Gil Feig is cofounder and CTO of Merge, which connects software products and AI agents to business systems. He writes about building reliable integrations.[1]
Start with What everyone is getting wrong about context graphs →
X · gilfeig Career
Feig cofounded Merge with Shensi Ding after both encountered integration problems in earlier jobs. In their joint interview, they describe founding the company in June 2020. Feig previously led engineering at Canvas, and a later podcast biography identifies engineering leadership at Untapped and software engineering roles at Wealthfront and LinkedIn. The platform’s original unified API let developers add integrations across business software without implementing each provider separately.[4][3]
Contributions
His Merge MCP announcement explains exposing integration endpoints as tools that a language model can call. The relevant contribution is an API and integration team adapting access to existing business systems for agent software. Feig’s context-graph essay extends that problem beyond connectivity: a system must select a useful, permitted slice of information rather than fetch everything it can reach.[2][1]
Ideas
Feig argues that a context graph is an operating system for choosing what information to fetch for each request. His examples involve checking freshness, permissions, latency and API limits before combining data from business systems. Provenance records what the model saw and under whose authority.[1]
Tools, writing & talks
Writing, research & talks
Start here
Explains live data selection, permission checks and provenance as the operational core of assembling context for AI.
Read the context-graph essay first for a concrete account of why connecting tools is only part of agent engineering. Familiarity with APIs helps; no graph database background is required.
Author[1]Merge's CTO lays out a context graph design: synced copies for semantic queries, live calls for lookups, a router and summarizer, company skills and memory, and traceability of where data came from and what invalidates it.
Catalog workOriginal source- We’ve launched Merge MCPmerge.dev
Describes how Merge exposes integration endpoints as tools through the Model Context Protocol.
Author; company product announcement[2] A conversation about maintaining integrations, data engineering and the evolution of software ecosystems.
Interview guest[3]