Jaya Gupta

Jaya Gupta is a partner at Foundation Capital, where she leads early-stage investments across enterprise software. Her AI interests span the infrastructure powering machine learning and generative AI’s ability to change enterprise applications and how software is coded. With Ashu Garg, she has argued that AI startups can build durable advantages by capturing the decisions and feedback generated inside customer workflows.[1][2][3]

In their December 22, 2025 essay, “AI’s trillion-dollar opportunity: Context graphs”, Gupta and Garg distinguish general rules from records of how those rules were applied in particular cases. They define a context graph as “a living record of decision traces stitched across entities and time so precedent becomes searchable.” This records inputs, policies, exceptions, approvals, and resulting changes; it is distinct from a model’s chain of thought.[2]

Their illustrative renewal scenario makes the distinction concrete. An agent proposes a 20% discount despite a policy cap of 10%. It gathers three severe incidents from PagerDuty, a cancellation escalation from Zendesk, and an earlier renewal thread documenting a similar exception. Finance approves the request. The CRM records the final discount; the decision trace preserves the evidence and approval that justified it.[2]

Gupta and Garg argue that startups operating in the agent orchestration path can capture this context as decisions happen, whereas downstream warehouse ingestion may preserve outcomes after the decision context has been lost. Their proposed approach begins with human approval: the agent gathers context, proposes an action, routes approvals, and records the trace. As precedent accumulates, more of the workflow can be automated. They identify exception-heavy work and functions spanning multiple systems—such as revenue operations and incident response—as promising places to build.[2]

An earlier essay with Garg, published November 5, 2025, addresses the risk that a startup’s model provider becomes its competitor. Their recommendation is to build learning systems around judgment-heavy, customer-specific processes whose interaction data providers cannot easily replicate. Domain-specific evaluators, rubrics, and reward functions define success and turn observed outcomes into feedback.[3]

Their advice extends beyond model training: own difficult execution surfaces, hire domain experts, and work alongside customers. They cite insurance startup Fulcrum’s use of human adjusters for ambiguous cases as a way to learn exception handling. In sales, they describe Regie’s customer collaboration and emphasis on conversions per representative per channel, giving agents a business-outcome target rather than simply rewarding emails sent.[3]

In a January 27, 2026 investment account coauthored with Joanne Chen, Gupta recalled: “I first met Arjun Mangla when we worked together at McKinsey.” She emphasized Mangla’s attention to messy enterprise workflows and described his work with cofounder Sambhav Anand as addressing the gap between how insurance brokerages should operate and how they actually do.[4]

Chen and Gupta said Foundation first met Fulcrum before it had customers, led its seed round two years before the announcement, and invested again alongside a CRV-led Series A. They credited the whole Foundation team with customer introductions, advisor sourcing, and help recruiting its first head of sales. Their investment argument applied the context-graph thesis to insurance: capturing workflow decisions could make institutional knowledge structured and searchable.[4]

Before her investing work, Gupta worked in McKinsey’s technology practice, conducting software diligence and helping startups develop go-to-market plans. She initially joined Foundation as chief of staff focused on portfolio operations and strategy, and built communities around university networks including IIT, Berkeley, and Georgia Tech. These earlier roles provide context for her work with founders and enterprise operations.[1]

Gupta’s startup experience began at age 19, when she built CardioBuddy, which failed. While studying at Georgia Tech, she also joined a startup that was later acquired. She worked at two startups that were acquired.[1]