Overview:
Working at Atlassian
Atlassians can choose where they work – whether in an office, from home, or a combination of the two. That way, Atlassians have more control over supporting their family, personal goals, and other priorities. We can hire people in any country where we have a legal entity.
Responsibilities:
About the team
------------------
Confluence is where organizations write down what they know: decisions, architecture, process, and policy. But a company's work does not happen in one tool. It happens in file drives, chat, code hosts, CRMs, and data platforms.
The Aggregation team makes Confluence fill and organize itself from those tools. Content arrives with its source system's permissions intact, gets organized so a team can navigate it, and stays current as the underlying work changes. That content is what Confluence's AI answers are built on, which makes its coverage and quality the difference between an answer a team acts on and one they go verify by hand.
About the role
------------------
You will own how Confluence turns connected content into knowledge that maintains itself: what a space pulls in, how it gets organized, how it stays current, and when it flags that something important is missing. Vision through launch and iteration, on problems that arrive undefined.
Quality is the job here, not a stage of it. Content that is stale, mis-grouped, or wrong does more damage than content that was never connected at all, because a team will have already acted on it.
You are building for three audiences, and the role requires you to name which one you are serving in every decision: the admin who approves connecting a system of record and will disconnect it the moment trust breaks, the space owner accountable for a body of knowledge staying current, and the end user who wants the answer without knowing which tool it came from.
### **What you'll do
- •Own the strategy and roadmap for what connected content becomes inside Confluence: what gets pulled in, how it is grouped, titled, and summarized, and what happens to content that has gone stale or contradicts a published page.
- •Define how messy inputs become usable structure. Documents, spreadsheets, meeting transcripts, chat threads, and CRM records do not arrive as pages, and deciding what shape they take in Confluence is the core of the work.
- •Own freshness and decay. Decide what updates on its own when a source changes, what needs a human to confirm, and how a user can tell which one they are looking at.
- •Set and defend the quality bar for retrieval: permission fidelity to the source, accuracy, and latency. Decide what does not ship, and hold that line when a date is at risk.
- •Own the quality evals for the parts of the experience powered by a large language model. Define what a correct output looks like, build the test sets, catch regressions as models and content change, and decide when output is good enough that a user stops double-checking it.
- •Decide where AI belongs in the experience and how it behaves when it is wrong. A suggestion that costs nothing to ignore and a suggestion that hijacks what someone was doing are the difference between a feature teams keep and one they switch off.
- •Drive execution across Engineering, Design, Data Science, Research, Marketing, Partnerships, and Security and Trust. Bring clear recommendations to executive forums.
- •Set the product bar. Mentor PMs and raise the quality of specs and decisions across the team.
### **What we're looking for
- •8+ years in product management, including at least one product where content, retrieval, or a model-powered experience was the core value rather than a feature added later.
- •You have taken a 0 to 1 product from strategy through launch and past the first disappointing metric.
- •You have shipped features powered by large language models and owned their evals. You can walk through how you defined a correct answer, built and maintained the test set, caught a regression before customers did, and made the call to ship or hold.
- •Fluency with unstructured and semi-structured content. You know why a meeting transcript, a spreadsheet, and a database record each behave differently when a system splits, retrieves, and summarizes them, and what breaks when it treats them the same.
- •AI product taste. Every product a knowledge worker touches will have AI features, and they get cheaper to build every year, so which ones are right becomes the differentiator. Shown two working versions, you can say which is better for a specific job, persona, and moment, and why. You know what a wrong output should cost the user, you keep the cognitive cost of a suggestion low, and you retire AI surfaces that stopped earning their place. Craft without impact is theatre, and we hire for both.
- •You understand retrieval-augmented generation, where the system looks up relevant documents before it answers, and why coverage and permissions decide whether an answer is