How product managers can get the most out of AI-native development processes

Watch on YouTube

TLDR

This episode, featuring Avi Zelenko, Principal Product Manager at Atlassian, explores how AI is transforming the traditional software development lifecycle (SDLC). Our discussion focuses on Atlassian’s Prompt-Eval-Iterate loop, using AI with PRDs, the creation and use of “golden datasets,” and the use of LLM judges to deliver higher quality AI products. Product managers will hear actionable insight into AI-native development processes and tips for involving cross-functional teams and customers in the journey.

Introduction

Is the traditional Product Requirement Document dead, along with the standard “Build-Test-Launch” cycle? AI-driven Software Development Life Cycles (SDLCs) are making changes in what has been standard practice. In this discussion we’ll explore the AI-native SDLC used at Atlassian. By the end of this episode, you’ll have a new framework to bring back to your team: The Prompt-Eval-Iterate loop. We’ll discuss why your PRD should be a “behavior contract,” how to build “golden data sets,” and how to use LLM judges to ship higher-quality software faster than ever.

Our guest is Avinoam Zelenko. He is a Principal Product Manager at Atlassian, where he is currently leading the transition to AI-native development for Confluence. With a career spanning leadership roles at LinkedIn and Feedvisor, and years spent teaching the next generation of PMs at Product School, he knows exactly how to bridge the gap between high-level AI strategy and day-to-day execution.

Summary of Concepts Discussed for Product Managers

Evolution of SDLCs:
We discuss the limitations of linear Software Development Life Cycle (SDLC) approaches like “build, test, launch” in the era of AI. Avi explains that product managers must now co-own quality, moving beyond handoffs and static PRDs, as AI-driven features require deeper, ongoing commitment.

Prompt-Eval-Iterate Loop:
Atlassian’s approach starts with collaborative prompt design and exploration, not lengthy specs. Instead of guessing feature outcomes upfront, teams build out golden datasets and use rapid iterations to let real data and metrics refine both the product and its requirements.

Golden Datasets:
A golden dataset is a living collection of well-curated real-world examples and edge cases from customers. It helps teams define what “good” looks like and allows continuous improvement of AI features, with new findings fed back into the dataset for better output and coverage.

Maintaining Customer Proximity:
Avi emphasizes that core product management tasks like customer interviews and understanding unmet needs remain vital. Atlassian leverages AI agents to automate customer feedback loops, enabling PMs to connect with more users and gather data on a much larger scale.

PRD as a Behavior Contract:
The Product Requirements Document (PRD) evolves into a behavior contract, encoding what the AI should do in specific scenarios, along with clear metrics, safety guardrails, and references to the golden dataset. This contract is drafted after substantial hands-on exploration and iteration, keeping specs grounded in reality.

Evals and LLM Judges:
Quality assurance uses two types of evals: deterministic checks (yes/no, hard criteria) and LLM judges (AI-based evaluators) for assessing nuances like faithfulness to source material, narrative, and tone. These automated evals create quality gates for each product milestone.

Collaboration and Transparency:
Atlassian encourages cross-functional teams—from engineering and support to sales and marketing—to participate early in the process. This open, inclusive approach gathers a breadth of perspectives and aligns objectives across the organization.

Useful Links

• Connect with Avi on LinkedIn
• Learn more about Atlassian

Innovation Quote

“Sometimes immersing works better than observing.” – Avi Zelenko

Application Questions

1. How can your team evolve its SDLC to better integrate AI-driven features and ongoing iteration?
2. What would a “golden dataset” look like for your product, and how would you begin building it?
3. In what ways can you involve more customers, support, sales, or marketing in defining the behavior of AI features?
4. How does shifting from a static PRD to a “behavior contract” change your collaboration with engineering and other teams?
5. What new skills or practices must PMs develop to balance automation with human judgment in AI product development?

Bio

Avinoam “Avi” Zelenko is a Principal Product Manager at Atlassian, where he leads product strategy for Confluence, the company’s flagship collaboration platform. With more than 16 years of experience in B2B SaaS, he has built and scaled products at companies including LinkedIn, where he helped shape the feed experience for hundreds of millions of users, as well as LivePerson, ClickTale, and Feedvisor, spanning intelligent chat, analytics, and algorithmic pricing. 

Thanks!

Thank you for taking the journey to product mastery and learning with me from the successes and failures of product innovators, managers, and developers. If you enjoyed the discussion, help out a fellow product manager by sharing it using the social media buttons you see below.

Source