From Startup to AI-First: How Braze CTO Jon Hyman Scaled Engineering for the Agentic Age

In the fast-moving world of software engineering, fifteen years is an eternity. It spans the birth of the mobile era, the rise of cloud-native architecture, and now, the seismic shift toward artificial intelligence. For Jon Hyman, co-founder and CTO of Braze, these milestones have been more than just historical markers; they have been the defining challenges of his career.

Recently, Hyman joined Jody Bailey, Chief Product and Technology Officer at Stack Overflow, on the Leaders of Code podcast to discuss the rapid evolution of Braze’s engineering organization. The conversation provided a rare, candid look at how a 300-person engineering team transformed into an AI-first operation in a matter of months, moving beyond mere code completion to the deployment of autonomous, feature-building agents.

The Evolution of Engineering Leadership

To understand Braze’s current AI strategy, one must first understand its roots. Founded nearly 15 years ago, Braze emerged alongside the mobile revolution, forcing its engineering team to prioritize real-time interactivity and global scale.

Hyman describes his leadership style as that of an "on-the-ground general." Rather than operating from a remote executive office, he maintains deep technical proficiency, often jumping into the weeds to prototype solutions or help teams break through architectural gridlock. As the company grew from a small startup to a global leader, Hyman’s role evolved from hands-on coding to managing a complex, divisional structure.

"The evolution has been that engineering managers and divisional leads need to manage increasing levels of complexity," Hyman explains. "But I still believe that to be effective, you must be able to go toe-to-toe with your best builders. My proficiency in AI has allowed me to remain motivating and credible in those high-level architectural conversations."

Chronology of an AI Transformation

The transition to an AI-first organization was not an overnight directive, but a calculated progression of enablement, experimentation, and eventual integration.

Phase 1: Enablement and Awareness (February 2025 – May 2025)

The catalyst for Hyman was the release of advanced coding tools in early 2025. After experimenting with Claude Code shortly after its release, Hyman was struck by the model’s efficacy. His initial strategy was simple: remove friction. Braze purchased enterprise licenses for Cursor, expanded access to GitHub Copilot, and provided access to various models via AWS Bedrock.

Phase 2: The "Greenfield" Proof of Concept (August 2025)

The turning point for the team occurred in August 2025, when Braze tasked a small team with building an MCP (Model Context Protocol) server. Rather than using traditional methods, the team utilized AI tools for the entire project. The result was a delivery six weeks ahead of schedule. This success proved to the organization that AI was not just a productivity "nice-to-have," but a transformative force for engineering velocity.

Phase 3: Scaling and Adoption (November 2025 – Present)

The release of more advanced models, specifically Opus 4.5 in November, solidified the shift. The technology moved from needing constant human guidance to being capable of building complex, meaningful features with minimal correction. Today, over 60% of the code committed to Braze’s main repositories is AI-generated, a metric that continues to climb as the "flywheel effect" takes hold among the engineering staff.

Supporting Data: The Reality of Inference Costs

While the productivity gains have been staggering, Hyman is quick to warn against the "vibe-coding" trap. As CTO, he is keenly aware of the financial implications of scaling AI agents.

"We are seeing that engineers are changing how they work," Hyman notes. "Some are working on models for six or more hours a day. It’s expensive."

He shared a cautionary anecdote about a single engineer generating $150 in inference costs in a single day. Across a 300-person organization, this represents a massive, unforeseen budget shock. This reality has forced Braze to pivot from "AI at all costs" to a more disciplined approach:

• Cost-Efficiency: Developing strategies to choose the right model for the right task to optimize token consumption.
• Measurement: Shifting from measuring code volume to measuring the cost of inference against the business value delivered.
• Standardization: Moving away from the "wild west" of individual experimentation toward a standardized "agentic infrastructure" that governs how models are used across the enterprise.

Strategic Integration and Acquisitions

A significant test of Braze’s engineering culture came with the acquisition of OfferFit, a reinforcement learning engine company. Hyman noted that despite the differences in size—Braze with 1,800 employees and OfferFit with roughly 170—the cultural alignment was immediate.

Braze integrated the OfferFit team by aligning them with existing divisional structures. "We move them over to our Jira system and our PR management workflows," Hyman said. "But we also learned from them. We adopted their use of Graphite for stack pull request management, which we are now trialing across the entire organization."

This ability to integrate, learn, and iterate is, in Hyman’s view, the mark of a mature, healthy engineering organization. It confirms that while AI can speed up the build, it cannot replace the necessity of sound organizational processes.

Implications for the Future of Software

Hyman’s outlook for 2026 is one of rapid, agent-driven development, but he cautions against the notion that software engineering is becoming obsolete.

The End of "Vibe-Coding"

"You cannot ‘vibe code’ your way to scale," Hyman asserts. He argues that modern, high-complexity systems require a human architect who understands not just the code, but the business domain, the customer pain points, and the system dependencies—context that currently exceeds the capacity of any model’s context window.

Competitive Velocity

Hyman believes that AI has effectively induced a "singularity event" in software production. Because competitors now have access to the same 2x or 3x productivity multipliers, the bar for the entire industry has been raised. "If you had 100 engineers and now have the output of 180, you don’t stop working," Hyman explains. "You simply tackle more of your massive, ever-growing roadmap."

The Road Ahead: Autonomous Agents

The next frontier for Braze is the deployment of autonomous agents capable of "working while the engineers sleep." This includes:

• Automated Bug Resolution: AI agents that can identify a bug report, write the fix, and submit a pull request without human intervention.
• Self-Service Design: Product managers and designers using AI to resolve minor UI/UX debt without needing to pull engineers away from core feature development.
• Continuous Improvement: Using AI to ensure that documentation, testing, and scaffolding remain consistent with internal standards, preventing the buildup of "AI-generated spaghetti code."

Conclusion: A Call to Action for Leaders

For those leading engineering teams, Hyman’s message is clear: do not wait for the perfect model. The era of AI-first engineering is not about replacing developers; it is about raising the expectations of what a team can achieve.

"If you are not launching agents in parallel and having them work on different features, you are falling behind," Hyman concludes. "My advice is simple: tinker. If you don’t have an AI app on your home screen, put it there. Start engaging with these tools as your primary interface. The transformation of your business starts with your own personal mastery of these tools."

As Braze enters this next chapter, the focus will remain on the delicate balance of velocity, cost-efficiency, and, most importantly, delivering value to the customer—a mission that remains unchanged, even as the methods to achieve it become increasingly autonomous.