Person
1 episodes · 13 quoted claims
Sherif Mansour describes an AI maturity pyramid in which layer one is asking questions and layer two is having AI generate artifacts.
“bottom layer, most people just ask a question, get an AI from an answer, layer one. Layer two is like, oh, hang on a second, AI can do some work for me”
Atlassian's Rovo chat tool has built-in personal memory that learns a user's preferences, such as Sherif Mansour's habit of putting emoji in headings.
“we've got inbuilt personal memory. So over time we start to remember, you know, for me, I like to write with a lot of emoji.”
Sherif Mansour says Atlassian's products such as Confluence and Jira default to open access, and he argues this helps AI give useful answers to new employees.
“our products are open by default. And I have to explain that a little bit. When you create a documentation space in Confluence or a Jira project or whatever it is, the default is open”
Atlassian's Teamwork Graph maps users, teams, goals and the work items related to them.
“we have what we call the team graph, which is an understanding of all the users, the teams, their goals, and how the work relates to that.”
Sherif Mansour says Atlassian uses retrieval-augmented generation rather than its graph for simple lookups such as a company's annual leave process.
“what's the process for taking annual leave in my company? RAG's probably a much better solution there. We don't traverse the graph for that.”
Atlassian operates an AI gateway that proxies a mix of locally hosted and cloud models so it can swap models and test alternatives.
“We have an AI gateway, and we have a mix of locally hosted models, models in the cloud.”
Sherif Mansour says adding AI agents to Jira workflows is probably the biggest or fastest growing use of agents among Atlassian customers.
“adding an AI agent in a Jira workflow, which is probably the biggest or fastest growing use case of agents in deployment”
A customer described to Sherif Mansour that using AI for simple string checks in a workflow was like swatting mosquitoes with a rocket launcher.
“I've got a rocket launcher and I'm swatting mosquitoes with it.”
Atlassian's agent framework lets customers give agents skills, which can be no-code or code, including deterministic functions.
“Our agent framework lets people create skills that the agent has. The skills can be no code or code skills.”
Sherif Mansour says general-purpose models solve 80 to 90 percent of typical knowledge-worker tasks and that customer demand for choosing a specific model is declining.
“the general purpose models are 80% to 90% easily solving the problem and we're seeing that desire for model selection just decline over time”
Sherif Mansour says Apple is encouraging app developers to build app intents, which let AI such as Siri use an app's functions as tools, rather than focusing only on user interfaces.
“Apple's got a big push right now with their app vendors of like, hey, spend time focusing on building app intents over UIs.”
Sherif Mansour describes a five-person sales AI startup that uses AI agents in a Jira workflow to publish content to its WordPress blog, with humans reviewing it.
“they have AI agents in their Jira workflow that automatically publish content to their WordPress”
Atlassian ran an AI Builders Week in which about a thousand product managers, designers, engineering leaders and researchers blocked time to experiment with AI.
“we just ran a AI Builders Week, a thousand product managers, designers, engineering leaders, researchers.”