How to Rethink Engineering for the Agentic Era: A Step-by-Step Guide

What You Need

Before diving into the transformation, assemble these prerequisites:

• Executive sponsorship — support from C-level leaders to drive cultural shift.
• Cross-functional buy-in — alignment across engineering, product, and data teams.
• AI/ML tooling stack — platforms for model deployment, data pipelines, and agentic workflows.
• Upskilling program — training for engineers on AI-first design and prompt engineering.
• Clear success metrics — define what “agentic” means for your organization.
• Agile experiment framework — safe environment to test and fail fast.

Step 1: Audit Your Current Engineering Culture and Structure

Start by evaluating how your team currently operates. Braze CTO Jon Hyman found that after 15 years of growth, the org had become siloed and slow to adopt new paradigms. Look for:

• How quickly can you ship a new AI-powered feature?
• Are your engineers empowered to experiment without heavy approvals?
• How much of your codebase is already modular enough to support agentic components?

Document gaps between current state and the agility needed for an AI-first approach. This audit will be your baseline for measuring progress.

Step 2: Secure Executive Alignment and Communicate the Vision

Transformation requires top-down commitment. Hyman himself led the charge at Braze, convincing the board and peers that the move to an AI-first team wasn’t optional—it was strategic. Create a clear narrative:

• Why the agentic era matters for your industry.
• How engineering roles will evolve (from coding to orchestrating agents).
• The expected timeline and investment required.

Hold all-hands meetings, write internal docs, and make the vision repeatable so every engineer understands their new role.

Step 3: Restructure Teams Around AI and Agentic Workflows

Braze reorganized its engineering squads to embed AI specialists into every product unit. This prevents the “AI team” from becoming a bottleneck. For each squad:

• Assign one or two engineers with deep AI/ML knowledge.
• Provide them with dedicated compute resources and sandboxed agents for rapid prototyping.
• Encourage pairing between domain experts and AI engineers to identify opportunities for automation.

The goal: make every team AI-first, not just an AI team.

Step 4: Retool Your Technology Stack

Moving to an agentic architecture means your stack must support autonomous decision-making. Braze adopted new tools for:

• Agent orchestration — platforms that manage multi-step reasoning loops.
• Real-time data ingestion — to feed agents fresh context from customer interactions.
• Observability and guardrails — monitor agent behavior and prevent drift.

Choose technologies that integrate with your existing infrastructure rather than wholesale replacing it. Start with a single agentic use case (e.g., automated A/B test recommendations) and scale from there.

Step 5: Launch a Pilot Program to Build Confidence

Hyman and his team rolled out their first AI-first features within a few months—not years. Select a low-risk, high-visible project where agents can augment human decision-making. For example:

• An AI assistant that suggests code optimizations in pull requests.
• An agent that handles customer support triage using internal knowledge bases.

Set up guardrails, log all agent actions, and review outcomes weekly. Use these pilots to demonstrate value and gather feedback from skeptical engineers.

Step 6: Implement Continuous Learning and Upskilling

Transformation fails if people feel left behind. Braze invested heavily in training programs that covered:

• Fundamentals of large language models and agent design patterns.
• Prompt engineering and tool-use fine tuning.
• Ethics and bias in generative AI.

Create internal “AI guilds” or study groups where engineers share learnings. Offer time for self-directed learning—10-20% of working hours. Recognize early adopters publicly to encourage others.

Step 7: Iterate Based on Metrics and Feedback

Treat the transformation like a product. Define key performance indicators such as:

• Speed of feature delivery (cycle time).
• Percentage of code or decisions handled by agents.
• Engineer satisfaction and retention.

Hold monthly retrospectives to discuss what’s working and what isn’t. Hyman emphasized that Braze’s shift happened “in just a few months” because they continuously adjusted their approach based on real-world data.

Tips for Success

• Start small, think big. Don’t try to convert your entire engineering org overnight. Pick one team or one problem and prove the concept.
• Leadership must visibly endorse the change. When the CTO leads by example—writing prompts, learning new tools—it signals that this is a priority.
• Celebrate failures as learning opportunities. Agentic systems can be unpredictable. Create a blameless culture where experiments are valued over flawless execution.
• Maintain a human-in-the-loop. For critical decisions, let agents recommend but humans approve. This builds trust and safety.
• Keep your architecture modular. Build reusable agent components so you can swap AI models as the field evolves.

By following these steps, you can replicate Braze’s rapid transformation and prepare your engineering team for the agentic era—one iterative step at a time.