|
This week, we’re exploring the system Atlassian created for employees to build agents — and how it gained so much popularity, it eventually became an external product. “People tell me they’ve heard this buzzword, ‘agent,’ and they ask me if an agent can do a certain task for them,” says Sherif Mansour, Atlassian’s Head of AI. “It’s best to start by asking them what they actually want to do, because an agent might not be the answer.” That’s the delta between executives pushing their companies to use AI and employees building tools that are impactful — often, they don't know what they need, and if they do, there’s no systematic way for people to create it. Atlassian faced this problem at scale. People wanted to use AI, they just didn’t know how. So the company established a team led by Mike Bentley, AI Principal Program Manager, responsible for supporting internal AI use cases across the 13,000-person global company. You can imagine how this went. “It was a complete, chaotic mess,” says Mansour. “We’d get a crazy number of Slacks, whether that was directly to me or Mike, or in the general AI help channel.” Bentley created an internal tool to build out all these requests, and also developed (and optimized) a ticketing system to manage them. Then he realized something. “When you think about good software, it’s scalable and extensible,” he says. Through the hundreds of requests, he noticed patterns in what people were asking for: pulling information from specific sources or pushingdata to another platform. This led him and Mansour to completely change their approach to building internal agents and workflows. Bentley rolled out his internal tool to the whole company with one key change. He gave everyone the pieces they needed to compose their own agents, calling them “AI LEGO blocks.” Users combine the different parts of an AI workflow — knowledge, instruction and action — to create custom agents. “We realized all the capabilities we built and how we built them were the ingredients for how to make AI work in general,” Bentley says. “Make your data available. Make sure your tools are available in ways AI can talk to them. And design and orchestrate those specific agents to work with the data and the tools — that’s how we rinse and repeat.” Atlassian now has more than 3,000 agents working alongside its 13,000 employees. One of its agents, Customer 360, saves account executives 1 - 2hrs of compiling a data report about a customer each time it’s used — and last month alone, it was used by 102 users 1,690 times. The internal agent-builder gained so much momentum that Mansour worked to turn it into a customer-facing product: Rovo. In this week’s essay, we sit down with Bentley and Mansour to learn how Rovo was built and scaled, the creation of its LEGO blocks and the most powerful use cases they've seen so far. Thanks, as always, for reading and sharing, -The Review Editors
|
