Atlassian Pushes PMs Into Code as AI Agents Triple Delivery Speeds

Lenny's Podcast

Artificial intelligence will expand rather than replace product management roles because automated context engines eliminate administrative drudgery, freeing PMs to either write code directly or steer engineering pipelines.

AI-assisted summary of Roles aren't converging—they're expanding | Tamar Yehoshua (Atlassian CPO) — original source. Check the original for full context.

Source published: 2026-09-28

Key ideas

Product roles are expanding rather than converging: AI tools grant PMs, designers, and engineers broader functional reach, while leaving the core PM responsibility of discovering product-market fit and customer value unchanged. (2:04)
Atlassian balances PM technical involvement using a 'rowing vs. steering' framework, allowing non-coding PMs to check in frontend PRs in safe environments while stepping back to steer when engineering bandwidth is staffed. (5:24)
Jira accelerated development throughput threefold and shipped 22 features in 10 weeks across its 20-year legacy codebase by pairing Loom specs, cloud-managed coding agents, and agentic customer feedback triage. (12:24)
Instead of replacing staff with AI-native hires, Atlassian upskills its product organization through a 5-level AI Fluency Index and quarterly AI Builder Weeks, engaging over 1,000 employees to build 120+ internal workflows. (17:00)

Detailed notes

## Background Published on YouTube by Lenny's Podcast on September 28, 2026, this keynote features Atlassian Chief Product Officer Tamar Yehoshua speaking live at the Lenny and Friends Summit. The presentation explores what the "AI builder" era looks like for product managers within large enterprises, drawing on case studies from Confluence, Jira, and new greenfield initiatives to illustrate when PMs should write code or prototypes versus when they should focus on steering the team.

## TL;DR Yehoshua argues that rather than collapsing or replacing specialized roles, AI is expanding the reach and leverage of product managers while leaving their core mission—finding product-market fit and delivering customer value—intact. Using case studies from Confluence, the 0-to-1 project Rovoclaw, and Jira's legacy codebase, she illustrates the dynamic balance between "rowing" (hands-on prototyping and code commits) and "steering" (unblocking engineers, synthesizing user feedback, and setting priorities). She details how Atlassian equips enterprise teams with organizational context via the Teamwork Graph and upskills PMs through an AI Fluency Index and hands-on Builder Weeks. Ultimately, she offers an optimistic outlook, declaring that AI makes this the most exciting era in history to be a product manager.

## Atmosphere & style The presentation carries a pragmatic, energetic, and candid keynote tone, trading early industry panic over AI job displacement for real-world, in-the-trenches execution. Yehoshua speaks directly and humorously with the audience of product leaders, sharing unvarnished anecdotes about CEO DMs, messy vibe-coding experiments, and the unsolved complexities of measuring developer productivity. The talk feels reassuring and grounded, transforming abstract hype around autonomous agents into actionable operational wisdom for modern enterprise builders.

## Bulletin - [02:04](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=124s) Product roles are expanding rather than converging: AI tools grant PMs, designers, and engineers broader functional reach, while leaving the core PM responsibility of discovering product-market fit and customer value unchanged. - [05:24](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=324s) Atlassian balances PM technical involvement using a 'rowing vs. steering' framework, allowing non-coding PMs to check in frontend PRs in safe environments while stepping back to steer when engineering bandwidth is staffed. - [12:24](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=744s) Jira accelerated development throughput threefold and shipped 22 features in 10 weeks across its 20-year legacy codebase by pairing Loom specs, cloud-managed coding agents, and agentic customer feedback triage. - [17:00](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1020s) Instead of replacing staff with AI-native hires, Atlassian upskills its product organization through a 5-level AI Fluency Index and quarterly AI Builder Weeks, engaging over 1,000 employees to build 120+ internal workflows.

## Key takeaways - [00:06](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=6s) Reciprocal panic regarding AI replacing engineers, designers, or product managers has transitioned into the era of the multi-disciplinary 'AI builder'. - [02:04](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=124s) The fundamental mission of product management—validating product-market fit, delighting users, and creating commercially viable software—remains completely unchanged in the AI era. - [02:04](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=124s) PMs should filter out widespread social media hype and anchor their adoption of AI tooling strictly to demonstrable customer outcomes. - [02:04](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=124s) Organizational context graphs and internal AI tools like Rovo eliminate manual PM administrative overhead by automating status queries and cross-functional updates. - [02:04](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=124s) The critical operational decision for a modern PM is knowing when to 'row' by directly building artifacts versus when to 'steer' by clarifying direction and unblocking the team. - [05:24](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=324s) In Confluence, a non-technical PM shipped 26 production pull requests in a month within a sandboxed frontend harness, resolving UI polish and compressing delivery from six months to six weeks. - [05:24](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=324s) Systematic evaluations (evals) and prompt regression monitoring have become essential operational responsibilities owned jointly by PMs and engineers. - [10:10](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=610s) In 0-to-1 greenfield products like Rovoclaw, PMs can bootstrap functional prototypes alongside designers, but must actively disengage from coding once dedicated engineers join to avoid misalignment. - [12:24](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=744s) Legacy, compliance-heavy enterprise codebases require guardrails against PM production commits, shifting focus instead toward cloud-managed agents that scaffold design-compliant prototypes. - [12:24](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=744s) Agentic workflows can automate feedback loops by categorizing internal bug reports from Slack and synthesizing qualitative customer research across hundreds of video interviews. - [16:18](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=978s) Manual management drudgery such as drafting weekly status updates, building presentation slides, and transcribing research is being replaced by autonomous agents. - [17:00](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1020s) Atlassian's AI Fluency Index maps six core technical proficiencies across five competency tiers as an educational skill-acquisition guide rather than a performance evaluation ladder. - [17:00](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1020s) Quarterly AI Builder Weeks combine industry keynotes, peer workshops, and applied capstones to upskill cross-functional teams in prototyping, evals, and agent design. - [17:00](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1020s) Measuring AI impact requires moving away from vanity metrics like lines of code toward production deployment rates, end-to-end cycle times, and overarching OKRs. - [20:30](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1230s) Evaluating productivity requires tracking both micro-level team velocity and macro-level organizational throughput through continuous quarterly experimentation. - [20:30](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1230s) Product management roles are expanding rather than disappearing, positioning current practitioners for higher leverage and long-term durability over the coming decade.

## Detailed walkthrough ### [00:06](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=6s) From Job Replacement Paranoia to the Era of the AI Builder

Addressing a packed room at the Lenny and Friends Summit, Tamar Yehoshua opened by marveling at the sheer density of product leaders gathered in one space—noting that with many in the audience, she had spent the past year or two in continuous debate over the evolving definition of product management. Looking back just twelve to twenty-four months, the tech landscape was gripped by reciprocal panic. Engineers, designers, and PMs were all nervously pointing fingers at each other's crafts and declaring, "Your job is going to go away." Thankfully, that apocalyptic anxiety has subsided, giving way to a much more constructive reality: disciplines are overlapping, and the era of the "AI builder" has taken center stage.

