Overview of How Braze’s CTO is rethinking engineering for the agentic area
In this Stack Overflow Podcast conversation, Braze co-founder and CTO John Hyman shares how engineering leadership has evolved across Braze’s 15-year journey—from mobile-first scaling challenges to the current shift toward AI-driven, agentic workflows. He explains how Braze is using AI not just for code completion, but for prototyping, bug fixing, QA, internal automation, and broader product development. The discussion also covers leadership structure, post-acquisition integration with OfferFit, the realities of AI costs, and why Hyman believes AI should be used to increase output and speed rather than simply reduce headcount.
Key Takeaways
-
Engineering leadership must stay close to the technical work.
- Hyman sees himself as an “on-the-ground general,” not a detached executive.
- His effectiveness comes from deeply understanding architecture, operational constraints, and how the product actually works.
-
Braze’s org structure evolved to match growth and complexity.
- The company moved from a startup-style pool of generalist engineers to specialized teams and a divisional model.
- Engineering managers now lead through increasingly complex layers, while division leaders own vision, architecture, and business outcomes.
-
AI adoption at Braze accelerated dramatically in 2025.
- The team moved from code completion tools to AI as a true engineering collaborator.
- In some cases, AI now helps generate fixes, handle test failures, and support more autonomous workflows.
-
The biggest shift came from model quality improvements.
- Early AI use was often inconsistent and sometimes created more work than it saved.
- Better models—especially later in 2025—made the value obvious and drove broader adoption.
-
Braze is using AI to increase velocity, not shrink ambition.
- Hyman pushes back on the idea that AI should simply reduce engineering headcount.
- His view: AI expands what teams can build, and companies should use that capacity to ship more.
AI Transformation at Braze
From enablement to agentic workflows
Hyman describes Braze’s AI rollout in three broad phases:
-
Enablement and awareness
- Provide tools like Cursor, GitHub Copilot, Claude Code, and internal training.
- Help employees understand where AI fits into their day-to-day work.
-
Agentic workflows
- Move beyond “AI as advisor” toward “AI as coworker.”
- Examples include:
- Auto-generating bug fixes from reports
- Fixing failing tests automatically
- Building code from Jira epics or projects
- Creating outputs that are closer to finished work, not just drafts
-
Business-metric impact
- Use AI to improve measurable outcomes like ramp time, support efficiency, and customer-facing productivity.
- Braze is still defining engineering-specific metrics, but cost, velocity, and output quality are top priorities.
A few notable milestones
- February 2025: Claude Code becomes a turning point for Hyman personally.
- Spring 2025: Braze rolls out tools and training across the org.
- August 2025: An AI-built MCP server is delivered about six weeks ahead of schedule.
- November 2025: Opus 4.5 becomes a major leap in model usefulness.
- Recent result: More than 60% of code committed to main repositories was written with AI assistance.
Leadership, Structure, and Team Management
Managing leaders vs. managing ICs
- As Braze scaled, Hyman had to shift from hands-on coordination to leading through directors and VPs.
- Division leads are expected to:
- Set goals
- Review architecture and design
- Articulate product vision
- Balance product health, feature adoption, and monetization
When leaders still need to get in the weeds
- Hyman believes executives still need enough technical depth to:
- Spot flawed estimates
- Break down oversized problems
- Prototype solutions when needed
- That hands-on involvement helps unblock teams and strengthen trust.
Acquisition and Integration: OfferFit
Cultural fit mattered
- Braze acquired OfferFit, a reinforcement learning company for marketing software.
- Hyman says the two companies had compatible cultures: open, direct, innovative, and leadership styles that aligned well.
Integration challenges and what Braze learned
- OfferFit was fully remote, while Braze is more hybrid and office-oriented.
- Braze borrowed some useful practices from OfferFit, including tooling and pull request workflows.
- OfferFit now operates as its own division within Braze’s broader divisional engineering structure.
Why engineering integration is easier to measure
- Hyman notes engineering is one of the clearest areas to assess during integration because velocity, workflow adoption, and product progress are visible and measurable.
Measuring AI Success and the Cost Problem
Braze is still defining the right metrics
Hyman says the company is early in figuring out how to measure AI value in engineering.
What they’re watching
- Adoption
- Velocity
- Output quality
- Inference cost
- Long-term maintainability
The cost surprise
- AI is productive, but inference is expensive.
- Hyman gives a concrete example of an engineer already spending $150 in a single day on inference.
- His concern isn’t just spend—it’s whether the output justifies the cost at scale.
Broader strategic insight
- Hyman argues that AI increases competition everywhere, not just within one company.
- Since competitors also get faster, teams have to keep moving quickly just to stay even.
- He warns against the simplistic idea that AI will make “vibe coding” enough to replace serious engineering or software buying decisions.
Product, Roadmaps, and UX Work
AI is changing how product teams work
Braze’s product process was already mature:
- Heavy customer feedback
- UX research
- Product management review
- Early access before general availability
AI now makes that process faster and more flexible.
Concrete product improvements
- Designers can now self-serve more of their work, including fixing UX debt
- PMs use tools like Vercel, Cursor, and V0 to build interactive prototypes quickly
- Teams can start building while design is still being refined, especially for beta customers who prioritize feature access over perfect polish
Why this matters
- Rework is cheaper than it used to be.
- Faster iteration means Braze can validate ideas sooner and move more quickly with customers.
Main Advice from John Hyman
For engineering leaders
- Stay close to the tools and workflows.
- Learn AI hands-on; don’t delegate understanding of it entirely.
- Raise expectations as model quality improves.
- Standardize AI workflows over time, but allow experimentation first.
For individual contributors and teams
- Tinker with AI tools directly.
- Put an AI app on your phone and use it regularly.
- Treat AI as something to build habits around, not just test occasionally.
Looking ahead
- Hyman expects 2026 to bring more standardization around:
- AI coding practices
- Shared workflows
- Agent infrastructure
- Team roles centered on agentic systems
Closing Thought
Hyman’s core message is that AI is not just a productivity boost—it is reshaping how engineering organizations are structured, how products are built, and how leaders should think about scale. His view is pragmatic: embrace AI early, measure it carefully, keep human domain expertise central, and use the gains to build more value rather than simply do less work.