From Rowing to Steering: How Atlassian's PMs Are Expanding Their Role with AI
Key points
Key takeaways from Lenny's Podcast at the Lenny and Friends Summit interview with Atlassian Chief Product & AI Officer Tamar Yehoshua (September 2026):
The AI builder is real — but different at 10,000 people. Tamar Yehoshua argues the "AI builder" role is genuinely emerging, with startups hiring builders rather than separate PMs and designers, while inside a company the size of Atlassian the PM job is unchanged in purpose — find product-market fit, build products people love and a business they will pay for — but expanding in method: roles are not converging, each one can now do far more with AI tools.
Context is the unlock. Atlassian built a "teamwork graph" that gives AI tools the full organisational context, freeing PMs from taking notes, chasing action items and hand-writing launch communications. Tamar recounts how CEO Mike Cannon-Brookes DM'd a product lead asking when a feature would launch — and five minutes later messaged back a screenshot from Rovo answering it himself.
Confluence Remix — PMs rowing. On the AI features Remix with Rovo and Confluence Slides, a PM who had never written code or used a terminal began checking in code through an engineer-built harness, landing 26 PRs in a month — more than most engineers — to fix UX faster. The team doubled eval throughput, automated design-bug fixes by mapping Figma designs to code with a coding agent (~14 bugs an hour) and cut test creation from half a day to 10 minutes, shipping in six to eight weeks instead of roughly six months.
Rovo — from rowing to steering. A zero-to-one project was vibe-coded from scratch by PM Josh and designer Kevin to a working internal alpha before engineers joined; Josh then stepped back from coding once it was no longer his highest-leverage activity, using his hands-on understanding of the blockers to steer and unblock the team — and built an agent inside the product to write its own weekly status updates.
Jira — PMs steering, not touching production code. Because Jira's 20-plus-year, heavily customised codebase made PM code-writes dangerous, the team instead went AI-first with PMs prototyping via Loom (recordings auto-generate work items that trigger coding agents), triaging feedback through a Jira agent in Slack, and using Rovo to categorise 900-plus pieces of customer feedback. Throughput roughly tripled and 22 user-facing features shipped in about 10 weeks.
What PMs now do — and stop doing. New work includes prototyping and testing, writing evals, going deeper on customer feedback and consuming information at a much higher rate; PMs no longer do manual status updates, hand-compiled research or building slide decks from scratch. All of it depends on AI tools understanding organisational context through the teamwork graph.
Building AI fluency at scale. Atlassian's "AI fluency index" grades PMs from 1 (curious) to 5 (pioneering) across six capabilities — expecting everyone to reach level 3 over time — and quarterly "AI builder weeks" on prototyping, evals, building agents and checking in code have trained more than 1,000 people and produced over 120 reusable workflows. Atlassian now offers an "AI builder week in a box" to customers.
Outcome measurement remains unsolved. Tamar concedes no one has cracked how to measure AI-driven impact yet. Atlassian tracks PRs deployed (not written) to production, time from idea to delivery in customers' hands, OKR attainment, and both team-level and organisation-level throughput, running new measurement experiments each quarter.
Read more: YouTube — Lenny's Podcast