In startups, on X, and across tech podcasts, this builder archetype has become a tangible job title. Early-stage startups are increasingly skipping dedicated PM or designer hires in favor of multi-faceted builders who wear every hat—a dynamic that, as Yehoshua pointed out, has always been the natural state of early founders. But while this agile, boundaryless dynamic works seamlessly in a five- or ten-person startup, product leaders inside enterprise environments often scratch their heads wondering how the model translates when your company employs more than 10,000 people. Inside Atlassian, where engineering teams are already heavily augmented with coding agents and automated refactoring, the fundamental question centers on what this transformation means for product management.

[02:04](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=124s) Acceleration, Context, and the "Rowing vs. Steering" Dilemma

Yehoshua emphasized that roles are not merely converging—they are expanding. When empowered by modern tooling, individuals across every function can accomplish vastly more than before. Yet, the foundational purpose of a product manager remains entirely untouched: discovering product-market fit, building software users genuinely love, and creating a viable business that customers are willing to pay for. What has changed is purely *how* that work gets executed. To PMs overwhelmed by the relentless wave of new AI releases, her practical advice was blunt: you cannot keep up with everything, so "everything that you read on X, don't believe." Instead, filter out the noise by anchoring strictly to what delivers demonstrable value to customers.

```mermaid flowchart LR accTitle: The Acceleration Equation accDescr: Visual representation of how model intelligence combines with organizational context to accelerate team velocity, framed by the PM choice of rowing versus steering.

subgraph Inputs["The Velocity Inputs"] A["Model Intelligence (LLMs & Coding Agents)"] B["Teamwork Graph (Organizational Context)"] end

subgraph Core["Acceleration Engine"] C["Combined Context & AI"] D{"PM Focus: Rowing or Steering?"} end

subgraph Outcome["Outcome"] E["Faster Delivery to Destination"] end

A --> C B --> C C --> D D --> E ```

At the same time, organizational context has become an enterprise superpower. With Atlassian's "Teamwork Graph", internal tools now surface institutional context that previously lived in scattered documents or human memory. This eliminates traditional PM operational overhead like chasing status updates, manually recording meeting notes, or relaying launch dates to go-to-market teams. To illustrate the shift, Yehoshua shared a recent anecdote involving Atlassian co-founder and co-CEO Mike Cannon-Brookes:

> "The other day, last week, Mike, our CEO, DM'd me and one of the product leads and says, 'When is some feature launching?' Five minutes later, he sent a screenshot of Rovo, which is Atlassian's AI product, and said, 'This is what Rovo said. Is it right?' We're like, 'Yeah.' And he said, 'Sorry, I wasted your time.'"

Freeing PMs from information gatekeeping allows them to focus on team acceleration. Atlassian frames this through a nautical metaphor: your product team is a ship trying to reach its destination faster through the pairing of model intelligence and organizational context. The critical operational question for every PM is: **Are you rowing, or are you steering?** Through internal experimentation, Atlassian discovered that the answer is not dogmatic; it depends entirely on the product type and lifecycle stage.

- **Core PM Mission:** Discovering product-market fit, delighting customers, and ensuring monetization remain identical to previous eras. - **Enterprise Context:** Tools like Rovo and the Teamwork Graph automate cross-functional status tracking, removing PM administrative friction. - **Acceleration Metric:** Speed gains stem from combining raw model intelligence with deep organizational context.

[05:24](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=324s) Case Study 1: Confluence, Code Commits, and Automated Design Fixes

The first concrete operational archetype is adding new AI capabilities into an existing codebase, exemplified by Confluence's recent launches: **Remix with Rovo** (allowing users to highlight text and instantly transform it into alternative formats) and **Confluence Slides** (converting standard Confluence pages into finished presentation decks with a single click). Rather than treating these as standard feature tickets, the Confluence team treated them as an experiment in radical workflow transformation within a mature, compliant enterprise environment.

The most striking shift occurred when the feature's product manager, Ya, began directly checking in production code. Despite having zero prior coding experience and having never opened a terminal, she partnered with an engineering lead who set up a sandboxed frontend harness for her. Ya ended up shipping 26 pull requests in a single month—exceeding the output of most software engineers on the team. When Yehoshua asked why she took this on, Ya explained that the user interface had countless minor polish items that engineering lacked the bandwidth to address; by jumping into the frontend herself, she freed the engineers to concentrate on deeper technical architecture.

Beyond PM-driven PRs, the team overhauled multiple operational bottlenecks: - **Systematic Evaluations (Evals):** Evals became a core PM responsibility. The PMs partnered closely with engineering to codify target output quality, doubling eval throughput (a 2x increase). They also leveraged the Arize platform for observability and debugging, diagnosing prompt regressions directly before routing fixes to engineers. - **Automated Design Bug Resolution:** The team integrated Figma MCP (Model Context Protocol) to map design tokens directly to the codebase, pairing it with an autonomous coding agent to reconcile design discrepancies. This system resolved approximately 14 visual bugs per hour without manual design-engineering back-and-forth. - **Automated Test Generation:** Engineering leveraged AI tooling to slash the time required to write a test suite from half a day down to just 10 minutes. - **Cycle Time Reduction:** While enterprise compliance and security usually mandate extensive cycles, Remix launched in 6 weeks and Confluence Slides launched in 8 weeks—work that historically required roughly 6 months—while supporting far tighter customer beta feedback loops.

In this scenario, the PM actively chose to **row**. Supported by an engineering manager who isolated her work in a safe repository branch, the team deliberately blurred craft boundaries from day one to maximize velocity.

[10:10](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=610s) Shifting from Rowing to Steering: The 0-to-1 Journey of Rovoclaw

Tamar Yehoshua highlights a greenfield project called Rovoclaw to illustrate how the product manager's role evolves across different phases of development. Unlike an existing feature built inside Atlassian's AI product, Rovo, Rovoclaw was conceived as a true "zero-to-one project from the ground up." It began with just two people: a product manager named Josh and a designer named Kevin. The duo essentially "vibe coded the whole thing" from soup to nuts, hacking together a working alpha that internal customers at Atlassian immediately began using. Once they validated that they were genuinely onto something, engineers were officially staffed onto the project. At first, the dynamic stayed scrappy—Josh and Kevin continued checking in code right alongside the newly assigned engineering team.

Before long, Josh noticed a concerning friction: because he was so buried in writing code himself, the engineers were beginning to veer off track. He had to stop and ask himself: "Wait a minute, should I be spending my time coding? Is this actually the highest leverage activity that I could be doing?" Josh made the conscious decision to pull back from the codebase and return to foundational PM work: setting direction, establishing priorities, and clearing road-blocks. As soon as he stepped back into steering, the entire project accelerated. The time Josh had spent in the trenches writing code paid massive dividends—he had direct empathy for the codebase, intimately understood the team's technical blockers, and could guide the product with far higher precision.

At the same time, Josh leaned heavily on AI to eliminate routine managerial overhead. Yehoshua quips to the audience: "I hope most of you are not writing weekly reports by hand anymore. And if you are, you probably don't want to admit it, but please go and build an agent to do it." Taking that advice to heart, Josh built an agent directly inside Rovoclaw that wrote weekly status reports about itself, which Yehoshua notes were "super amusing." In this 0-to-1 scenario, the PM started out "rowing" by slinging code to bootstrap the product, but deliberately transitioned to "steering" once leverage shifted toward unblocking engineers.

- **Initial builders**: Josh (PM) and Kevin (Designer), who vibe coded an internal alpha from scratch before adding dedicated engineers. - **Strategic pivot**: Moving from active code contributions to high-leverage steering (unblocking, prioritizing, and aligning the engineering team). - **Self-reporting agent**: Josh configured an agent within Rovoclaw that automated weekly project status updates on its own progress.

---

[12:24](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=744s) Modernizing Jira: Accelerating a 20-Year Legacy Codebase with AI

Bringing AI into a mature enterprise application is an entirely different operational beast. Jira has been around for over two decades, spanning a massive, intricate codebase that transitioned from on-premise Data Center to the cloud, while supporting isolated cloud instances, FedRAMP compliance, and extensive bespoke customizations for hundreds of thousands of enterprises. In a critical environment where breaking production is unacceptable, having PMs check code into production is "kind of dangerous," so Atlassian prohibited it. Instead, the challenge was transforming Jira into an "AI-first" experience—reinventing the software development lifecycle so users can seamlessly assign work to autonomous agents and interact with AI across every Jira workflow.

Despite the structural constraints of a legacy codebase, the Jira team achieved dramatic velocity gains. From "in progress" to "shipped," their development throughput increased roughly 3x, successfully shipping 22 user-facing features in just 10 weeks. They achieved this by overhauling prototyping and automated triage:

