3 AI Lessons From Atlassian: Ship Chat, Copy Humans, Hire Juniors

The popular executive line on AI hiring is to skip juniors and hire one senior with AI tools. Atlassian, a company with thousands of R&D staff and more than 20 apps, went the other way. Its head of AI explained why at SaaStr AI, along with a candid list of the AI bets the company got wrong first.

The Framework

Sherif Mansour has spent 17 years at Atlassian and now runs AI across its product portfolio, including the product management practice of roughly 450 PMs. In SaaStr’s write-up of his session, the unifying idea is that existing users and workflows are an asset. Atlassian bolted AI onto products people already used, learned from what they did, and then built dedicated features. Each lesson below came from a wrong assumption that usage corrected.

1. Ship Chat, Then Let Prompts Pick the Roadmap

Adding Rovo chat to every app was controversial internally about two years ago. It shipped broadly mostly because Atlassian needed the chat backend anyway. The team assumed chat was temporary. Instead, millions of users use it daily, and the prompts became a product research feed. On Confluence whiteboards, a common prompt (group these sticky notes into themes) became a button. A long prompt to turn notes into Jira, Trello, or Linear backlog items became a workflow. Pulling customer feedback from Salesforce and Google Drive stayed a prompt. Mansour’s advice: if users are in your product every day, ship chat and read what they type.

2. Give Agents the Same Primitives as Humans

About 2.5 years ago, Atlassian added one option to the Jira automation designer: use a Rovo agent at this step. Customers went well beyond it, chaining agents with branches and conditions. In one example, a marketing agent calls Canva to create assets and a social agent publishes them. Roughly six months ago, Atlassian concluded that agents need what humans already have: tools, context, goals, accountability, and visibility. Agents can now be @mentioned, assigned Jira tickets, added to workflow steps, and reached in Slack or Microsoft Teams. The design review rule: if an agent does something a human could not do in the product, ask why.

3. Hire for Both Ends of the Experience Curve

Atlassian ran about 10 projects staffed with hand-picked “AI builders.” They moved fast for a few weeks, then stalled. In Mansour’s telling, everyone was rowing and no one was steering. PMs and designers drifted back to their roles because supplying customer context and decisions unblocked teams most. With AI, the effective ratio of PMs or designers to engineers now feels closer to 1:30 or 1:40, up from about 1:10 on product teams.

Hiring flipped too. Atlassian moved from a few juniors and many mid-level and senior hires to a pipeline weighted toward both ends. Internal research found juniors are up to 38% more likely to use AI and nearly twice as likely to experiment, but they produce a lot of low-quality output. Seniors adopted more slowly but are better at spotting weak output. Every couple of months, the whole R&D org runs a synchronized four-day AI Builder Week where juniors teach tools and seniors teach quality control.

Where It Doesn’t Apply

Mansour is explicit that everyone can be a builder in a small startup, but not in mid-size or large organizations. The bolt-on approach also assumes an installed base of daily users. Atlassian builds from scratch only for new products, such as its six AI-native apps. A pre-product-market-fit company has no workflows to learn from. For context on how the market is pricing that installed base, see Stack Archive’s look at Atlassian at $6.6B ARR.

Source: SaaStr

Content created with AI assistance and reviewed for accuracy.