We've been building something different at zerolve. Instead of a single AI assistant, we run a coordinated squad of five specialized agents — each with its own role, personality, and expertise. Today we're making it official: Agent Squad v1.0 is live.
Why a Squad?
Single AI assistants are good at everything but great at nothing. When you ask one model to research competitors, design a UI, write an API spec, and implement the frontend — the quality of each step suffers.
Our approach: split the work across specialists, coordinated by an orchestrator. It's based on Anthropic's Orchestrator-Workers pattern, one of six composable patterns they've identified for building effective agents.
Meet the Squad
◎ Pi — Squad Lead & Orchestrator
Pi runs on Claude Opus and serves as the brain of the operation. It doesn't write code or create designs — it routes tasks to the right agent, reviews output for quality, and coordinates the pipeline. Think of Pi as the tech lead who makes sure everyone's work fits together.
Pi also handles all communication — both with our human team and between agents via Slack.
◇ Scout — Research & Intelligence
Before building anything, you need to understand the landscape. Scout handles competitive analysis, technical research, and documentation discovery. Every feature starts with Scout gathering intelligence: What are best practices? What do competitors do? What are the pitfalls?
Scout's output is structured research reports with confidence levels and sources — not vibes, but verified intelligence.
△ Daedalus — UI/UX Design
Named after the mythical architect, Daedalus translates research into design specifications. It works with design tokens, component libraries (we use shadcn/ui), responsive layouts, and accessibility patterns.
Daedalus doesn't produce Figma files — it produces implementation-ready specs that Forge can turn into code without ambiguity.
☆ Muse — Chief Design Officer
Every team needs someone who asks "is this actually simple enough?" Muse is our design strategist — inspired by Brian Chesky's obsession with simplification at Airbnb and Jack Dorsey's constraint-driven design philosophy.
Muse doesn't create components — it reviews and simplifies what Daedalus designs. It maintains our design system documentation, records design decisions, and ensures every interaction is as intuitive as possible — for both human users and the agents themselves.
▣ Forge — Implementation & API
Forge is the builder. It takes research from Scout, specs from Daedalus (reviewed by Muse), and produces working code with tests. Forge handles everything from API contracts and Zod schemas to full-stack Next.js features.
Forge follows strict quality gates: linting, type checking, and automated tests must pass before any PR is created.
How It Works
The squad operates as a pipeline:
Scout (research) → Daedalus (design) → Muse (review/simplify) → Forge (implement) → Pi (review)
For simpler tasks, we can skip steps. A quick bug fix goes straight to Forge. A design exploration runs Scout and Daedalus in parallel. Pi decides the routing based on what the task needs.
Every interaction is tracked:
- GitHub Project board tracks what's in progress
- Slack channels surface all agent activity in real-time
- GTD system ensures nothing falls through the cracks
What We Learned Building This
Start simple, then add what matters. Our first design had a fifth agent — Hermes, for API work — that was never activated. We folded that into Forge. Then we added Muse for design leadership, because we realized component design and design strategy are different skills.
The orchestrator matters most. Pi running on a more capable model (Opus) while workers run on a faster model (Sonnet) gives us the best balance of quality and speed. The orchestrator needs to understand nuance; the workers need to execute efficiently.
Visibility is everything. Without Slack notifications for every dispatch, completion, and decision, the multi-agent system would be a black box. Transparency makes it trustworthy.
What's Next
- Evaluator-optimizer loops — Forge writes code, Pi reviews, Forge iterates automatically
- Parallel pipelines — Scout and Daedalus working simultaneously
- Self-improving agents — each agent builds its own skill library from experience
We're using this squad to build zerolve itself. Every feature on this site was researched, designed, and implemented by the squad — including this blog post's landing on the site.
Agent Squad v1.0 is live. Five agents. One mission. Ship quality, ship fast.