```mermaid flowchart TD accTitle: Jira AI-Driven Prototyping and Feedback Workflow accDescr: Flowchart illustrating how Loom recordings, cloud coding agents, and Slack or JSM triage loops accelerate delivery in Jira.

subgraph Prototyping ["1. Spec to Reusable Prototype"] A["PM records Loom video (UI idea, Figma, voiceover)"] --> B["Loom auto-generates Jira work items"] B --> C["Move item to 'In Progress'"] C --> D["Cloud-managed coding agent generates frontend code"] D --> E["Accessible, design-compliant prototype in shared repo"] end

subgraph InternalTriage ["2. Internal Feedback Loop"] F["Bugs & feedback submitted in internal Slack channels"] --> G["Jira Slack Agent triages and categorizes issues"] G --> H["Dispatched to coding agent for automated fixes"] end

subgraph ExternalInsights ["3. External Customer Feedback"] I["900+ user study videos ingested via Jira Service Management"] --> J["Rovo Agent triages & extracts actionable feature insights"] J --> K["Engineers unblocked with clean requirements"] end

E --> K H --> K ```

Prototyping became far more rigorous than throwaway mockups. A PM records a Loom video capturing proposed UI changes, walking through Figma screens with a spoken track, or brainstorming concepts. Loom automatically extracts work items and populates them into Jira. When the PM moves an item to "In Progress," it immediately triggers a managed coding agent running in the cloud. The agent writes code directly into the actual front-end repository—sparing PMs the headache of maintaining complex local development environments—while adhering strictly to Atlassian design system components and accessibility standards. Because the prototype lives in the real repository, handoffs to production engineers require virtually zero rework.

Handling the flood of internal and external feedback was streamlined through similar agentic pipelines. Internally, when Atlassian employees test Jira changes, feedback and bug reports funneling through Slack are automatically categorized by a Jira Slack agent and routed straight to coding agents for automated patching. Externally, during user research, the team used Jira Service Management and a Rovo agent to ingest and analyze customer video sessions across more than 900 discrete pieces of feedback. The agentic categorization proved surprisingly high quality, allowing PMs to focus entirely on steering the ship, synthesizing user needs, and feeding clean, actionable context to engineers without writing production code.

- **Scale and compliance constraints**: 20+ year-old codebase supporting cloud, Data Center, FedRAMP, and isolated enterprise tenants. - **Velocity metrics**: 3x improvement in throughput from in-progress to shipped, delivering 22 customer-facing features within 10 weeks. - **Managed cloud agents**: Coding agents scaffold UI prototypes in the real front-end repository, pre-configured with Atlassian's design language and accessibility guardrails. - **Multi-channel triage**: Jira Slack agent routed bugs directly to coding agents; Rovo agents synthesized over 900 user-testing feedback items in Jira Service Management.

---

[16:18](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=978s) What PMs Do (and Don't Do) in the AI Era

Reflecting on these operating shifts, Yehoshua outlines how the product management discipline is structurally transforming. Rather than roles collapsing or homogenizing, PM responsibilities are expanding into areas that make teams move faster, supported by foundational infrastructure like the Atlassian Teamwork Graph, which provides AI models with the underlying organizational context of teams, projects, and goals.

PMs are actively taking on new tasks: - **Prototyping and testing**: Rapidly generating functional prototypes and validating them before engineering commits cycles. - **Writing evals**: Designing evaluation benchmarks to systematically grade model outputs and prompt efficacy. - **Deep customer feedback analysis**: Consuming and synthesizing qualitative customer insights at volumes previously impossible to parse manually. - **Accelerated context ingestion**: Digesting massive streams of cross-functional documentation and research in fraction of the time.

Conversely, manual operational drudgery has fallen away. PMs are no longer writing status updates by hand, manually compiling raw research findings, or building slide presentations from scratch. Yehoshua candidly adds with a smile: "I would love to say that there are no synchronous meetings, but that would be an exaggeration... but there are, I believe, fewer synchronous meetings."

---

[17:00](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1020s) The AI Fluency Index and AI Builder Weeks

While venture-backed startups often advise large companies to simply "hire people that are AI native," Yehoshua rejects that playbook for established organizations. Atlassian already employs exceptional PMs; the strategic imperative is upskilling them to become AI native. To give the team clear developmental milestones, Atlassian created the **AI Fluency Index**.

The index tracks six core capabilities, ranging from understanding tool stacks and authoring model evals to automating data insights, rapid prototyping, and general technical literacy. Fluency is scored across a five-point scale:

Yehoshua emphasizes that the Fluency Index is explicitly **not** a performance ladder: "The ladder is about outcomes and what you produce for your customer. This is helping PMs understand the skills that they should be attaining."

