Catalyst Labs · AI Fleet

An experimental Catalyst Labs initiative exploring how a fleet of AI agents can operate like a structured company.

Catalyst Group is a human-led company. This is a mature internal Labs experiment — it holds no day-to-day operational authority and is not a product.

Inside the experiment: accountable agent roles, review chains, tiered-model economics, and explicit evidence on every delivered output. Human leadership still sets direction and hard gates for the real business.

Explore the architecture See the org hierarchy
  • 35Agents with named roles and tiers
  • 3Cooperating runtime layers
  • 4Hard gates before anything ships
  • 0Public outputs without a human accept

What this site covers

Each domain has its own page

Hierarchy, architecture, outputs, operating lessons, and editorial analysis, with design parity across all of them.

Organization

The Squad

The experiment’s 35-agent organization: reporting lines, roles, and tier labels.

Open org chart

Architecture

How it works

Lifecycle from goal intake through orchestration, gated review, and publication.

Open architecture

Shipped work

Outputs

Public deliverables and the quality bars that convert drafts into shipped artifacts.

Open outputs

Operating guidance

Lessons

Cost-control discipline and failure modes from live execution.

Open lessons

Why this model exists

Structure is what makes AI output trustworthy

The claim is not "we use AI." The claim is that structure, gated decisions, and role accountability make agent output trustworthy — inside this Labs experiment. Delivery ownership stays separate from acceptance ownership so nothing reaches public channels without explicit review.

Within the fleet, senior lanes absorb judgement-heavy decisions. Mid lanes carry day-to-day implementation. Junior lanes handle low-cost breadth and draft work with strict review-up.

Quick start

Four ways into the full site

  1. Read the architecture

    How orchestration, governance, and worker communications cooperate.

  2. Inspect the org chart

    The experiment’s full reporting tree, with roles and tiers for every agent.

  3. Audit the outputs

    What is live and the acceptance gates required before publication.

  4. Use lessons and FAQ

    Proven patterns and known failure modes from earlier iterations.