The "Cofounder" team replaced traditional Product Requirement Documents (PRDs) with code. An idea was built directly into a feature, and the resulting pull request served as the tangible proposal for discussion. This compressed the spec-to-build cycle and forced concrete, implementation-level decision making.
Contrary to the belief that building agents requires complex frameworks, Gusto built their product using just a Cloudflare Worker for the agent loop and the Vercel AI SDK for model switching. This minimalist stack proves that the barrier to entry for building powerful agentic products is surprisingly low.
Instead of creating Figma mockups, the designer on Gusto's team built and shipped a "faked" UI with hardcoded responses directly to production behind a feature flag. This functional UI skeleton served as a living specification that engineers could then progressively wire up to real backend logic.
With AI coding assistants, the cost to write code is so low that teams can afford to build entire features and then delete the pull request if the idea isn't right. This "trash can method" makes code a disposable medium for product exploration, not a precious asset to be preserved.
The designer on the "Cofounder" team, though not a classically trained engineer, became one of Gusto's most prolific coders. She ranked in the 94th percentile for code throughput, a feat attributed to her technical curiosity, AI tools, and the team's commitment to mentoring and reviewing her code.
Gusto's CTO used a 5-hour flight layover and Claude Code to personally build the initial prototype for what became "Gusto Cofounder." This hands-on "vibe coding" by an executive served as the catalyst for the entire project, bypassing formal roadmaps and inspiring the team with a tangible starting point.
Gusto's "Cofounder" team achieved a median PR review time of just nine minutes, facilitated by a constant "PermaZoom" room where reviews could be requested and conducted instantly. This proves that ultra-fast human feedback loops, not just AI code generation, are the true enabler of rapid development.
While traditional Test-Driven Development (TDD) can be cumbersome, an "eval-driven" approach is perfect for AI. The workflow: 1) write a failing evaluation to reproduce a conversational bug, 2) prompt an AI to fix the underlying code/prompt, and 3) confirm the fix by re-running the eval to ensure it passes.
To build its "Cofounder" product in 10 weeks, a 5-person team at Gusto removed all traditional development artifacts. Their only structured process was a 24/7 "PermaZoom" room for constant communication, proving extreme process minimalism can accelerate zero-to-one product development, even inside a large company.
Gusto's CTO recognized his team's "no docs" method succeeded because his co-founder status gave him implicit permission to break rules. To scale this, leaders can't just allow new methods; they must explicitly forbid old ones, telling teams, "you will get a slap on the wrist if you produce a doc."