Braze CTO Jon Hyman shares how the company transformed its 300-person engineering org into an AI-first team over roughly three months. Key milestones included early adoption of Claude Code (which he credits as a game-changer), shipping an MCP server six weeks ahead of schedule using only AI, and reaching a point where over 60% of committed code is AI-generated. Hyman discusses the cultural challenges of driving adoption, noting that improved model quality — not mandates — won over skeptics. He also addresses the steep and often underestimated cost of LLM inference at scale (one engineer spending $150/day in tokens), why 'vibe coding your way to scale' is naive, and how autonomous agents building features overnight is the near-term future. He argues AI is inducing demand for more software rather than reducing headcount needs.
Sort: