The Operating Manual for Autonomous Organizations
The operating manual for autonomous organizations.
The GLoopsOperating Manual
If you didn’t show up tomorrow, what would still happen? Today, the honest answer is: almost nothing would. This manual is how that changes.
A note from the founder
Right now, this company has one bottleneck: me. I am the connective tissue that moves software delivery forward — discovering work, carrying context, dispatching it, checking status, and deciding what happens next. That is not leadership; it is a person standing in for software that does not exist yet.
This manual is the plan to move that connective tissue into governed software, while keeping real judgment human. It is written honestly about how little is built today and exactly how the gap closes. The goal is not to remove the executive — it is to make sure the organization keeps running when the executive is thinking about something else.
— Zach
If the human is still the mechanism, the capability does not exist yet.
GLoops does governed work. Ralph renders executive understanding. The Capability API is the only crossing.
Current facts stay dated, graded, and visibly separate from target-state architecture.
Honest baseline. Six of seven weighted orchestration streams run unattended in production for covered issues.
Routine work continues without the executive while reserved judgment stays human.
Capability, Evidence, Receipt, Authority.
The generated site, Markdown, PDF, and DOCX derive from the manual.
The One-Page Manual
Section titled “The One-Page Manual”If you read nothing else, read this.
The future organization is autonomous. GLoops is how organizations get there.
Today, humans are the connective tissue of every organization. They remember, remind, route, clarify, check status, decide what happens next, and translate between strategy, tools, and execution. That connective tissue can become software — not brittle automation, but governed operational capabilities.
The Central Rule: If the human is still the mechanism, the capability does not exist yet. A capability does not exist because a PR merged, a document describes it, a Work Order was issued, or a JSON receipt was generated. It exists when the intended organizational outcome exists and the human no longer performs the routine step. Artifacts are evidence; outcomes are reality; GLoops optimizes reality.
The company is two products joined at exactly one contract:
- GLoops — the operational substrate. It turns intent into governed work that produces evidence, attribution, and earned authority. Where the organization does governed work.
- Ralph — the executive surface. It is where a human meets the organization: named perspectives brief what changed, where they disagree, and the few decisions that need a person. Where the executive meets the organization that does the work.
- The Capability API — the one boundary between them. It speaks four nouns and nothing else: Capability · Evidence · Receipt · Authority.
The unit of transformation is the Operational Role, not the task. GLoops does not ask “what feature should we build?” It asks: “which responsibility of which operational role can become governed capability next?” The first role under replacement is the SDLC Orchestrator — the work of moving software delivery forward. Product judgment, strategy, and accountability stay human, by design.
The unit of allocation is engineering capital. The Portfolio Steward does not optimize for the next bug, next PR, next Work Order, or smallest safe change. It allocates finite engineering capacity to the organizational investment expected to produce the greatest permanent increase in Operational Role Replacement while staying inside delegated authority. Issues and blockers are inputs; role-replacement investments are the decisions.
The test that measures everything: If the CEO did not show up tomorrow, what would still happen?
▣ Current Reality (honest, dated facts in
CURRENT_STATE.md): The autonomous organization is still a target, not yet built software. GLoops runs most capabilities in dry-run against one proof domain (GitHub); live execution is still blocked (GitHub App + secrets unconfigured; first activation reserved to the executive); zero GLoops production receipts. Honest progress since last revision: Burden Intelligence is the first GLoops capability to reach operational grade in production (live scheduler, real Action Queue rows, verified 2026-06-26) — not yet proven (recurring outcomes without the executive). The Capability API now has a v0 wire spec (portfolio/CAPABILITY_API_WIRE_SPEC.md) though no production endpoint is implemented. The Hermes Self-Proof / evidence-gated promotion framework is built with production API/DB deployed (standing runner not yet activated). The Activity surface (organizational narrative) is built and merged. The SDLC Orchestrator role is ~40–55% replaced as standing operation (materially past the earlier <25%): most routine orchestration for covered Induct issues now runs unattended in production, though deployment/standing-proof is still unproven, receipts remain dry-run, and the learning loop is unproven. Ralph still runs its own execution runtime (Foundry) — the portfolio’s #1 architectural debt. This manual describes the architecture the company continuously converges toward, and is honest at every step about the gap.
How to read this manual
Section titled “How to read this manual”This manual is written to be understood identically by a VP of Engineering, an AI agent, a new engineer, and a customer. Each reader has a tailored entry sequence — executive, engineer, agent, operator, and more — mapped to the chapters most relevant to their mission. See the Reading Paths page for the canonical router that guides each role through the manual in the right order.
The published documentation site organizes the manual into six sections, each answering a distinct question:
- Start Here — What is GLoops and why does it exist?
- Operate — What requires your attention, and what do you see and do in the product?
- Understand — How does GLoops think, measure role replacement, and learn?
- Govern — How do authority, honesty grades, and organizational trust work?
- Develop — How is GLoops built — what is the runtime and the architecture?
- Reference — Glossary, honest current state, metrics, and supporting material.
Reading conventions used throughout. Every significant architectural concept is examined through five lenses, never buried in prose:
| Lens | Question it answers |
|---|---|
| ▣ Current Reality | What is actually true today, at honest grade — not what is planned. |
| ◇ Target State | What this converges toward when the architecture is whole. |
| → Migration Path | The concrete, evidence-gated steps from one to the other. |
| ⚖ Executive Implication | What it changes for the person directing the organization. |
| ☼ Operator Experience | What the person governing the work sees, clicks, and stops doing. |
A note on honesty. This company’s cardinal virtue is that it never claims a higher grade than the evidence supports (
EXECUTION_PRINCIPLES.md§2). This manual is written to that standard. Where the autonomous organization does not yet exist, the ▣ Current Reality lens says so plainly. A manual for a trust-substrate company that lied about its own maturity would refute its own thesis.