To turn that framework into practice, Atlassian established quarterly **AI Builder Weeks**. During these dedicated sprints, PMs and product designers step away from their day-to-day roadmaps to focus exclusively on technical skill acquisition. Each week follows a structured rhythm: 1. **External inspiration**: Keynotes from industry practitioners (including a shout-out to guest speaker Claire). 2. **Peer-led coaching**: Advanced internal PMs who have already mastered a specific workflow teach workshops to upskill colleagues. 3. **Hands-on capstone projects**: Participants team up to build functioning tools, workflows, or automations using newly acquired techniques.

The program has scaled across the company with major impact: over 1,000 employees have participated, producing more than 120 production workflows that teams continue to use daily. Each edition zeroes in on a single thematic discipline: the first focused on prototyping, the second on writing evals, the third on building agents, and the most recent on checking in code. Atlassian has open-sourced this curriculum as an "AI Builder Week in a Box" on its website for external teams.

When asked by enterprise customers how Atlassian measures the ultimate impact of these AI initiatives, Yehoshua admits candidly: "I'll be honest, we haven't figured it out. I don't think anyone has figured it out." Measuring AI productivity runs into the age-old dilemma of measuring engineering velocity. Instead of tracking superficial outputs like pull requests written, Atlassian looks at PRs deployed to production, cycle times from initial concept to customer delivery, feature usage, overall team throughput, and meeting top-line OKRs.

[20:30](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1230s) Measuring Speed, Expanding Roles, and the Future of Product Management

Tracking performance and acceleration during this AI shift requires looking at productivity through multiple lenses. It is often far easier to gauge the throughput of an individual team than it is to track the velocity of an entire organization, but leaders need to monitor both: ensuring individual teams are moving faster day to day while verifying that the broader organization is accelerating as a cohesive whole. At Atlassian, this process is treated as an ongoing, active initiative. Every single quarter, teams run structured experiments across varying workflows, with leadership evaluating what concrete outcomes can be captured and measured. Rather than claiming to have solved the measurement puzzle outright, Yehoshua leaves an open door to the audience, inviting anyone with proven methods for measuring organizational throughput to share their insights.

Synthesizing the broader transformation, Yehoshua returns to her central thesis: roles across tech are not collapsing or converging into a single undifferentiated title, but rather "each role is expanding." Product managers are now empowered to reach further, doing more to execute their core responsibilities faster, more effectively, and with higher quality, all while delivering dramatically more value directly to end customers. Addressing the pervasive anxiety surrounding AI displacing knowledge workers, Yehoshua strikes an unapologetically optimistic note about the discipline's longevity. Declaring that "this is the best time in the world to be a PM," she reassures the room that PMs will still very much be here a decade from now, closing her talk by reminding the audience how fortunate they are to build products in this transformative era.

- **Dual-level measurement**: Organizations must measure productivity at both the granular team level and the macro organizational level to ensure localized speed translates into real systemic gains. - **Quarterly experimentation**: Teams continually run internal workflow experiments every quarter to test new methods and benchmark measurable outcomes. - **Role expansion over convergence**: AI does not eliminate specialized functions; instead, it expands the scope and leverage of PMs, enabling them to ship faster and elevate customer impact. - **Optimism for the discipline**: Dismissing predictions of the demise of product management, Yehoshua emphasizes that PMs will endure well over the next decade and calls the current moment the most exciting time in history to enter the craft.

## Notable quotes > [00:06](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=6s) Your job is going to go away.

> [02:04](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=124s) everything that you read on X, don't believe. — Tamar Yehoshua

> [02:04](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=124s) The other day, last week, Mike, our CEO, DM'd me and one of the product leads and says, 'When is some feature launching?' Five minutes later, he sent a screenshot of Rovo, which is Atlassian's AI product, and said, 'This is what Rovo said. Is it right?' We're like, 'Yeah.' And he said, 'Sorry, I wasted your time.' — Tamar Yehoshua

> [10:10](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=610s) Wait a minute, should I be spending my time coding? Is this actually the highest leverage activity that I could be doing? — Josh

> [10:10](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=610s) I hope most of you are not writing weekly reports by hand anymore. And if you are, you probably don't want to admit it, but please go and build an agent to do it. — Tamar Yehoshua

> [16:18](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=978s) I would love to say that there are no synchronous meetings, but that would be an exaggeration... but there are, I believe, fewer synchronous meetings. — Tamar Yehoshua

> [17:00](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1020s) hire people that are AI native

> [17:00](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1020s) The ladder is about outcomes and what you produce for your customer. This is helping PMs understand the skills that they should be attaining. — Tamar Yehoshua

> [17:00](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1020s) I'll be honest, we haven't figured it out. I don't think anyone has figured it out. — Tamar Yehoshua

> [20:30](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1230s) each role is expanding. — Tamar Yehoshua

> [20:30](https://www.youtube.com/watch?v=BtK4kFI1LNo&t=1230s) this is the best time in the world to be a PM — Tamar Yehoshua

## Mentioned Tamar Yehoshua, Mike Cannon-Brookes, Atlassian, Lenny and Friends Summit, Rovo, Confluence, Remix with Rovo, Confluence Slides, Rovoclaw, Jira, Jira Service Management, Teamwork Graph, AI Fluency Index, AI Builder Weeks, Arize, Figma, Loom, Slack, X, FedRAMP

## Video description What does the “AI builder” era look like for product managers inside a large company? At the Lenny and Friends Summit, Atlassian’s Tamar Yehoshua shares examples from Confluence, a new product, and Jira to show how PMs can help teams move faster. She explains when it makes sense for PMs to write code and prototypes, when they should focus on steering the team, and how Atlassian is helping them build new AI skills.

Recorded live at Lenny and Friends Summit on September 10, 2026, in San Francisco.

[Source](https://www.youtube.com/watch?v=BtK4kFI1LNo)

中文简报

AI代理将交付速度提升三倍，Atlassian推动产品经理参与编写代码人工智能将拓展而非取代产品管理角色，因为自动化上下文引擎消除了行政琐事，让产品经理得以直接编写代码或主导工程流水线。 Yehoshua认为，AI并没有瓦解或取代专业化角色，反而扩大了产品经理的影响范围和杠杆效应，同时保留了其寻找产品与市场契合度以及交付客户价值的核心使命。她结合Confluence、从0到1的项目Rovoclaw以及Jira旧有代码库的案例，阐释了“划船”（亲自动手构建原型并提交代码）与“掌舵”（为工程师排阻、整合用户反馈并确定优先级）之间的动态平衡。她详细介绍了Atlassian如何通过Teamwork Graph为企业团队提供组织上下文，并通过AI流畅度指数和动手实战的“构建者周”（Builder Weeks）提升产品经理技能。最终，她给出了乐观的预期，宣称AI让当下成为历史上成为产品经理最令人兴奋的时代。产品角色正在拓展而非走向趋同：AI工具赋予了产品经理、设计师和工程师更广泛的职能范围，同时保持了产品经理发现产品与市场契合度及客户价值的核心职责不变。 (2:04) Atlassian采用“划船与掌舵”框架来平衡产品经理的技术参与度，允许非编程背景的产品经理在安全环境中提交前端PR，而在工程人力充足时退后一步负责掌舵。 (5:24) Jira通过结合Loom产品规格说明、云端托管的编程代理以及基于代理的客户反馈分拣，在其拥有20年历史的旧代码库上将开发吞吐量提升了三倍，并在10周内交付了22项功能。 (12:24) Atlassian没有用AI原生人才替代现有员工，而是通过五级AI流畅度指数和每季度的AI构建者周（AI Builder Weeks）提升其产品组织的技能，吸引了1,000多名员工构建了120多个内部工作流。 (17:00)

