Roles aren't converging—they're expanding | Tamar Yehoshua (Atlassian CPO)
In a Nutshell
Product managers' roles aren't shrinking—they're expanding as AI tools give everyone new capabilities while the core work of finding product-market fit remains unchanged. At scale, teams like Atlassian's are seeing 3x productivity gains by having PMs write code, run evaluations, build prototypes, and automate feedback loops, while eliminating manual updates and meetings. Success requires building AI proficiency through structured training and focusing on what actually benefits customers rather than chasing every new tool.
These notes were generated by AI and may contain inaccuracies.
Product managers were previously worried about their roles disappearing as other specialties claimed expertise in each other's areas. Engineers were told their jobs would disappear, designers told product managers their roles would vanish, and everyone was concerned about job loss. This fear has now subsided as roles have begun to overlap rather than converge.
A new era of the "artificial intelligence builder" has emerged. The "Lenny" podcast, startup conversations, and X platform discussions frequently reference "AI builders." Some startups now hire "builders" instead of separate product managers or designers. This role has always existed for founders who wore multiple hats when establishing companies.
Atlassian, with over 10,000 employees, provides insight into how these changes manifest in large enterprises rather than 5-10 person startups. Engineering teams are using programming agents and restructuring code rules extensively, though the focus remains on product manager roles.
Roles have become overlapping and are expanding as everyone gains ability to perform tasks previously impossible. The core product manager job remains unchanged: finding product-market fit, building products people love, and creating sustainable business models. The methods for accomplishing these goals have transformed.
Product managers frequently ask how to keep up with rapidly improving AI tools. The advice given is that no one can fully keep up with everything, and not everything shared on the X platform should be believed. The focus should remain on discovering what truly benefits customers through AI tools.
Teamwork graphs provide organizational context that was previously unavailable. At Atlassian, a "teamwork graph" enables visibility into the full organizational context. Product managers no longer need to take meeting notes, follow up on work items, or communicate with marketing teams about launch dates, as tools can handle these tasks.
Mike, Atlassian's CEO, recently asked about a feature release timeline and received the answer through "Rovo," Atlassian's AI product, within five minutes. This demonstrates how information access has changed, though change management is still required to help people effectively use these tools.
Product managers can focus on accelerating team progress. Using the ship metaphor, the question becomes whether product managers are rowing or steering to reach destinations faster. Intelligence (model capabilities) combined with organizational context enables faster movement.
Three specific examples demonstrate how product manager roles vary based on product type and development stage:
- New features in existing software databases
- Products built from scratch
- Improvements to large, existing software bases
The Confluence team launched "Remix" with Rovo and "Confluence Slides" features. These AI-heavy features were developed with a focus on building faster and radically changing work styles rather than simply specifying requirements.
Product managers began entering code check-ins. One product manager who had never written code before created a utility with their engineering partner for easier front-end code input. She filed 26 pull requests in one month, exceeding most engineers on the team. Her motivation was fixing user experience issues without sufficient engineering resources, allowing engineers to focus on higher-level activities.
Evaluations have become crucial, with product managers sometimes conducting them and engineers sometimes handling them. For Confluence Slides, product managers traditionally led evaluations while working with engineers to ensure proper evaluation approaches. This resulted in double productivity in evaluations.
Product managers used Atlassian's language modeling platform "Rise" to correct errors by identifying prompt problems and sending them to engineers more quickly.
Design flaws were automated using Figma MCP to link Figma designs to actual code, then using software agents to automatically fix mismatches. This fixed approximately 14 errors in one hour, significantly improving product quality.
Engineers reduced test creation time from half a day to 10 minutes through AI assistance. Remix was released in 6 weeks and Confluence Slides in 8 weeks, compared to the previous 6-month timeline before AI tools, with tighter iteration loops enabling faster customer feedback incorporation.
"RoboClaw" was built from scratch by product manager Josh and designer Kevin, who programmed everything based on intuition. They reached an initial beta version used internally before adding engineers. Josh initially contributed to coding but later stepped back to focus on direction-setting, prioritization, and obstacle removal when he realized engineers were going off track.
Josh's initial programming experience helped him understand obstacles and challenges, creating better guidance for the team. He built an agent into RoboClaw to provide weekly updates automatically, eliminating manual weekly report writing.
Jira, over 20 years old with data center origins, cloud migration, isolated cloud support, FedRAMP standards compliance, and hundreds of thousands of customers requiring careful handling, presented different challenges. Product managers did not enter code in the actual work environment due to risk.
The goal was making Jira AI-based first, building AI features to help customers use AI in software development. The team achieved 3 times normal productivity, launching 22 user-oriented features in 10 weeks.
Prototypes were enhanced using Loom to record user interfaces, Figma designs with voiceover, or brainstorm ideas. Loom automatically creates work elements from recordings, which product managers can move to implementation, triggering software agents. This streamlined process wrote code in the actual front-end repository using cloud-managed agents, ensuring compatibility with Atlassian design language.
Feedback received via Slack was classified using Jira agents and automatically sent to programming agents for fixing. Over 900 comments from user studies were reviewed using Jira Service Management, with automated agents classifying and organizing insights. The quality of automated ranking results was surprisingly high.
In the Jira case, product managers focused on obstacle removal for engineers, processing internal and external feedback more effectively, and building reusable prototypes without entering code in the actual work environment.
Product managers now build models and tests, write evaluations, delve deeper into customer feedback, and consume information at higher rates when AI tools understand organizational context through teamwork graphs.
Product managers no longer perform manual updates, collect research manually, or create presentation slides from scratch. The number of simultaneous meetings has decreased.
Atlassian created an "AI Proficiency Index" with six abilities: using tools, writing evaluations, automating data insights, creating prototypes, and having necessary technical knowledge. The index ranges from 1 (curious) to 5 (entrepreneurial), with 3 being capable. Everyone is expected to reach level 3 across all aspects, with focus on reaching level 5 in areas most important to their team.
"AI Builders Weeks" occur quarterly, dedicating a week to training product managers and designers. The program includes external speakers, internal product managers with advanced skills training colleagues, and project work to apply learned skills. Over 1,000 people have participated, creating over 120 new workflows still used after the training.
Each AI Builders Week focuses on one skill area: prototyping, evaluations, building proxies, and code review. An "AI Builders Week in a Box" is available on Atlassian's website with schedules and implementation guidance.
Measuring AI impact remains unsolved. Current measurements include pull requests posted to production environments, features delivered from idea to customer use, OKRs, and overall productivity. Team-level productivity measurement is often easier than organizational measurement. Experiments are conducted quarterly to identify measurable results.
Keep Lenny's Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
