Sam Fertig is a deployed engineer at Cognition, helping enterprises adapt the Windsurf and Devin coding agents to established software environments. His work addresses a practical obstacle to autonomous programming: useful code must fit an organization’s existing repositories, infrastructure, development conventions, and legacy systems.
After studying at Oberlin College and working at C3 AI, Fertig became a deployed engineer at Windsurf. His AI Engineer World’s Fair 2025 talk described how coding assistants infer developer intent from both immediate behavior and surrounding technical context. At Cognition, he has worked with teams at Brazil’s Itaú Unibanco on enterprise AI engineering; a colleague involved in Devin deployments identified him as the team’s COBOL specialist.
- Developer state and codebase state: Effective coding agents combine behavioral signals—including cursor position, recently opened files, terminal activity, and clipboard contents—with repository code, documentation, organizational rules, and agent-generated memories.
- Relevant context over larger context windows: Expanding model inputs cannot substitute for finding the right information quickly. Fertig emphasizes embedding search, reranking, remote indexing, and syntax-aware parsing to improve relevance without sacrificing responsiveness.
- Legacy-system modernization: Enterprise deployment extends beyond editor integrations into older languages such as COBOL, institutional infrastructure, security requirements, and established engineering workflows.
For Fertig, coding-agent performance depends on whether generated software reflects the developer’s immediate intent and the technical constraints of the organization receiving it.