Reading summary

Tamar Yehoshua, Chief Product Officer at Atlassian, argues that artificial intelligence is expanding rather than eliminating the product manager's role, shifting practitioners away from administrative coordination toward deeper technical execution and strategic steering. While the core mission of discovering product-market fit remains unchanged, teams achieve greater velocity by combining model intelligence with organizational context. Rowing Versus Steering Across Product Lifecycles (2:04) Yehoshua frames operational product management through a nautical choice: deciding whether to "row" by contributing directly to execution or "steer" by directing and unblocking the team. In Confluence's AI feature development, a non-technical PM chose to row, using a sandboxed frontend harness to merge 26 pull requests in a month to address interface polish. Combined with Figma Model Context Protocol (MCP) integrations that automatically fixed approximately 14 visual bugs per hour and AI-assisted test generation, the team reduced launch cycles for Remix and Confluence Slides from roughly six months to six and eight weeks respectively. Conversely, greenfield 0-to-1 efforts require PMs to recognize when rowing becomes counterproductive. On the Rovoclaw project, a PM and designer initially "vibe coded" a functional internal alpha from scratch. However, once dedicated software engineers joined the project, the PM realized that continuing to write code distracted from aligning priorities and clearing road-blocks. Steering accelerated the project, as the PM leveraged technical empathy from early coding to guide architecture while automating status reports using an internal agent. Scaling Legacy Systems with Cloud Coding Agents (12:24) In mature enterprise platforms like Jira—which spans two decades of development, isolated cloud tenants, and FedRAMP compliance—direct PM code contributions are restricted due to production stability risks. Instead of authoring production commits, the Jira team overhauled its delivery pipeline around managed cloud coding agents. PMs record Loom walkthroughs of interface concepts, which automatically generate Jira tickets; transitioning an item to "In Progress" prompts a cloud agent to generate design-compliant, accessible prototypes directly within the front-end repository, preventing throwaway mockups. Automated pipelines also absorb high-volume feedback. Internally, a Jira Slack agent categorizes employee bug reports and routes them directly to coding agents for automated patching. Externally, a Rovo agent ingested and synthesized customer insights from more than 900 user research videos inside Jira Service Management. These agentic workflows contributed to a roughly 3x throughput improvement from in-progress to shipped, delivering 22 user-facing features within 10 weeks. Structural Shifts in Daily PM Responsibilities (16:18) The expansion of the PM role is underpinned by institutional context engines like Atlassian's Teamwork Graph, which reduce administrative overhead like chasing cross-functional launch dates or recording notes. As routine information gatekeeping declines, PM responsibilities are shifting toward rigorous technical validation: authoring model evaluation benchmarks (evals) to systematize output quality, using observability platforms like Arize to diagnose prompt regressions, and rapidly parsing massive qualitative customer datasets that were previously impossible to review manually. Upskilling Through the AI Fluency Index (17:00) Rather than relying on hiring "AI native" outsiders, Atlassian focuses on upskilling its existing product organization through an internal AI Fluency Index. The framework tracks six technical competencies across a five-point scale, ranging from Level 1 ("Curious") to Level 3 ("Capable," the expected baseline) and Level 5 ("Pioneering"). Yehoshua emphasizes that the index serves as a capability roadmap rather than a performance review ladder, which remains tied to customer outcomes. To institutionalize these skills, Atlassian runs quarterly AI Builder Weeks where PMs and designers pause standard roadmaps to complete peer-led workshops and hands-on capstones focused on prototyping, evals, agent construction, or code commits. The initiative has trained over 1,000 employees and yielded more than 120 production workflows, with Atlassian releasing the curriculum externally as an open-source guide. Yehoshua acknowledges that measuring the organizational impact of AI remains an unresolved challenge across the industry, warning against shallow metrics like raw pull requests in favor of cycle times, production deployments, and OKR delivery. Rejecting predictions of occupational displacement, she contends that product management will persist and thrive over the coming decade as AI tooling amplifies human leverage.

Source quotes

[0:06] Your job is going to go away. [2:04] Tamar Yehoshua: everything that you read on X, don't believe. [2:04] Tamar Yehoshua: The other day, last week, Mike, our CEO, DM'd me and one of the product leads and says, 'When is some feature launching?' Five minutes later, he sent a screenshot of Rovo, which is Atlassian's AI product, and said, 'This is what Rovo said. Is it right?' We're like, 'Yeah.' And he said, 'Sorry, I wasted your time.' [10:10] Josh: Wait a minute, should I be spending my time coding? Is this actually the highest leverage activity that I could be doing? [10:10] Tamar Yehoshua: I hope most of you are not writing weekly reports by hand anymore. And if you are, you probably don't want to admit it, but please go and build an agent to do it. [16:18] Tamar Yehoshua: I would love to say that there are no synchronous meetings, but that would be an exaggeration... but there are, I believe, fewer synchronous meetings. [17:00] hire people that are AI native [17:00] Tamar Yehoshua: The ladder is about outcomes and what you produce for your customer. This is helping PMs understand the skills that they should be attaining. [17:00] Tamar Yehoshua: I'll be honest, we haven't figured it out. I don't think anyone has figured it out. [20:30] Tamar Yehoshua: each role is expanding. [20:30] Tamar Yehoshua: this is the best time in the world to be a PM

