper may struggle, while allowing that such a wrapper could be a starting point for a different business.

Stebbings asks whether per-seat software pricing will disappear. Cannon-Brookes identifies usage-based pricing and pricing by delivered value as alternatives, but says customers often dislike uncertain consumption bills. In his view the final answer may mix approaches rather than replace every seat-based model.

Outcome-based pricing raises its own problems: buyer and seller must agree on how to measure the outcome, and a vendor cannot indefinitely charge for the same cost saving as the customer’s baseline changes. His support-ticket example is a hypothetical demonstration of the problem.

Asked what Atlassian must protect, Cannon-Brookes answers: its creativity. A company hoping to last multiple decades must navigate successive technology transitions by creating, changing and sometimes retiring what it previously built, rather than only defending an existing product.

He thinks Atlassian is well positioned for AI but says the company still has to prove it. Whatever follows AI will also demand creative adaptation. Maintaining a culture where people can propose killing one product and building another is, in his view, part of that work.

Cannon-Brookes pushes back on unconstrained thinking: constraints force choices. Talent is difficult to acquire, but he does not see Atlassian as constrained like a three-person startup. He describes its scale and resources as a reason to build ambitious things while still choosing carefully among competing priorities.

He recalls a line from Shopify’s Tobi Lütke: the founder’s job is to fight the entropy of ambition. Despite Atlassian’s size, Cannon-Brookes thinks it must retain the sense that it is small in the global economy and still has to earn each customer.

After 23 years, he has experienced physical and mental exhaustion. He recommends breaks and self-awareness, and says a founder needs to find meaning in the ups and downs of the work. He acknowledges his own perspective may be unusual because he continues to enjoy the job.

Stebbings asks whether becoming wealthier made him a better leader by reducing anxiety. Cannon-Brookes says money is not his primary motivation; instead, the richness of his experiences, including raising children, has made him more empathetic and, he believes, a better leader.

Would he found the company again knowing how hard it was? Averaged over the whole journey, he says yes, while acknowledging luck and hard work. He emphasizes the people around him: a founder should want to spend time with colleagues, because a company is a group pursuing a common goal.

In the quick-fire round he recalls an unsuccessful experiment with status updates in Confluence roughly two decades earlier. He also expects OpenAI and Anthropic to become large and important companies, while saying nobody knows precisely how large; Stebbings’s multitrillion-dollar valuation possibility is a speculative scenario.

Cannon-Brookes says turning AI demonstrations into delivered customer value may take longer than many expect. Businesses have to understand the problem, rethink workflows and address people, data quality, access and security. A striking demo does not remove that implementation work.

Asked for advice to a new parent, he says to relax and enjoy the ups and downs, reflecting on his own experience with four children. He also sees strengths and weaknesses as two sides of the same trait: being unreasonable can drive change but can also make someone difficult.

For a serious work problem, Cannon-Brookes says Farquhar is still likely to be his first call. Looking toward 2035, he hopes Atlassian remains a creative, competitive company that attracts talented people and builds useful things.

On a personal level, he says being remembered as a good father matters deeply. Eventually Atlassian must continue without either founder; his hope is that it remains a vibrant company solving problems, rather than merely surviving in name.