How Braze’s CTO is Navigating the Shift to Agentic AI Engineering

How Braze’s CTO is rethinking engineering for the agentic area

Braze CTO Jon Hyman transformed his 300-person engineering organization into an AI-first team in just months. Over 60% of Braze’s committed code is now AI-generated following the adoption of advanced models like Opus 4.5.

Why This Matters

The technical reality of AI-first engineering involves a shift from simple code completion to autonomous agents, yet it introduces significant financial friction. Hyman notes that inference costs can reach $150 per day for a single engineer, challenging the “vibe-coding” narrative with the necessity of optimizing token spend and infrastructure. Furthermore, while AI accelerates greenfield projects—like an MCP server finished six weeks early—it cannot “vibe-code scale.” High-complexity, high-traffic systems still require deep domain knowledge and architectural oversight that current context windows cannot fully encapsulate.

Key Insights

• An MCP server project built entirely with AI finished six weeks ahead of schedule at Braze in 2025.
• Opus 4.5, released in November 2025, marked a shift from guided assistance to autonomous feature building with minimal human correction.
• Token costs for active AI-assisted development can reach $4,500 per month per engineer according to Jon Hyman’s 2026 projections.
• Designers use Vercel v0 and Cursor to self-service UX debt and interactive mock-ups without engineering intervention.
• Braze acquired OfferFit in 2024 to integrate reinforcement learning and specialized machine learning divisions.
• Over 60% of code committed to Braze’s main repositories is now AI-generated as of early 2026.

Practical Applications

• Use case: Designers at Braze use AI to resolve ‘UX debt’ items, such as adding dynamic links to error messages. Pitfall: Relying on vanilla AI scaffolding can introduce anti-patterns that violate internal React standards.
• Use case: Product managers use Vercel and Cursor for rapid prototyping to show customers interactive mock-ups before full engineering commitment. Pitfall: Assuming AI-generated code is production-ready for high-scale environments without rigorous architectural review.
• Use case: Engineering teams use Graphite for managing stacked pull requests to maintain velocity during high-volume AI code generation. Pitfall: Unmonitored inference spend during agentic workflows can lead to significant budget shocks.

References:

• https://stackoverflow.blog/2026/05/13/rethinking-engineering-for-the-agentic-area/