Matt Dailey is the founder and chief executive of Ref., which builds tools to help software teams preserve shared context and human judgment as coding agents accelerate implementation. His central concern is that faster individual engineers can leave their organizations buried in pull requests, competing priorities, and consequential decisions hidden inside disposable agent conversations.
Dailey studied at Carnegie Mellon University and worked on search and indexing at Palantir before joining Figma, where his work included infrastructure, growth engineering, applied AI, and the company’s public API. In 2018, he introduced an integration connecting Figma with the animation tool Principle.
While at Figma in late 2024, Dailey experimented with connecting its API to the emerging Model Context Protocol. A February 2025 hackathon produced Ref.’s initial MCP documentation server, designed to supply coding agents with accurate, relevant library documentation without exhausting their context windows. His account of these early projects also describes experiments with voice interfaces, sales-research agents, background agents, and interactive components. After initially building Ref. independently, he brought in former Figma colleague Suvir Copparam.
The engineering problems he focuses on
- Information access constrains autonomy. Models cannot reliably use unfamiliar libraries if they lack current documentation and precise API details. Dailey advocates specialized documentation retrieval that supplies relevant information efficiently across multiple agents.
- Velocity sickness means output without impact. AI-generated code can overwhelm reviewers, scatter teams across conflicting initiatives, and obscure architectural choices. Dailey measures productivity by whether faster work improves the product, not by the volume of code produced.
- Agent bankruptcy wastes work and context. Abandoning tangled sessions forces engineers and agents to reconstruct decisions, repeat research, and spend additional tokens. His alternative uses durable decision documents to preserve project goals, technical context, tradeoffs, and agreed decisions independently of any individual conversation.
- The decision layer belongs to humans. Dailey separates collaborative planning and engineering judgment from agent-driven implementation. Shared plans let teams review important choices earlier, coordinate parallel agents, and increase idea velocity by exploring possibilities before committing to prototypes. His AI Engineer World’s Fair talk frames human ownership of consequential decisions as essential to retaining ownership of the product itself.
Dailey has also addressed the economics of paid MCP services, proposing usage-based credits to support the indexing and retrieval costs behind reliable agent infrastructure.
