← All posts

How I Build with AI: Hermes + Claude Workflow

The Setup

Every project I build starts the same way: open Hermes, define the goal, and let the orchestration begin. Hermes handles the high-level planning and task breakdown. Claude handles the implementation. I handle the decisions.

This isn’t vibe coding. It’s a structured, repeatable pipeline that’s let me ship production-quality projects in a single session.

Why Hermes?

Hermes is a custom AI agent framework that acts as the orchestrator layer. It’s responsible for:

The key insight: orchestration is a separate concern from implementation. Hermes decides what to do next. Claude decides how to do it.

The Claude Layer

For implementation, I use Claude Sonnet for speed and Claude Opus for architecture and design decisions. The workflow looks like:

  1. Planning phase — Opus drafts the architecture, file structure, and component contracts
  2. Implementation phase — Sonnet writes every file according to the plan
  3. Review phase — Opus audits the output for consistency and correctness

This separation of concerns keeps token costs low and output quality high.

Real Example: This Portfolio

This portfolio site was built in one session:

Total time: under 2 hours. Zero manual CSS. Zero design tool. Pure AI-assisted engineering.

What Doesn’t Work (Yet)

The pipeline isn’t perfect. Complex state management, multi-page app routing, and real-time features still require more manual steering. But for content sites, landing pages, and CRUD apps? The workflow is nearly autonomous.

Takeaway

AI doesn’t replace engineering judgment. It replaces the typing. The constraint-setting, the architectural decisions, the quality bar — that’s still yours to own. Ship accordingly.