Alan Braun

AI capability is advancing faster than organizational productivity. Frontier models have moved from handling tasks that take humans minutes to tasks that take hours. But making individuals more productive does not automatically make teams and companies more productive, especially when work spans functions, systems, and dependencies.

Collaboration has always required transferring context. An engineer and a product manager need to share requirements, technical constraints, and past decisions. Human-to-agent and agent-to-agent collaboration extend that challenge: a coding agent assigned a Jira issue needs the relevant requirements, related bugs, dependencies, and rationale. That context lives across Slack and Teams action items, Google Docs requirements, Jira epics and workflows, meeting notes, and bespoke systems.

Connected context can improve both agent performance and efficiency. In this talk, we share how Atlassian approaches this with a graph that connects context sources and the relationships among people, work, and decisions, so agents get relevant context without brute-force searching every tool. We walk through a recorded benchmark of the same agent answering the same question with siloed tool access versus graph-based context, and show how we score answers with a blind LLM judge on groundedness, accuracy, comprehensiveness, and usefulness. Our internal benchmarks suggest graph-based context improves response quality while reducing token costs.