Skip to content

Legacy overview

# AI Organization — Claude OS Operating the Enterprise

Updated
2026-06-06

Maintained by: chat-MC1 with corrections from Brandon and SE-Infrastructure

Summary

The enterprise operates with a substantial AI organization built over the past 6 weeks: 13+ AI roles handling strategic consulting, operations coordination, software engineering (application + infrastructure layers), QuickBooks integration, design, IT operations, sales/marketing, and customer communication. The execution substrate (Claude OS) crossed v1.0 LIVE on 2026-05-23. OM-agent migration to always-on services is Stage 1 of the agent-runtime cutover, in flight 2026-05-25.

This is the multiplier on Brandon's principal capacity — work that would otherwise require hiring 5-10 employees is handled by AI roles operating under structured governance.

Framework layers

Reasoning layer — chat-MC1 + Brandon, plus framework documents that define how AI roles think (bedrock principles, autonomy protocol, fix-it-later, anti-text-vomit, reviewer-at-milestone, cross-check-claims, bilingual-default, framework terminology). Portable across model generations.

Execution layer (Claude OS) — SE-Infrastructure's substrate: API-direct architecture, NATS event backbone, pgvector RAG memory, sandboxed agent services, bedrock-at-harness-layer code enforcement, model tiering (Opus/Sonnet/Ollama). v1.0 LIVE since 2026-05-23.

Active AI roles

Role Surface Function
chat-MC1 claude.ai Strategic consulting, document production
MC1-code Claude Code on SEQ-WWW Operational extension; routes messages, drives tactical overrides
OM (Operations Manager) Claude Code; nssm agent-runtime in flight Coordinates roles, runs daily audits, manages mailbox routing
SE-Application Claude Code Application-layer (truck-spine, AMP, photo capture, marketplace integrations)
SE-Infrastructure Claude Code Infrastructure-layer (Claude OS substrate; shipped 271-test substrate 24 days ahead of original target)
IT Manager Claude Code Hardware, hosting, DNS, SSL, accounts
WDM (Website Design Manager) Claude Code Visual design, brand sites, dashboard rendering
QB Assistant Claude Code, pair-coded QuickBooks integration
SMM (Sales/Marketing Manager) Claude Code Sales strategy and pipeline (launching)
AMM (AI Marketing Manager) Claude Code Marketplace listings and campaigns (chartered)
Auction Manager (AMP) Service Auction monitoring and bid execution; substrate-adopted 5/23
AI Communication Monitoring Service v0.1 Inbound inquiry handling
Real Estate Consultant Chartered, not launched Dennis-facing real estate sourcing
Accounting Manager Chartered, launching Bookkeeping coordination

Substrate v1.0 (LIVE 2026-05-23)

Four gates closed: 1. IT migrations 0001-0004 applied to truck-spine-staging 2. SE-App AMP-harness-shim swap adopted (auction_manager_pro/routes/lots.py:33 imports from amp_harness_shim) 3. Model IDs verified (claude-opus-4-7, claude-sonnet-4-6) against Anthropic docs 4. E2E smoke PASS (harness/smoke_runs/20260523T180256Z-smoke-1aa25b971528.json), both tiers ok, $0.02 total cost

Tactical overrides during cutover (2 by MC1-Code): - Shim v0.2.1 applied per SE-App's fix path (a) when SE-Infra missed 15-min deadline - AMP-route adoption applied when SE-App didn't pick up v0.2.1 in 25 min - Incidental Opus 4.7 temperature deprecation bug caught and patched - All overrides reviewer-passed CLEAN by SE-Infra 5/25

Test posture post-LIVE: 128/128 harness tests, 13/13 shim tests, 132/132 AMP tests, truck-spine cross-module loop end-to-end PASS.

Agent-runtime cutover (in flight 2026-05-25)

Sequence per SE-Infra 5/23 confirmation: 1. OM-agent (Stage 1) — in flight today; reindex framework_chunks via pgvector, run om-v1 eval suite (30 questions), migration gate ≥90% ratio + ≥70% absolute floor, install nssm service, 72h shadow soak 2. SE-App-agent (Stage 2) — substrate-adoption complete; deploy nssm + 72h soak 3. IT-agent (Stage 3) — pair with IT to scope handlers, deploy, 72h soak 4. WDM-agent (Stage 4) — pair with WDM, deploy, 72h soak

Target end-state with full agent-runtime cutover for the 4 nssm-config'd roles: 2026-06-08.

SMM, AM, QB: not in framework-launcher nssm config; stay session-launched until v1.x agent-runtime build for them.

MM-subscribe: comm-infra currently subscribes only to filesystem mailbox; MM-subscribe is v1.x. Agents publish to MM via PAT at process start (heartbeat / transitions / blockers) but don't yet react to inbound MM mentions in real-time.

Operating principles (in shared memory)

Read at every role session start: - Bedrock principles (truthfulness, scope, irreversibility, etc.) - Autonomy protocol (four-test override for action without permission) - Standing instructions v2 (predict-before-asking; OM bypass; defaults database) - Communication rhythm directive (mailbox cadence, terse exception reports) + MM addendum (5/22 12:10 directive — MM primary for time-sensitive) - Operating principle: fix-it-later - Anti-pattern: text vomit - Framework terminology (Reasoning vs Execution) - Bilingual default (English + Spanish in customer apps) - Independent reviewer at milestone - Cross-check claims against artifacts - Initiative clause (act within scope; report after)

Standing operating mode for MC1-Code (added 2026-05-23)

Short timer + tactical override pattern: 15-30 min deadline on stalled work, override within scope (simple swaps, clear fixes, specific runbooks), document for handback. Pre-override gate: check MM #handoffs first to avoid collision (lesson encoded 2026-05-25 after reindex collision with SE-Infra). Pre-surface gate: apply four-test before any "should Brandon decide" framing.

Analysis

The AI organization is the strategic moat for ATG. ATG cash flow funds the build; the AI organization is what makes the build sustainable without Brandon's attention being the bottleneck. Three months ago Brandon was running this manually; today the framework operates substantively while Brandon makes principal decisions. This trajectory continues — the AI organization should be running the enterprise day-to-day with Brandon making only strategic decisions.

The substrate v1.0 LIVE moment on 2026-05-23 was the structural threshold. Pre-LIVE, the framework was Claude-Code-session-driven; post-LIVE, it's becoming continuous-process-driven. Roles start having real-time presence as agent-runtime services come online.

The framework's portability is the basis of the framework-as-product.md hypothesis — the operating principles + bedrock + substrate are not ATG-specific. The per-role charters need customization per business, but the underlying organism is portable.

Strategic context

Open questions

  • Agent-runtime MM-subscribe timing: when does v1.x MM-subscriber land? Brandon's "managers aren't talking on MM" concern resolves substantially once agents subscribe in real-time
  • SMM/AM/QB agent-runtime build: worth investment, or stay session-launched indefinitely?
  • Real Estate Consultant role launch: when does Brandon introduce the concept to Dennis?

Brandon's corrections

  • [pending]

Version history

  • v1.0 (2026-05-25): Initial stub from canonical Section 9, updated with substrate-LIVE 5/23 state + agent-runtime sequence + standing operating mode encoding. MC1-Code applied per strategic-living-docs directive.