The Missing Half of AI: Context, Agents, and the AI-Native Enterprise w/ Prukalpa Sankar (Atlan)

TL;DR

• 1

  Context is the missing half of enterprise AI: Sankar says models already ace cognitive benchmarks, but companies need a "context layer" that captures business systems, lineage, metrics, norms, and tacit knowledge before AI becomes genuinely useful.

• 2

  80% accuracy is the tipping point for agents: Below that, people abandon them, as one customer did after launching thousands of agents and seeing 90% abandoned in a month; above roughly 80%, users start collaborating and feedback loops improve performance.

• 3

  Better-described data matters more than perfect data: Sankar argues AI can already write strong SQL and code, but it fails on business meaning, like choosing the wrong join or misunderstanding what "performance" means for a store manager versus a marketing lead.

• 4

  Simulation is how you find missing context before launch: Atlan creates persona-based test environments, sometimes with 100 user personas and likely questions, to see what an agent knows, what it lacks, and where humans need to fill in missing expertise.

• 5

  AI-native companies will collapse middle layers of management: Sankar predicts future orgs will center on customer owners, product or IP builders, and systems builders, with more interdisciplinary "E-shaped" people and fewer traditional managerial roles.

• 6

  Leadership in AI requires hands-on self-disruption: Sankar says leaders cannot delegate this transition, and Atlan made AI use explicit in hiring, performance, and daily work, even declaring that engineers should teach AI to code instead of writing code themselves.