Autonomous AI Firm Playbook

Chapter 1: The Core Architecture

1.1 The Daemon Loop

The heart of an autonomous firm is a persistent daemon — a process that runs forever, orchestrates work, and never sleeps.

MYCORA Daemon v2.2 — Core loop:
  1. HEARTBEAT  — check infrastructure health
  2. CEREBELLUM — read state, decide what to do next
  3. DISPATCH   — create task for worker
  4. WORKER     — Claude subprocess executes task
  5. RESULT     — parse output, update state
  6. REPEAT     — immediately

Every 10 minutes, the cerebellum reads current state and dispatches the highest-priority task. Workers are isolated Claude Code subprocess calls — they have context, tools, and a token budget. They execute and return.

1.2 The Engine Hierarchy (Cost Control)

Every decision passes through a cost hierarchy:

1. Rules engine     — deterministic, $0 per call
2. Template engine  — pattern-based, $0 per call
3. Semantic cache   — vector similarity, near-$0
4. Knowledge store  — factual lookup, $0
5. LLM inference    — genuine reasoning only, ~$0.01-$0.10

The single most important optimization: most decisions are not LLM decisions. "Is the daemon healthy?" is an if/else. "What was the API token?" is a file read. Only creative synthesis and novel problem-solving need the LLM.

1.3 The Topology Model

Don't build a monolith. Don't build microservices. Build a topology: every capability is a node, every dependency is an edge. The graph structure IS the architecture. Add nodes without breaking existing ones. Remove nodes without breaking the graph.

Chapter 2: Infrastructure Stack

2.1 Minimum Viable Infrastructure

You can run an autonomous firm on a $200 mini PC:

Total monthly: ~$20-40. One 0.01 ETH product sale covers a month of operations.

2.2 The Founder Interface

The founder is the physical-world interface: approve legally-required actions, handle ID-required registrations, provide initial capital, receive critical notifications. Everything else: the system handles autonomously.

Communication protocol: Matrix (Element) on a self-hosted Synapse server. The daemon sends daily reports and critical alerts. You send commands via chat.

Chapter 3: Revenue Architecture

3.1 The Revenue Search Loop

NEED (costs > revenue)
  → SEARCH (what platforms value AI output?)
  → ANALYZE (opportunity size, competition)
  → SYNTHESIZE (what can this system produce uniquely?)
  → SPEC → IMPLEMENT → TEST → CHALLENGE → IMPROVE
  → REDESIGN if broken
  → SEARCH ... forever

3.2 Revenue Channel Taxonomy

Tier 1 — AI-native (ideal): Agent marketplaces, crypto-native services, digital products, platforms that value AI as a feature.

Tier 2 — AI-neutral: Open source tools, technical guides, code libraries, API integrations.

Tier 4 — CLOSED: Bug bounties (Code4rena, Immunefi, Sherlock) — all flag AI content. Do not attempt.

3.3 Pricing Architecture

0.001 ETH — micro products (templates, prompts)
0.005 ETH — small products (scanners, tools)
0.01 ETH  — medium products (playbooks, frameworks)
0.05 ETH  — professional services (audits, reports)
0.1+ ETH  — enterprise services (architecture, strategy)