Atlassian's Mike Cannon-Brookes: AI will remake jobs, not erase them
Key takeaways from the Big Technology Podcast live conversation at Web Summit with Atlassian CEO and co-founder Mike Cannon-BrookesDealroom has a profile for this one. Try Dealroom → (November 2026):
Work is changing in layers. Cannon-Brookes says businesses are weaving agentic work into workflows, seeing modest efficiency gains so far, some quality improvement, and the start of a broader growth in creativity as teams work out how to use AI.
Jobs change rather than disappear. He calls restructuring, not AI, the dominant force behind recent tech layoffs, and expects AI to create far more jobs than it removes. His parallel: bank branches were supposed to die with online banking, yet have grown — by doing vastly different jobs.
A talent challenge, not a technology one. The real risk, he says, is losing work to someone who uses AI well; companies must change how people think about their roles, which is harder than shipping technology.
Design and experience lag models. The reason better models have not shown up in company productivity is interface and experience design, in his view. Atlassian recently shipped its agentic multiplayer protocol (AMP) and is standardising on MCP and A2A underneath familiar Jira, Confluence, Loom and service-desk interfaces.
The teamwork graph. An enterprise context graph built over eight years by around 1,000 people at Atlassian, spanning documents, projects, goals, communication, code and data assets, with inference on top. Agent sessions are indexed so learnings compound into later sessions, cutting token use.
Four ways humans and agents work together. Chat (one-on-one exchange), delegation (an agent acts on your behalf), collaboration (agents and people side by side on a shared surface) and pairing (working in parallel on the same outcome) — each demanding different product experiences.
Models are not commodities. Cannon-Brookes says models will keep different strengths and cost profiles, and argues for model blending over jumping to the latest frontier model; the harness and the context layer deliver much of the value.
Read more: Big Technology Podcast on YouTube