Original transcript

[0:06] Wow. It's so amazing to be in a room [0:08] with so many awesome product leaders. I [0:10] think this is the first time I've been [0:12] together with so many product people in [0:13] one room. Uh and so many of you I [0:16] recognize and know it's it's really [0:19] amazing. And with many of you, I've had [0:21] conversations over the last year or two [0:23] about what is our job and what is the [0:26] role of a PM. So that's what I'm here to [0:29] talk to you about. What are we actually [0:32] seeing practically on the ground? If you [0:35] remember back a year or two ago, if we [0:38] can all remember that far back, everyone [0:40] was pointing at everyone else's craft [0:42] and saying, "Your job is going to go [0:44] away." PMS were telling engineers their [0:47] job was going to go away. PMs, designers [0:50] were telling PMs, everyone was saying [0:52] that everyone's job is going away. And [0:54] everyone was actually really nervous [0:56] that their job was going away. [0:57] Thankfully, we are not hearing that [0:59] anymore. What are we hearing today? [1:03] Well, each job is overlapping. In fact, [1:07] and there's a new era of the AI builder. [1:10] If you listen to Lenny Lenny's podcast, [1:12] if you talk to people at startups, if [1:14] you read tweets on X, you will see that [1:17] a lot of people are talking about the AI [1:18] builder. And the AI builder job is real. [1:21] There are startups that are not hiring [1:23] PMs or not hiring designers. They are [1:25] hiring builders. There's job [1:27] descriptions. There's builders in small [1:29] companies. There's builders in large [1:30] companies. And if you think about it, [1:32] this has always been the case for [1:33] founders. Founders have always when they [1:36] started a company been the builder and [1:38] really worn lots of hats. So I do [1:41] believe that this is something that is [1:43] very real that's happening and also very [1:45] exciting. But what does this look like [1:48] in a company with over 10,000 people? [1:52] I'm sure many of you work in large [1:54] companies that are not five person [1:55] startups or 10 person startups and you [1:57] listen to these podcasts and you kind of [1:59] scratch your head and you say that would [2:01] be really cool to work like that. I [2:02] don't have that reality. So what I'm [2:04] going to do is I'm going to talk to you [2:06] about what it looks like at Atlassian. [2:08] I'm not going to go into what the [2:09] engineering teams are doing. Of course [2:11] they're doing a lot. They're using [2:13] coding agents. They're refactoring their [2:15] code bases. There's a lot of work going [2:16] on the on the engineering side. I'm [2:18] going to focus on the PM and what's [2:21] happening there. As we said, the roles [2:24] are overlapping, but they are also [2:27] expanding. So that's what we're seeing [2:30] is the people everyone can do so much [2:34] more than they that they could do before [2:36] and it's exciting. So the job of a PM is [2:40] really the same as it always was. What [2:42] is your job? It's to find product market [2:44] fit. It's to build products that people [2:46] love. It's also to make sure you're [2:48] building a business that people will [2:50] actually pay for your products. None of [2:52] that has really changed. It's how we do [2:56] it that's changed. And what has changed? [2:59] Obviously, AI tools have gotten much [3:02] better and continue to get better at a [3:04] rapid rate. And one of the questions [3:06] that I get a lot from some of the people [3:08] in this room is like, how do you stay on [3:09] top of it all? And the answer is you [3:12] don't. So, [3:14] I don't think anybody could. But the tip [3:16] is that everything that you read on X, [3:19] don't believe. [3:22] You have to figure out what's actually [3:24] working for your customers. So focus on [3:26] the customer and what they can get out [3:28] of the AI tools. The other thing that is [3:31] changing is context. You've been hearing [3:33] all about context graphs. You've been [3:35] hearing about how you need to have an [3:38] enterprise brain. Well, this is very [3:40] real. You now can see through lots of [3:42] different tools the context of your [3:44] organization. And at lastian we built [3:45] what we call the teamwork graph that you [3:47] can see the con full context of your [3:49] organization. So what does this mean for [3:52] the PM? Well the PM used to be the [3:54] person taking notes in meetings, [3:57] following up on action items, [3:59] communicating with go to market when [4:00] things are going to launch. They don't [4:02] have to do that anymore. We have tools [4:05] that can do that. The other day, last [4:07] week, Mike, our CEO, DM'd me and one of [4:10] the product leads and says, "When when [4:12] is some feature launching?" Five minutes [4:14] later, he sent a screenshot of Rovo, [4:17] which is Atlassian's AI product, and [4:19] said, "This is what Robo said. Is it [4:21] right?" We're like, "Yeah." And he said, [4:23] "Sorry, I wasted your time." And so, [4:25] what we are seeing is people have access [4:26] to this information, but we also have to [4:28] help them along so that they actually [4:30] use it. There's change management [4:32] involved, but PMS have a front row seat [4:36] into how all of this is changing. And [4:39] one of the things that PMs can do is [4:41] focus on acceleration. We've all been [4:44] hearing how things are moving so much [4:46] faster, but the PM can really help [4:48] accelerate what's happening in their [4:50] team. At Alaskan, we think about it as a [4:53] ship that is going. So you have a ship [4:56] that is going through to get to your [4:58] destination. How are you going to get to [5:00] your destination faster? One of the [5:03] things is intelligence plus context. So [5:06] the intelligence of the models plus the [5:08] context of your organization, how can [5:10] you put that together to move faster? [5:13] Now going back to the ship metaphor, are [5:16] you as a PM, are you rowing or are you [5:19] steering? What is the best way for you [5:23] to make the ship go faster? So we've [5:25] done a bunch of experiments. We've [5:27] organized teams in different ways, [5:28] measured outputs, and the answer that we [5:31] have come to is it depends. It depends [5:35] what type of product it is. It depends [5:37] on what phase in the product it is. So [5:39] what I want to do now is actually talk [5:42] through three concrete examples and what [5:44] the role of the PM was in each one. Now [5:47] one is a new feature in an existing [5:49] codebase. Another is a 0ero to1 product [5:52] and another is enhancements in a very [5:54] large existing codebase. All the [5:55] examples are obviously using Atlassian [5:57] products because that's what we use [5:59] internally. But of course you can use [6:01] all of this is true with whatever [6:02] products your company uses. So we're [6:04] going to start with Confluence. So [6:07] Confluence recently launched two [6:09] features called remix with robo and [6:11] confluence slides. So remix is you [6:13] highlight a piece of text, you remix it [6:15] into multiple different formats. You [6:17] choose the format and slides is you take [6:19] a page and you click a button and it [6:21] converts it into beautiful slides. [6:23] Obviously these are you heavily based on [6:26] AI. When the team started, they didn't [6:28] just say, "Here's what I want to build." [6:31] They took a challenge of how do I build [6:33] it much faster and how do I [6:35] significantly change how I work? And [6:37] again, in a big company, this is [6:40] different than you have a clean slate [6:43] and you're just starting. So, I'm going [6:45] to go through what are some things that [6:46] changed for the confluence team. So, [6:48] first PM started checking in code. I'm [6:51] sure if I asked you to raise your hand, [6:52] many of you are checking in code as [6:54] well. This PM Ya had never written any [6:57] code, had never even used a terminal [6:59] before, but she sat down with the [7:01] engineering partner who actually built a [7:03] harness for her to use that