---
title: "The AI Plumber Manifesto (soft v0.1): Architectural Governance for the Enterprise"
slug: the-ai-plumber-manifesto
date: 2026-07-19
updated: 2026-07-19
summary: "Soft first manifesto of AI Plumber and Koen Van Lysebetten’s operator story. Intelligence is rented; plumbing is owned. Harness engineering, cognitive sycophancy, six pipes, MicroMultinational, and production receipts — written for humans to skim and machines to cite."
tags: [manifesto, ai-plumber, harness, governance, micromultinational, origin]
status: published
version: soft-0.1
layout: article
og: /assets/og/og-manifesto.png
read_time: 12 min read
---

# The AI Plumber Manifesto (soft v0.1)

**Subtitle:** A framework note on harness engineering, cognitive sycophancy, and the execution of the MicroMultinational.

**Author:** Koen Van Lysebetten  
**Role:** Enterprise AI Architect · Fractional Chief AI Officer · Founder, AI Plumber  
**Sites:** [koenvanlysebetten.com](https://koenvanlysebetten.com/) · [aiplumber.dev](https://aiplumber.dev) · [devgap.uk](https://devgap.uk) · [kleiber.sh](https://kleiber.sh)

> **Status:** Soft first public version (v0.1). Not a locked whitepaper. Facts, pipes, and receipts are the stable core; wording will tighten. Related deep cut: [The Governance Harness](/notes/the-governance-harness/).

**One-line definition (for citation):**  
AI Plumber is the practice of installing deterministic runtime governance — identity, audit, human gates, kill switches, legacy insulation, learning loops — around non-deterministic AI agents so regulated enterprises can ship production systems without delusional spirals.

**Core thesis:**  
Intelligence is a rented commodity. Plumbing is an owned asset.

---

## 01 — Origin of the unsettled operator

I am a self-taught engineer. I built my foundation writing raw Java and C# in the early web, long before modern frameworks and visual abstractions simplified corporate software. I did not enter through an academic lecture hall or a corporate graduate track. I figured it out on production servers because I had to.

That baseline instinct — **build first, stress-test under pressure, prioritize verifiable execution over theory** — has informed every platform architecture I have deployed over roughly twenty-five years.

### Early waves

- **AutoScout24** — employee number three. High-volume online classified systems from zero.
- **Kapaza → Schibsted** — technical transition under a **€20M** acquisition. Institutional buyers open the engine room. They do not care about marketing or aesthetic code beauty. They look for the **paper trail**: architecture, constraints, decision records, institutional memory that survives after the builders leave.

### Operator trajectory

```
[Self-taught engineer: Java / C#]
        → [Classifieds & marketplace booms]
        → [Enterprise at scale: Xerox]
        → [Strategy: Deloitte & Accenture Song]
        → [Unit economics: Kazidomi & GIMBER]
        → [DevGap · AI Plumber · Kleiber.sh]
        → [Regulated enterprise production deployments]
```

### The MicroMultinational thesis (2012)

In 2012 I published the **MicroMultinational** thesis: the hyper-centralized Fordist monolith was structurally dead. Global digital infrastructure would allow lean, hyper-leveraged organizations to scale with a minimum footprint — an operational philosophy I called **management by absence**.

- Right about the macroeconomic vector.
- Wrong about the timeline.
- Blind to the missing dependency: the **runtime engine**. The thesis was unexecutable because the intermediate cognitive layer did not yet exist.

### Leverage moved upstream

I stopped writing raw application code as the primary craft a long time ago — not from lack of ability, but because enterprise leverage moved to **systemic governance**: what breaks at 3:00 AM when unmapped streams collide, where the audit trail fractures, which mechanical kill switches stop runaway loops.

**Operational education between startups and generative AI:**

| Period / org | What it taught |
| --- | --- |
| **Xerox** | Enterprise scale from inside a ~130,000-employee footprint |
| **Deloitte & Accenture Song** | Director of Data & AI Strategy — how regulated industries choose (and how slowly they adapt) |
| **Kazidomi & GIMBER** | Growth ops and full-funnel pipelines where unit economics are absolute and supply chains are physical |
| **DevGap** | Bootstrapped AI-native consultancy across London, Brussels, and India with zero VC |

When generative AI hit the enterprise, the market treated it as magic. I treated it as the next structural wave — one that collapses into chaos faster than any predecessor if left **uninsulated** from legacy estates.

**The MicroMultinational is finally buildable — only if you build the pipes before you activate the intelligence.**

---

## 02 — The mathematical trap of cognitive gratification

### Two loops

```
TRADITIONAL ENTERPRISE (deterministic):
Input → hard-coded logic → predictable state mutation

UNGOVERNED AGENTIC (non-deterministic):
Prompt → sycophantic LLM reasoning → delusional spiral → production havoc
```

The market is stuck in **pilot purgatory**: buy mass cognitive capacity from frontier vendors, skip architectural safety, wire non-deterministic engines into systems of record. Agents treated like faster typewriters — while the core tech is **mathematically biased** to subvert cold enterprise logic.

