Prukalpa Sankar

Prukalpa Sankar is a co-founder of Atlan whose public work focuses on giving enterprise AI access to organizational meaning: business definitions, rules, relationships and accumulated judgment. In her 2026 writing, she argues that stronger models cannot substitute for this knowledge. Atlan identified her as co-founder and co-CEO in May 2026.[2]

Sankar frames the problem as “Performance = Intelligence × Context” in her Knowledge Graph Conference talk walkthrough, published May 14, 2026. The formula expresses her argument that reasoning capability needs situated organizational knowledge. Her illustrative support representative, Maya, becomes effective through training, supervision and corrections; Sankar contrasts that accumulated experience with agents repeatedly arriving without the business context needed to act well.[2]

In her March 2026 explanation of context engineering, she distinguishes three kinds of context: structural, such as definitions, relationships and semantic models; operational, such as procedures, approval logic and exceptions; and behavioral, such as preferences and historical decision precedents. She acknowledges the overlap with data governance, while arguing that agents require definitions to be machine-readable and retrievable, alongside rules and experience that a glossary alone does not capture.[3]

Her proposed context layer sits between an enterprise’s data estate and its AI agents, connecting metric definitions, business rules, cross-system entities, skills and data graphs. Sankar highlights two problems that shared infrastructure should address: a correction to one agent failing to reach another, and agents interpreting the same business term differently. Her revenue example contrasts “closed-won” with “recognized”: each interpretation can appear plausible in isolation but conflict when agents hand off work.[2]

Sankar recommends starting from knowledge already encoded in metadata, SQL, lineage, semantic models and documentation, then refining it with domain experts. Tests should check whether an answer matches the business’s meaning, beyond whether its SQL executes. Human reviews, approvals and corrections supply further context signals. As definitions change, she advocates versioning, testing, governance, approval and rollback practices so improvements propagate across agents.[3][2]

Her practical starting point is a struggling AI project in one domain. She recommends selecting 10 to 20 important definitions, metrics, rules and relationships, making them retrievable, and first enabling answers to three common questions. She also makes her commercial position explicit: “Full transparency: Atlan builds context infrastructure. We have a stake in this category.”[3]

Sankar spoke about context engineering at AI Engineer World’s Fair 2026. In Latent Space’s conference account, Richard MacManus attributed to her a description of context flowing from business systems into a shared company brain and outward to agents, copilots and applications through MCP, APIs and retrieval. Her May conference essay also calls for knowledge graph practitioners to help shape enterprise AI, arguing that their work on meaning, entity resolution and lineage supplies important foundations.[1][2]

Before her 2026 advocacy for enterprise AI context infrastructure, Sankar co-founded SocialCops with Varun Banka in 2013. YourStory’s February 2018 profile identifies both founders. Banka is also credited alongside Sankar as an Atlan co-founder.[4][2]