Agent-native debugging: why we built Glass

Most session replay tools assume a human will sit down, scrub a timeline, and eyeball what went wrong. That workflow doesn't map onto how coding agents operate.

An agent wants compact, structured, token-efficient answers: which errors are happening, what sequence of events led to a failure, and a screenshot of the exact moment — not a multi-megabyte replay blob.

Glass treats the agent as a first-class consumer. Every capability is an API and MCP tool first, a dashboard view second. Recording is free and unlimited; we charge for the agent usage that actually saves engineering time.

Built on Cloudflare, the economics work: zero egress on R2, cheap columnar search in ClickHouse, and on-demand screenshot rendering that only runs when an agent explicitly asks for it.