An Engineering Leader's Blueprint for the Agentic Era: Lessons from Braze's AI Transformation

Introduction

As artificial intelligence reshapes the technological landscape, engineering leaders face a critical question: how do you retool a team built for traditional software development to thrive in an agentic, AI-first world? Jon Hyman, co-founder and CTO of Braze, navigated this exact challenge over the past few years. Under his guidance, Braze's engineering organization—spanning hundreds of engineers and nearly 15 years of growth—transformed into an AI-first team in just a few months. This guide distills the actionable steps Hyman and his team took, offering a practical roadmap for CTOs and engineering VPs who want to lead their own organizations into the agentic era.

What You Need

• Leadership commitment from the C-suite and engineering management.
• A baseline understanding of agentic AI (e.g., large language models, autonomous agents).
• An engineering culture that values experimentation and rapid iteration.
• Access to cloud infrastructure and modern AI tooling (e.g., vector databases, model APIs).
• Dedicated budget for training, tooling, and potential re-architecture.
• Cross-functional buy-in from product, data, and security teams.
• Patience and a long-term mindset—transformation doesn't happen overnight.

Step-by-Step Transformation Guide

Step 1: Assess Your Current Engineering Culture and AI Maturity

Before any major shift, Hyman emphasizes the importance of understanding where your organization stands. Evaluate how your teams currently approach problem-solving, the degree of autonomy engineers have, and the existing infrastructure for building AI-powered features. Create a maturity matrix: Are you using basic ML models? Do you have data scientists embedded in product teams? Are engineers comfortable with probabilistic outputs? At Braze, this assessment revealed that while the company had strong data foundations, it lacked a unified strategy for integrating generative AI agents across its customer engagement platform. Document these gaps—they will inform every subsequent step.

Step 2: Commit to an AI-First Mindset from Leadership Down

This step is non-negotiable. Hyman and Braze's executive team publicly declared an AI-first strategy, aligning the entire company around a shared vision. For your organization, this means updating your engineering principles to prioritize AI capabilities in all new projects, rewriting team charters to include agentic objectives, and communicating the shift clearly in all-hands meetings. Leaders should model the behavior—experiment with AI tools themselves, share learnings, and adjust performance metrics to reward AI-driven innovation rather than just traditional feature throughput. At Braze, this top-down commitment created the urgency needed to move from months of planning to execution in weeks.

Step 3: Redefine Team Structures for Agentic Workflows

Traditional engineering teams organized around microservices or feature areas may not optimally support agentic systems. Hyman reorganized Braze's engineering squads to include dedicated AI platform teams, as well as embedded AI specialists within product teams. Consider forming a small, central AI platform team that builds shared infrastructure (model hosting, prompt management, guardrails) and then pairs with product teams to deploy agents into specific use cases. This hybrid model prevents silos while enabling deep expertise. Also, create new roles such as “AI architect” or “agent engineer” to bring structure to the work.

Step 4: Invest in Rapid Upskilling and Experimentation

One of Braze's most effective tactics was launching a company-wide “AI hackathon” within the first month of the transformation. Engineers, even those with no prior ML experience, were encouraged to build small agentic prototypes using pre-approved APIs and LLMs. This hands-on approach demystified the technology and surfaced potential use cases quickly. To replicate this, set aside two weeks for an internal sprint where teams explore agentic capabilities. Provide sandboxes, access to models, and light mentorship. Follow up with formal training programs—certifications, workshops, or partnerships with AI vendors. Hyman reports that this rapid exposure built confidence and accelerated adoption across the entire engineering organization.

Step 5: Build an Internal AI Platform and Toolchain

To scale agentic development, you need a robust platform. At Braze, engineers created internal tools that abstract away complexity—custom SDKs for agent orchestration, a prompt version control system, and reusable guardrails for safety. Your team should identify the most common bottlenecks in building AI agents: model selection, data retrieval, response validation, and monitoring. Then, build or integrate tooling that standardizes these components. Invest in observability specifically designed for AI agents (e.g., tracing LLM calls, tracking agent decisions). This platform acts as a force multiplier, enabling multiple product teams to build AI features without reinventing the wheel.

Step 6: Implement Guardrails and Governance for Safety

Agentic AI introduces new risks—unpredictable outputs, bias, and security vulnerabilities. Hyman made sure Braze established clear governance from the start. Define policies for when an agent can act autonomously vs. when human approval is needed. Create red-teaming processes to test agents before deployment. Build automated content filters and usage quotas. Also, establish an ethics review board that includes engineering, legal, and product leaders. Document your guardrails as code so they can be tested and versioned like any other software component. This step builds trust both internally and with customers.

Step 7: Continuously Iterate Based on Feedback and Metrics

Transformation is never complete. After the initial push, Braze embedded continuous improvement loops: weekly retrospectives on AI features, user satisfaction surveys, and dashboards tracking adoption, latency, and error rates for agents. Establish similar feedback mechanisms for your team. Measure not just engineering outputs (e.g., number of models deployed) but business outcomes (e.g., improved customer engagement, reduced support tickets). Be willing to kill underperforming agents and double down on successful patterns. Hyman stresses that the agentic era demands a “build-measure-learn” pace faster than traditional software development.

Tips for Success

• Start with a high-impact, low-risk pilot to prove value—Braze began with an agent that helped marketers write better campaigns before moving to more complex autonomy.
• Don't underestimate the cultural shift. Many engineers are used to deterministic systems; help them embrace probabilistic thinking through paired programming and code reviews of AI features.
• Watch out for “AI washing”—make sure your agents truly augment human work rather than simply regurgitate data. Focus on real user needs.
• Invest in data quality. Agents are only as good as the data they consume. Clean, structured, and well-governed data will make your AI platform much more effective.
• Communicate early and often with all stakeholders—engineering, product, sales, and customers. Transparency about the capabilities and limitations of your agents builds trust and reduces friction.
• Plan for cost management. Agentic AI can be expensive to run. Implement cost monitoring per agent and create a feedback loop between usage and budget.
• Stay informed about the fast-moving field. Hyman and his team regularly attend industry events and maintain close relationships with model providers to stay ahead of the curve.

By following these steps and adapting them to your organization's unique context, you can lead an engineering transformation that prepares your team not just for the next product cycle, but for the agentic era that is already unfolding.