In an era where artificial intelligence is reshaping industries, Jon Hyman, co-founder and CTO of Braze, has steered the company’s engineering organization through nearly 15 years of growth. Recently, he led a bold transformation—making the team AI-first in just a few months. This Q&A explores how Braze is rethinking engineering for the agentic era and what other leaders can learn from their journey.

What prompted Braze’s engineering team to adopt an AI-first approach?

The shift was driven by the rapid emergence of agentic AI—systems that can act autonomously to achieve goals. Jon Hyman recognized early that traditional engineering processes, built for deterministic outputs, would struggle in an environment where AI agents make decisions in real time. Braze’s platform, which powers customer engagement for brands, needed to become smarter and more adaptive. The team saw an opportunity to leverage machine learning not just for recommendations but for full-fledged autonomous actions. Hyman also noted that competitors were moving fast, and staying relevant meant embracing AI at the core. The decision wasn’t just about adding features; it was about rethinking the entire engineering mindset—from how code is written to how systems are tested and deployed. This urgency led to a company-wide initiative to become AI-first, fundamentally changing how Braze engineers approach problems and build solutions.

How did Jon Hyman lead the transformation in just a few months?

Hyman focused on three pillars: culture, tooling, and experimentation. First, he communicated a clear vision to the entire engineering org—that AI would be embedded in every product and process. He encouraged teams to experiment with generative AI and large language models, providing resources and sandbox environments. Second, he invested in new internal tooling, such as AI-powered code assistants and automated testing frameworks that could handle probabilistic outputs. Third, he established rapid feedback loops: small, cross-functional squads would build prototypes, test them with real customers, and iterate in days rather than weeks. Hyman also led by example, personally coding alongside engineers and showcasing AI integrations. By removing bureaucratic hurdles and fostering a “fail fast” culture, he compressed what could have been a year-long transition into just a few months. The key, he says, was not waiting for perfect solutions—instead, shipping early and learning from real-world usage.

What specific changes were implemented to support the agentic era?

Braze made several concrete changes. The engineering team adopted an agentic architecture, where autonomous microservices—called “agents”—could make decisions and trigger actions without human intervention. For example, a campaign optimizer agent can A/B test creative elements and automatically allocate budget to the best performer. They also overhauled their data pipelines to support real-time, streaming data, enabling agents to react instantly. Testing evolved: instead of relying solely on unit tests, they built simulation environments where AI agents could run thousands of scenarios to validate behavior. On the operational side, Braze introduced observability for AI, monitoring agent decisions for bias or drift. Team structures changed too—traditional backend/frontend splits gave way to small, autonomous “agent squads” that own end-to-end features. Hyman also implemented weekly “AI sprints” where engineers could work on passion projects, from natural language interfaces to predictive models. These changes collectively moved Braze from a reactive platform to a proactive, intelligent engagement system.

How does the agentic era differ from previous tech paradigms?

The agentic era represents a shift from software that simply follows commands to software that understands goals and acts autonomously. In previous paradigms—like the web or mobile revolutions—engineers built deterministic systems: user clicks button, system responds. The agentic era introduces probabilistic, self-optimizing systems. An AI agent might decide where, when, and how to communicate with a user based on real-time context, without a developer coding each path. This changes everything about engineering: debugging becomes harder because the system’s behavior can vary; reliability demands new monitoring approaches; and security must account for autonomous actions. Hyman notes that the shift is as profound as moving from mainframes to personal computers. It requires engineers to think like “orchestrators” rather than “builders” of static features. For Braze, this meant rethinking legacy APIs, rewriting backend services for flexibility, and retraining the entire engineering team on AI principles and agent safety.

What role does AI play in Braze’s current engineering strategy?

AI is now the foundation of Braze’s engineering strategy, not an add-on. Every product decision starts with the question: “How can an AI agent improve this?” For example, their messaging engine now uses reinforcement learning to optimize send times per user. The experimentation framework automates A/B test design and analysis using generative models. Even internal tools are AI-first: engineers use custom LLM-based assistants to write documentation, generate code, and triage bugs. Hyman emphasizes that AI is also embedded in the development lifecycle—continuous integration pipelines run AI-driven tests that simulate edge cases. On the infrastructure side, Braze uses AI to auto-scale resources and predict traffic spikes. The company’s long-term roadmap includes fully autonomous campaign management, where AI agents handle everything from creative generation to budget allocation, with humans only supervising. This strategy has already led to measurable improvements in customer engagement metrics and development velocity.

What advice does Braze’s CTO have for other companies undergoing similar shifts?

Hyman offers three pieces of advice. First, start with culture, not technology. The biggest hurdle is getting engineers to embrace uncertainty and iterate rapidly. He suggests creating safe spaces for experimentation—like hackathons or “AI Fridays”—where failure is accepted. Second, invest in observability and safety. Agentic systems can behave unexpectedly, so and teams need robust monitoring, guardrails, and rollback mechanisms from day one. Third, think in terms of agent ecosystems, not isolated features. Design your system so that agents can communicate and coordinate—this unlocks exponential value. Hyman also warns against over-engineering: “Ship a simple agent that works 80% of the time, learn from the 20%, then improve.” He believes the agentic era will reward companies that move fast and put AI at the core of their engineering DNA. For Braze, this transformation was a bet on the future—one that has already paid off in agility and innovation.