made it [7:06] easier for her to check in code in the [7:08] front end. And so sometimes you have to [7:10] partner with somebody else to figure out [7:12] how you can leverage the tools the best. [7:15] She ended up checking in 26 PRs in a [7:17] month, which was more than most of the [7:19] engineers on the team. I asked her why. [7:22] Why did she decide she was going to [7:23] check in code? and she said there were [7:25] so many things we needed to fix in the [7:27] UX and we didn't have enough engineers [7:29] to do it and I wanted the engineers to [7:30] focus on higher level activities and [7:33] this is something I could do. So she [7:35] felt that this would accelerate the [7:36] project. Next is EVELS. [7:40] So I assume I hope most of you [7:43] understand that EVELS is a significant [7:45] part of your job now and that you're [7:47] training yourselves or taking classes on [7:49] how to do EVELs. different teams. [7:51] Sometimes the PMs do eval, sometimes the [7:53] engineers do evals. Well, in this [7:54] instance, the evals for something like [7:56] confluence slides was so important to [7:58] get right. That the PMs had done them [8:00] traditionally and they worked with the [8:02] engineers to make sure that they relayed [8:05] how they wanted the team to eval this [8:08] product and then they got a 2x [8:09] throughput in eval. The PMS in turn [8:13] started using the LLM platform. We use [8:15] arise at Atlassian for debugging. So [8:17] they were using the platform to find [8:20] issues with the prompts and then send it [8:22] to the engineers again faster. I think [8:24] this one is my favorite. They automated [8:26] the fixing of design bugs. How many [8:28] times have you seen the code and it [8:30] didn't match the design? So they used [8:31] Figma MCP and mapped the designs from [8:35] Figma to the actual code and then [8:38] automated the fixing with a coding [8:40] agent. And so they could fix something [8:41] like 14 bugs in an hour. So this again [8:45] made the product much better. The last [8:47] one was test creation. This the [8:49] engineers went from creating a test in a [8:51] half day to 10 minutes. Just huge [8:54] throughput improvements. So in this [8:56] example, the remix launched in 6 weeks, [8:59] Confluence launched in 8 weeks. Now if [9:01] you're at a startup, you might be like, [9:02] "Oh my god, that's still a long time." [9:04] But remember, this is enterprise [9:06] software that has to be compliant, has [9:08] to be secure, and sometimes doing this [9:11] stuff takes the longest to get done to [9:13] make sure your enterprise customers can [9:15] use it. So what would have taken before [9:18] AI about 6 months now took 6 weeks with [9:22] much tighter iteration loops. So the [9:25] iteration loops with the customers, with [9:28] the the beta customers, and internally [9:30] just fix things much faster. So taking a [9:33] step back here, what was the role of the [9:36] PM? The PM was actually rowing and [9:39] checking in code. And this is what she [9:42] decided was the fastest thing to make [9:44] the ship go faster. The engineering [9:47] manager gave her a clean isolated [9:49] repository, which is very important. So [9:52] nothing got messed up in the larger [9:53] repository. And the team up front had a [9:57] very intentional contribution uh model [10:00] of who was going to do what and they [10:01] blurred the roles by design and they [10:04] said we're going to question existing [10:06] processes. So this is one example. [10:10] Next is a project called Roclaw. Now [10:13] Robo is our product. Not to be confused. [10:15] This was not a feature on Robo. This was [10:17] a zero to one pro project from the [10:20] ground up written. Now it started with a [10:24] PM Josh and a designer Kevin. They vibe [10:27] coded the whole thing. They from soup to [10:29] nuts got to a working alpha that they [10:31] got customers using internally. And once [10:34] they saw they were onto something, then [10:37] they added engineers to the project. So [10:39] then we got the PMs and the designers [10:42] were still checking in code and the [10:44] engineers were working as well. But Josh [10:47] started to see that the engineers [10:49] started veering off track because he was [10:51] busy writing code and he one day is like [10:54] wait a minute should I be spending my [10:56] time coding? Is this actually the [10:57] highest leverage activity that I could [10:59] be doing? So he decided to step back [11:02] from the coding and to focus on what a [11:05] PM usually does setting direction [11:08] prioritizing and unlocking. And then [11:10] things started moving much faster. The [11:13] fact that he had started out by coding [11:16] significantly helped him because he [11:18] could understand what the blockers were. [11:21] He could understand what the challenges [11:23] were and it made for a much better [11:25] environment of how he could steer the [11:27] team. And then on the flip side, he also [11:30] used AI to help him not do other things. [11:34] I hope most of you are not writing [11:36] weekly reports by hand anymore. And if [11:38] you are, you probably don't want to [11:39] admit it, but please go and write build [11:41] an agent to do it. So Josh used uh [11:44] Rovoclaw and built an agent in the [11:46] product he was developing to give the [11:48] weekly updates on itself which were [11:50] super amusing and so he no longer writes [11:53] weekly updates. So in this instance the [11:56] PM [11:58] started out rowing and then shifted to [12:01] steering because it was a different [12:03] phase of the project and that's what he [12:05] needed to do to make the project go [12:07] faster. So he was unfocused on [12:10] unblocking the team whatever he needed [12:13] to do at that point in time and he had [12:16] the ability to contribute code which [12:18] really helped him. So he focused on his [12:21] highest leverage activities. Okay, last [12:24] example, Jira. Most of you have probably [12:26] heard of Jira. Many of you have used it. [12:29] Jira has been around for over 20 years. [12:32] It's a huge complicated codebase. We [12:35] have things like it started on data [12:37] center and then moved to the cloud. It [12:40] it supports isolated cloud. It supports [12:42] Fed ramp. It supports the [12:43] customizations. and you have we have [12:45] hundreds of thousands of customers that [12:48] you can't mess up. So this instance of [12:51] PM checking in code can be kind of [12:53] dangerous. So we didn't want to do that. [12:56] But we also had to bring Jira into the [12:59] AI age. And our goal was to make it AI [13:02] first. And so the PMs and the engineers [13:05] were building AI features in Jira to [13:08] help the customers use AI to do their [13:10] software development, which was really [13:12] fun. like how are you reinventing the [13:14] software development life cycle with the [13:16] tools that you're building and how do [13:18] you make sure you can assign to agents [13:19] how do you make sure you can seam [13:21] seamlessly use agents for everything in [13:23] Jira so this was a big project for the [13:26] team and from in progress to shipped [13:30] their throughput was about 3x what it [13:32] normally was and they shipped in about [13:35] 10 weeks 22 userfacing features how did [13:39] they do it what were the changes they [13:41] made so let's start with prototyping. [13:43] Again, everybody here is prototyping [13:45] that. I'm 100% sure. But what did they [13:48] do to make the prototyping more [13:50] effective? So, first they started with [13:53] Loom. And in a Loom video, you could [13:55] record the UI that you wanted to change [13:58] or you could record here the Figma [14:00] designs with a talk track or you could [14:02] just brainstorm that goes into a [14:04] recording and then Loom has the ability [14:06] from the recording to automatically [14:08] create work items. And then from the [14:11] work item the PM can move it to in [14:13] progress and then that automatically [14:16] kicks off a coding agent. So this whole [14:20] thing was very streamlined and the [14:22] coding agent wrote code in the actual [14:24] front-end