Atlassian’s Head of AI: We Bolted AI Onto 20+ Apps, Almost Didn’t Ship Chat, and Flipped Our Hiring Toward Juniors

We talk a lot about 'AI natives' in the crypto and dev world. It is easy to build something cool when you are starting from a blank terminal. It is a different beast entirely when you have to shove large language models into 20 legacy applications that millions of people actually rely on for their daily work. Sherif Mansour, who has been at Atlassian for nearly two decades, recently pulled back the curtain on how they did exactly that. It was not a smooth ride, and for builders looking to scale, there are some hard lessons in their messy transition.

The Multi-Product Trap

Atlassian is not one company; it is a federation of tools. You have Jira, Confluence, Trello, and a dozen others. Mansour is responsible for the product management craft for about 450 PMs. When the generative AI wave hit, they did not have the luxury of picking one winner. They had to figure out how to make AI feel native across a portfolio where some apps were born in the cloud era and others are practically legacy software by modern standards.

The biggest mistake most founders make is thinking they need a 'feature.' Atlassian realized quickly that users do not want an AI button; they want their work done. They ended up bolting AI onto 20 different apps. Six of these were built specifically during this new era, but the rest required major surgery. If you are building in the space right now, pay attention to that ratio. If you are spending 70% of your time fighting your own architecture to make an LLM work, you might be building on a foundation that is too brittle for the next decade.

The Chat Trap

One of the most revealing things Mansour shared was that Atlassian almost did not ship a chat interface. In the early days of ChatGPT, every SaaS company panicked and threw a sidebar chat into their app. It became a meme for a reason. Atlassian's internal debate was whether a chat box actually solved a problem for a Jira user or if it was just UI theater.

They eventually shipped it, but only after they realized that chat is not a feature—it is a navigation layer. For builders, the takeaway here is skepticism. Just because OpenAI provides a chat completion API does not mean your product needs a text box. Sometimes the best AI is the one the user never sees, like a background process that automatically links Jira tickets or summarizes a Confluence page without being asked. If your primary value proposition is 'Chat with your X,' you are competing with every other company on earth. Find the friction points in the workflow instead.

The Hiring Flip: Why Juniors Win Now

Perhaps the most controversial move Mansour described was their shift in hiring. Historically, big tech companies have been obsessed with 'Senior Plus' talent. They wanted the specialists who had seen it all. But Atlassian is flipping that script, moving their hiring focus toward juniors. This is not just a cost-saving measure; it is a structural bet on the future of work.

Mansour's logic is simple: junior developers and PMs do not have to unlearn the old way of doing things. They are using Copilot and ChatGPT from day one. They do not view AI as an 'add-on' to their workflow; they view it as the baseline. In a world where an LLM can handle the heavy lifting of boilerplate code and initial product specs, the value of a senior who refuses to adapt drops significantly. For founders, this is a signal. Stop overpaying for 'expertise' that was earned in a pre-LLM world if that expert is resistant to changing their process. You want people who can prompt, iterate, and ship at 10x speed, regardless of their years on a resume.

The Internal 10% Rule

How do you get 450 PMs to care about AI without it becoming a distraction? Atlassian did not just send a memo. They embedded it into the 'craft.' They looked at how much time was being spent on 'toil'—the boring stuff like writing status updates or documentation. They aimed to automate at least 10% of that immediately.

For builders, this is the 'eat your own dog food' phase. If your team is not using AI to build your AI product, you are going to lose. Mansour's team uses these tools to manage the very products they are building. This creates a feedback loop that you cannot get from user surveys. If a feature feels clunky to an Atlassian PM, it is definitely going to feel clunky to a customer.

Platform Over Features

The secret sauce for Atlassian was not the individual AI features in Trello or Jira; it was the 'Atlassian Intelligence' platform. They built a centralized layer that handles the prompts, the models, and the data privacy. This allowed their individual product teams to move fast without reinventing the wheel every time they wanted to add a summarization tool.

If you are a founder building a multi-tool ecosystem or even just a complex app, stop building one-off AI features. Build a small internal service that handles your LLM calls and context injection. It makes your app more modular and allows you to swap out models—moving from GPT-4 to Claude or an open-source Llama—without breaking 20 different front-end components. Atlassian’s scale proves that the 'platform' approach is the only way to survive the rapid pace of model updates.

The Reality Check

Despite the success—over 5 million users now interact with these AI features—Mansour is honest about the fact that they are still in the early innings. They are still figuring out what sticks. The most important metric they watch is not 'AI usage' but 'user task completion.' If the AI is being used but tasks are taking longer, the AI is a failure.

As builders, we need to get over the novelty. The era of 'wow, the computer talked back' is over. We are in the era of utility. Atlassian’s journey shows that even the biggest players are stumbling their way through this, but the ones who win will be the ones who focus on the unglamorous work of integration and the willingness to hire a new generation that works differently.

The goal is not to be an AI company. The goal is to be a company that solves problems so efficiently that the user forgets AI was involved at all.

The Takeaway for Builders

• Ignore the UI hype: Don't ship a chat bot just because you can. Look for the 'toil' in your user's day and kill it.
• Hire for the future: Look for juniors who are 'AI-fluent' over seniors who are 'AI-resistant.'
• Centralize your intelligence: Build a platform layer for your LLM logic so you can iterate on models without touching your core UI.
• Focus on completion: If your AI doesn't help a user finish a job faster, it's just noise.

Read the original at SaaStr →