Specialization
Welcome to the afternoon session of the Agentic Software Development workshop.
During the morning core, we worked together to build the foundation of our HyperBytes application, establishing basic guardrails, lint check loops, and Spec-Driven Development (SDD) compliance.
In the afternoon session, we transition from structured training to an exploration mission. The developer group will divide into specialized roles, each assigned a specific capability boundary to research, configure, and conquer.
Rather than following a rigid, step-by-step instruction guide, your assignment is to explore and figure it out in collaboration with your agent and the internet. We provide the mission objectives, setup pointers, and the friction points you are expected to encounter. Your goal is to configure the proper guardrails, rules, and tool integrations (MCP servers, scripts, or hooks) that empower the agent to achieve the goals.
The Mission
For each lab assignment, we outline the following key telemetry points:
- The Mission Role & Focus: Your assignment and the technical area you are specializing in.
- The Target Feature: The user-facing improvement or addition to build in
HyperBytes. - The Agent Friction: The specific boundary or limitation the agent will run into by default (e.g. failing to maintain visual fidelity, violating architecture guidelines, or bloating the token budget).
- Explorer Setup & Pointers: The starting configuration commands, files, or parameters you need to initialize to begin the lab.
- Mission Objectives: What you and your agent hope to accomplish and discuss during our sharing session.
Show-and-Tell
At the end of the afternoon labs, we will reconvene for a Show-and-Tell session. Be prepared to share:
- The custom rules or configurations you established.
- The friction points you encountered and how you guided the agent around them.
- A quick demonstration of the final capability in action.
- Key takeaways on developer-agent collaboration in your specialized area.
The Specializations
Choose one of the specializations below to explore. You can choose to collaborate with another developer on the same specialization, or work alone.
The Surveyor
Specializes in design-to-code alignment by giving the agent visual spatial awareness, ensuring UI layouts match Figma mockup specifications perfectly without manual pixel-pushing.
The Guardian
Specializes in TDD (Test-Driven Development) loops, allowing the agent to self-correct its own business logic in a closed loop until all unit tests pass.
The Constructor
Specializes in custom Model Context Protocol (MCP) servers, writing Node.js tools that expose private APIs, component libraries, or compiled internal structures directly to the agent.
The Architect
Specializes in architectural governance and component boundaries, establishing ADRs (Architectural Decision Records) and strict TypeScript models that keep the agent from creating monolithic code.
The Orchestrator
Specializes in non-blocking parallel workflows, spawning headless agents inside isolated Git worktrees so features can be built asynchronously in the background.
The Inspector
Specializes in visual QA and layout auditing, capturing actual UI screenshots to give the agent multimodal vision feedback to resolve alignment, padding, and CSS visual bugs.
The Governor
Specializes in security compliance, commit hooks, and size budgets, preventing the agent from installing vulnerable packages, bloating the bundle size, or pushing messy Git logs.
The Navigator
Specializes in browser-native E2E (End-to-End) integration testing, using browser automation to verify complex user flows, animations, and transitions.
The Optimizer
Specializes in context pruning and output speed, minimizing the agent's input/output token overhead to bypass rate limits, reduce costs, and accelerate generation times.