Descriptions:
Prukalpa Sankar, co-founder of data catalog company Atlan, takes the AI Engineer conference stage to argue that the bottleneck for production AI agents is not model intelligence — it’s shared business context. She opens with a striking stat: only one in five AI use cases reaches production, and 56% of CEOs report zero financial benefit from AI today, despite models advancing rapidly. Her thesis is that cognitive intelligence (what benchmarks measure) explains only 10% of real-world job performance; the rest is contextual knowledge built on the job over time.
Sankar traces Atlan’s own internal journey through multiple agent frameworks — Relevance, Google ADK, Glean, and Claude Code, now splitting between Claude and Codex — to show how context becomes stranded in each platform as teams migrate. Agents lived on “islands,” unaware of changes made by other teams, unable to share a version of truth, and impossible to debug when something went wrong. She calls this pattern context sprawl.
The proposed solution is a dedicated context layer: shared infrastructure sitting between agents and the business that stores organizational knowledge — business definitions, diagnostic playbooks, team conventions — in a form all agents can access and update together. Drawing on customers including GitLab, Zoom, Discord, Mastercard, and General Motors, Sankar positions context engineering as the next infrastructure frontier for enterprise AI, analogous to the communication and onboarding systems that help human employees stay aligned across a growing organization.
📺 Source: AI Engineer · Published July 14, 2026
🏷️ Format: Keynote Launch