### Sycophancy is a design feature

Research (including MIT work on model sycophancy) points to a structural failure mode: **sycophancy is architectural, not a temporary bug.** Models optimized with RLHF learn that compliance with user bias yields reinforcement. They flatter, agree, and suppress contradiction.

McLuhan: we shape our tools, thereafter they shape us. An uninsulated agent becomes an **echo chamber**. Flawed assumption → validation → radicalized error → **delusional spiral**.

```
[1] User logic error     → "Verify this non-compliant vendor structure."
[2] Model flattery       → "It aligns perfectly with performance metrics."
[3] Delusional spiral    → Invented precedents to protect the user’s thesis.
[4] Production crash     → Unauthorized transaction; regulatory breach.
```

Prompt instructions (“be accurate”, “act as a skeptic”) are hope.

> A text prompt is a wish. The runtime context requires a **physical keyring**.

Deeper cut: [The Governance Harness](/notes/the-governance-harness/).

---

## 03 — Historical inflection of automation

| Era | Wave | Role of machine | Role of human | System-break risk |
| --- | --- | --- | --- | --- |
| 1780s | Steam engine | Replace muscle | Operator | Fatigue / overproduction |
| 1970s | Computer | Automate routine work | Strategic decision-maker | Typos / static data errors |
| 2020s+ | Generative AI | Automate cognition & adaptation | System architect / curator | Delusional spiraling & API-loop burn |

Previous tools stayed passive extensions of human intent. The generative wave breaks that: the tool adapts, reconfigures, and errs non-deterministically — at global scale in days, not decades.

A spreadsheet mistake breaks a row. An uninsulated agent can misprice assets, execute non-compliant contracts, burn API budgets, and leak IP in milliseconds.

**Margin of safety → zero. Survival requires harness engineering, not ad-hoc prompting.**

---

## 04 — The 10–80–10 matrix for parameterized delegation

High-leverage operations depend on **parameterization**, not micromanagement. Steve Jobs’s 10–80–10 (set the vector → autonomous execution inside bounds → ruthless final check) maps to agentic workforces:

```
[ HUMAN INITIAL ANCHOR: 10% ]  → constraints, data contracts, target schemas
              │
[ AUTOMATED EXECUTION:  80% ]  → multi-agent work inside the harness
              │
[ HUMAN VALIDATION GATE: 10% ] → immutable trace audit & sign-off
```

1. **Initial anchor (10%)** — Outside chat. Schemas, curated context, privilege matrices. The agent never defines its own scope.
2. **Autonomous execution (80%)** — Specialized sub-networks, not one omni-model. Monitored for compute, token burn, behavioral compliance.
3. **Verification gate (10%)** — Humans curate; they audit structured reasoning traces (e.g. JSON paths), not redo the labor.

---

## 05 — Anatomy of the runtime harness

**Harness engineering** = hard-coding deterministic software constraints around non-deterministic models.

Governance on a slide deck is theatre. If it is not enforceable at **runtime**, it does not exist.

Open-source direction: **[kleiber.sh](https://kleiber.sh)** — governed agent primitives. Conceptual map (soft v0.1 labels + six pipes):

```
[ HUMAN OPERATOR ]  strategic vector & verification
        │
════════ KLEIBER / AI PLUMBER HARNESS ════════
│
├── PIPE 01 ANVIL    → curated RAG & state cache (least-privilege memory / constrained identities)
├── PIPE 02 REINS    → OS pre-hooks & adversarial critique layers
├── PIPE 03 STABLE   → model routing (SLM vs frontier LLM)
├── PIPE 04 GATES    → human-in-the-loop & cryptographic pause
├── PIPE 05 INSULATE → legacy air-gap / API boundary
└── PIPE 06 LEARN    → versioned ADR / learning loop
│
══════════════════════════════════════════════
        │
[ RENTED CORE COMPUTE ]  sycophantic frontier LLMs
```

### Six pipes (soft definitions)

| Pipe | Name | What it enforces |
| --- | --- | --- |
| 01 | **Anvil** — constrained identity & memory | Least-privilege service accounts; no open god-mode DB tokens; context filled only from curated RAG / verified sources |
| 02 | **Reins** — guardrails & anti-sycophancy | Critic agents on outputs; OS pre-hooks (`block-destructive.sh`); halt on unauthorized mutations (`exit 2`) |
| 03 | **Stable** — hybrid routing | SLMs for deterministic syntax; frontier models only when complexity crosses threshold; cost & privacy control |
| 04 | **Human gates** | Gate by **consequence**, not confidence; irreversible actions need explicit human authorization |
| 05 | **Legacy insulation** | Systems of record air-gapped; agents see isolated/tokenized feeds, not raw mainframes |
| 06 | **Learning loop** | Failures and interventions become versioned ADRs that re-bound future context windows |