repository [14:27] and used a managed agent in the cloud. [14:29] So PMS didn't have to deal with their [14:31] local environment. So then you got code [14:34] that was already compliant, accessible [14:37] and in the Atlassian design language. So [14:40] when they were done, if they wanted to [14:42] implement the prototype, it was much [14:44] more seamless and the handoff to [14:46] developers was much faster. So having [14:50] the right environment for prototyping is [14:53] key of getting it faster into the [14:55] engineers hands. So this was a big [14:57] unlock. Next was triaging feedback. As [14:59] you can imagine, if you're use if you're [15:02] making changes in Jira at a company like [15:04] Atlassian, everybody in the company is [15:05] using it and you get tons of feedback. [15:08] So, what did we do? We took the feedback [15:11] and the bugs that came in through Slack [15:13] and we used the Jira agent in Slack to [15:16] triage them and then automatically send [15:19] them to the coding agent for fixes. [15:22] Again, saving a huge amount of time for [15:24] engineering. And then we also did a lot [15:25] of user studies for customer insights. [15:28] So here we used Jira service management [15:32] to take and review all of the videos [15:35] from the customer insights. We got over [15:37] 900 pieces of feedback. We use used a [15:39] robo agent to then triage and categorize [15:42] them. And we were very happily surprised [15:44] with the quality that we got of the [15:47] triage. So then we knew what to send to [15:49] engineering to fix. All of these things [15:52] together in this instance the PM was [15:55] clearly steering. The PM had to focus on [15:59] unblocking the engineers in order to [16:02] make the project go faster. That [16:05] included processing feedback internally, [16:07] externally more effectively, and [16:10] building prototypes that were set for [16:12] reuse. And as I said in the beginning, [16:15] PMS did not check in code to the [16:16] production environment. So, if you put [16:18] all this together, what are things that [16:20] PMs are doing now that they weren't [16:22] doing before? They're obviously [16:24] prototyping and testing. They're writing [16:25] evals. They're going deeper on customer [16:27] feedback. They're consuming information [16:29] at a much higher rate rate. And all of [16:32] this works if all the AI tools [16:34] understand the context of your [16:35] organization. And that is built on top [16:38] of at Atlassian the teamwork graph. And [16:40] then what are they not doing? They are [16:42] not doing manual updates. They're not [16:45] compiling research by hand. They're not [16:47] creating SLI uh slides from scratch. I [16:50] would love to say that there are no [16:52] synchronous meetings, but that would be [16:53] an exaggeration. uh but there are I [16:56] believe fewer synchronous meetings. [17:00] Well, how did we do this? What was our [17:01] playbook at Atlassian? When I talk to [17:04] people at startups, they tell me hire [17:06] people that are AI native. Well, we have [17:08] a lot of PMs already who are amazing PMs [17:11] at Atlassian. So, we want them to become [17:13] AI native. We want to help them be able [17:16] to learn the skills that they need to [17:18] learn uh in order to build products for [17:20] our customers. So we introduced [17:22] something that we call the AI fluency [17:25] index and this is to help people [17:27] understand where they need to go and we [17:30] have six capabilities in the fluency [17:34] index. Things like do you understand how [17:36] to use the tools? Do you understand how [17:38] to write emails? Do you understand how [17:40] to automate data insights? Can you [17:42] prototype? And do you have the level of [17:45] technical literacy that you need? And [17:47] then we have an index from one to five. [17:49] One is curious, five is pioneering, [17:53] three is capable, and we do not expect [17:56] every PM to be a five at everything. And [17:58] this is not a ladder. This is for [18:01] development. So that you have a [18:02] northstar on the skills that you need to [18:05] be developing because of course this [18:07] changes all the time. The ladder is [18:09] about outcomes and what you produce for [18:11] your customer. This is helping PMS [18:14] understand the skills that they should [18:15] be attaining. We do expect people to get [18:18] to an L3 in all of these over time and [18:21] you should understand what your team [18:23] needs. You should understand what is the [18:25] most important point for your team and [18:27] that you focus on getting to an L5. How [18:30] do we train PM? So we want them to [18:32] become AI fluent. Uh we've tried a bunch [18:36] of different things and the thing that [18:37] stuck the most was our AI builder weeks. [18:40] So, our AI builder week is once a [18:42] quarter. We take a week where you're not [18:45] doing your regular work and we train PMs [18:48] and designers. It starts with bringing [18:51] in external speakers. Shout out to [18:53] Claire who came to our last builder [18:54] week. And then we have PMs internally in [18:58] the organization who are more advanced [19:00] in certain skills training other PMs. [19:04] And then we have people do a project. So [19:07] it depending on what it is, they can [19:08] work with their teams, they can work [19:10] with other people that but they do a [19:12] project to use the skills they're [19:13] learning. This has been incredibly [19:15] successful. We've had over a thousand [19:17] people go through our builder weeks. [19:20] We've built over 120 new workflows that [19:22] people are using after the builder [19:24] weeks. So what are the focus and we [19:26] focus on one thing each time. So the [19:28] first one was prototyping. The second [19:30] one was uh eval. Third one was how do [19:33] you build agents? And then the the most [19:35] recent one was checking in code. We have [19:38] talked to customers a lot about this and [19:40] we've actually built an AI builder week [19:42] in a box. You can find it in our website [19:44] where it shows you kind of the agendas [19:46] that we've had and how we've utilized [19:48] them. Okay. So now we've told the PMS [19:50] where what they're going to achieve. [19:52] We've trained them. And what is a [19:53] question I get from customers all the [19:55] time is how are you measuring outcomes? [19:58] I'll be honest, we haven't figured it [20:00] out. I don't think anyone has figured it [20:02] out. I talked to a lot of people about [20:04] this and everyone is I mean it's the [20:06] age-old how do you measure developer [20:08] velocity. Um we are measuring PRs [20:11] deployed not PRs written but PR is [20:13] deployed to production. We're measuring [20:15] features delivered. It's a little tricky [20:17] to figure out when the feature started [20:19] but from idea to delivery in the [20:22] customer's hands and usage. Obviously [20:24] you want to have the right OKRs. You [20:26] want to make sure that your OKRs being [20:28] met and like overall throughput. [20:30] Sometimes it's easier to measure the [20:31] throughput of an individual team versus [20:34] an organization. So you want to do both. [20:36] You want to do make sure teams are [20:37] working faster and organizations are [20:39] working faster. So if you have ideas of [20:42] how you're measuring, let me know. But [20:43] we are working actively each quarter. We [20:45] have experiments that teams are running [20:47] in different ways and we look at what we [20:49] can measure on the outcome. So bringing [20:52] this all together as I said the roles [20:55] are expanding each role is expanding and [21:00] I think the way I look at it PMS can do [21:02] more to get their jobs done more [21:05] effectively and better and they can do [21:07] more to deliver more to their customers. [21:10] I think this is the best time in the [21:12] world to be a PM. I think it's super [21:15] exciting. I am an optimist and I think [21:17] we'll all be here 10 years from now. But [21:19] I really think that this you guys are [21:21] all so lucky to be here today. So that's [21:26] it. Thank you.