Atlassian's research found junior engineers are up to 38% more likely than seniors to use AI tools in daily work. The company has flipped its R&D hiring pipeline in response, weighting both ends of the experience ladder and cutting the middle.

The data comes from Sherif Mansour, head of AI at Atlassian and the executive who runs the 450-person product management organization. He delivered the numbers at the SaaStr Annual conference. "Juniors are up to 38% more likely to use AI, nearly twice as likely to experiment, and use more tools," Mansour said. They also produce more slop and report lower productivity. Seniors adopt slower but catch slop faster and extract better output, he added.

The conclusion drives a counterintuitive pipeline. "He hears the same line from C-level execs all the time: stop hiring juniors and hire one senior with AI," Mansour said. Atlassian went the other way. The pipeline that used to lean mid-level now leans both junior and senior, with less in between.

What does this change for the rest of B2B product?

Atlassian tested three contested AI decisions across a 20-app portfolio with more than 5 million users on the AI capabilities. Six of those apps were built in the generative AI era; the rest predate it. The team made calls in public, at scale, with real users.

Why Atlassian shipped chat it planned to remove

Two years ago, Rovo chat in Jira, Confluence and the other apps sparked an internal fight. The case against: chat fails as an interface for forms and structured tasks. The case for: nobody knew what users would attempt, so give them an open box and observe. The backend forced the decision. Atlassian needed a chat layer to power AI features anyway. Once it existed, it shipped to all 20+ apps.

Mansour framed chat as the DOS of AI. Command-line interfaces once ran spreadsheets and games before dedicated apps absorbed common use cases. Chat absorbs the unlimited long tail and points engineers at the features worth a real interface. That intuition is now grounded in usage. "Millions of users use Rovo chat every day," Mansour said. The original plan was to learn, build features, then remove chat. That plan died.

How the Jira bolt-on became a portfolio rule

The standard advice in B2B software: never bolt AI onto a legacy product; reimagine from scratch. Atlassian bolted anyway.

Two and a half years ago, the Jira automation designer gained one new box: run a Rovo agent at this step. Customers went further than expected. They added branching, conditions and chained agents. One pattern Mansour described: a ticket arrives, an agent triages it, a marketing agent calls Canva for assets, a social agent publishes them. Workflows that existed for humans were automated for agents.

The portfolio-wide rule that emerged about six months ago: anything a human can do, an agent can do. "@mention a human" became "@mention an agent." Assigning a Jira ticket to a person became assigning it to an agent. The design review test now runs both ways. If a feature lets an agent do something a human couldn't, the team has to explain why.

Why the AI builder experiment stalled

About 10 teams staffed entirely with AI builders — engineers, plus PMs and designers who could ship code with AI — moved fast for weeks. Then they slowed badly. "Everyone was rowing and no one was steering," Mansour said. The PMs and designers drifted back to their old jobs because supplying customer context and making calls unblocked the team faster than any code.

The numbers explain why. Atlassian runs roughly 1 PM and 1 designer per 10 engineers on product teams, and 1:20 on platform teams. With engineers shipping more, the effective ratio becomes 1:30 or 1:40. Engineers come back sooner asking what to build next. A PM vibing in an editor cannot.

Mansour's conclusion on the people question: "hire AI builders if you can find them, but build for a 10x team over a 10x individual." AI builder is now an Atlassian role, but it will not absorb most PMs or designers. In a small startup, everyone builds. In a mid-size or large organization, that model fails.

The response across R&D is operational, not aspirational. Every couple of months the entire R&D org runs AI Builder Week, a four-day synchronized sprint. Juniors present new tools and techniques. Seniors present quality control and slop detection. Atlassian has open-sourced the schedule, sessions and videos. The pattern pairs the cohort that adopts fastest with the cohort that judges output best, on a recurring cadence. Mansour will run AI Builder Week again within two months, with the schedule, program plan, and session videos on a public repository for any product team to clone or contest.