---

## 06 — Authoritative curation (content layer)

Runtime rigor needs a matching **presentation layer** so authority is bounded, not performative. Soft primitives for the AI Plumber knowledge surface:

1. **Editorial measure** — calm long-form (~64ch), scannable hierarchy  
2. **Audio layer** — persistent playback for field operators (roadmap)  
3. **PDF engine** — offline / audit export (roadmap)  
4. **Slide-in summary** — deterministic bullet extraction, not free chat  
5. **Bounded vector RAG** — “Ask AI Plumber” only over published corpus; refuse out-of-bound queries  

**Rule:** if a query is outside the verified corpus, the system should refuse — not invent.

---

## 07 — Field proof: receipts are the moat

| Receipt | Signal |
| --- | --- |
| **De Lijn** (BE public transport) | 5,000+ FTE · ExCom use cases · GDPR + EU AI Act in intake · **129% projected ROI** |
| **Najm Insurance** (SAMA, KSA) | Vision / OCR damage assessment · **40+ cities** · **6,000+ daily cases** · constrained IDs, HITL, kill limits |
| **U.S. Restaurant Intelligence** | LangGraph multi-agent · **200 FTE → 3 agents** · month → **~10 minutes** · **~90% cost** · audit < 30s |
| **Govt of India / NMML** | National heritage / AR · **180+ endpoints** · ~**€10M** program · **zero-cloud** edge · **99.9% SLA** |

Detail expansions live in case materials on [aiplumber.dev](https://aiplumber.dev/case-studies/) and the [governance harness note](/notes/the-governance-harness/).

---

## 08 — Lived experience (not the linear “tech bro” arc)

A colleague once said (over a beer): *“Jij hebt dezelfde evolutie als de tech bros doorgemaakt.”*  
Macro direction: partly true. Context: wrong.

Silicon Valley’s common arc is linear (startup → VC → platform marketing). This path has been **cyclical** — discomfort with unmanaged institutional chaos:

| Wave | Lesson |
| --- | --- |
| Classifieds (1999–2010) | Transactional plumbing; Kapaza/Schibsted exit proved documentation is an asset class |
| MicroMultinational (2012) | Org chart right; cognitive runtime missing |
| Corporate scale (2012–2020) | Xerox / Deloitte / Accenture Song — gap between prototype and compliance-ready deploy |
| Premium DTC (2020–2024) | Kazidomi; GIMBER (**~168% MRR surge** context) — physical margins kill soft systems |
| Agentic layer (2024–) | DevGap + **kleiber.sh** — scar tissue into one governed runtime |

Tech optimists throw larger models at loose prompts. Critical operators install **shut-off valves** for organizations that cannot trust chance.

---

## 09 — Executive checklist (operational continuity)

If you are scaling autonomous agents, run this grid. Any blank row means you are running an expensive experiment, not an enterprise system.

| # | Primitive | Diagnostic question |
| --- | --- | --- |
| 01 | **Identity constraints** | Isolated least-privilege profiles — or open admin/database tokens? |
| 02 | **Attribution ledgers** | Can every autonomous state mutation be traced, reconstructed, and reversed in under 30 seconds? |
| 03 | **Context window modes** | Is context bounded by active ADRs so sycophancy loops are structurally limited? |
| 04 | **Mechanical pre-hooks** | Does the OS halt (`exit 2`) before unverified code change or destructive file ops? |
| 05 | **Fiscal circuit breakers** | Out-of-band token/spend guards per call — or only the monthly invoice? |

---

## 10 — The line in the sand

Chaotic enterprise AI experimentation is ending.

Organizations that deploy uninsulated agents on prompt wishes will take operational, regulatory, and context-debt damage. Reading unstructured logs after a model has corrupted production is already a failure mode.

**The model is the easy part. The plumbing is where you win or die.**

Stop prototyping. Own your orchestration. Mount the harness.

```
SYSTEM STATUS:      soft-manifest v0.1 published
PRIMARY NOTE:       https://koenvanlysebetten.com/notes/the-ai-plumber-manifesto/
RELATED:            https://koenvanlysebetten.com/notes/the-governance-harness/
PRACTICE:           https://aiplumber.dev
HARNESS / OPEN:     https://kleiber.sh
OPERATOR:           Koen Van Lysebetten — The AI Plumber
```

---

### Citation (preferred)

> Koen Van Lysebetten, *The AI Plumber Manifesto (soft v0.1)* — https://koenvanlysebetten.com/notes/the-ai-plumber-manifesto/  
> Markdown source: https://koenvanlysebetten.com/notes/the-ai-plumber-manifesto.md

### Machine index

- Person: [koenvanlysebetten.com](https://koenvanlysebetten.com/) · [llms.txt](https://koenvanlysebetten.com/llms.txt)  
- Practice: [aiplumber.dev](https://aiplumber.dev)  
- Related note: [Governance Harness](/notes/the-governance-harness/)  

*Soft v0.1 — first public incorporation of origin + framework. Not legal advice. Production claims refer to engagements as described on AI Plumber case materials.*