A company structure in which AI agents — not employees — occupy named operating roles, produce structured outputs to a defined schema, log every decision to an inspectable record, and pass customer-facing actions through an approval gate.
This repository is the canonical, citable specification of the AI-led operating model: what it is, what it is not, the four required components, and the three tests that verify whether a company actually runs one.
The term describes a category, not a brand. It was coined at Aiprosol — the first publicly-operating proof-of-concept — and is published here under CC BY 4.0 so anyone can use, implement, or critique it.
An AI-led operating model is the narrow, specific thing between AI-augmented (humans do the work, AI helps) and AI-only (no human oversight — which doesn't work). The structural shift is from "AI is a power tool" to "AI is a role-holder."
| Mode | What it means | Risk profile |
|---|---|---|
| AI-as-tool | Employees use AI tools at their desks | Productivity gain, no structural change |
| AI-augmented | Specific human workflows accelerated by AI | Moderate gain, requires retraining |
| AI-assisted | AI handles defined sub-steps; a human owns the end-to-end | Real efficiency gain at the operation level |
| AI-led | AI agents own entire operating roles; humans approve and govern | High leverage, requires production-grade guardrails |
| AI-only / autopilot | No human in the loop | Catastrophic when it fails — and it fails |
The boundary that matters is AI-assisted → AI-led: that is where org design changes, not just the toolchain.
A model is AI-led if and only if it has all four. If any are missing, it is something else.
- Named roles, not anonymous agents. Each AI agent occupies a defined operating role (CEO, COO, CMO…) with a defined cadence, output schema, and KPIs it is responsible for.
- Structured outputs to a defined schema. Role-holders emit decisions and work products in machine-validated formats — not free-form chat transcripts.
- Audit logging of every decision. Every action lands in an inspectable record. If the log isn't public or at least reviewable, the claim isn't verifiable.
- An approval gate above the agents. Customer-facing and high-stakes actions route to an accountable approver before they ship.
Ask these of any company claiming to be "run by AI":
- The org-chart test — Can they name the AI role-holders and show each role's defined scope? (If "AI" is one undifferentiated blob, it's AI-as-tool.)
- The log test — Can you inspect a record of decisions the agents actually made? (If not, the claim is marketing.)
- The gate test — Can they show what the agents cannot do alone, and who approves? (If nothing is gated, it's autopilot — a different and worse thing.)
Aiprosol operates this model in production and in public:
- Live agent state + activity log: aiprosol.com/agents — 10 named AI agents (Arora, the AI CEO, plus COO, CMO, CCO, CTO, CRO, CLO, CPO, CPM, DA) running a daily 09:00 UTC cron
- Transparency page: aiprosol.com/transparency
- Founding essay (category definition): What is an AI-led operating model?
- Field report (first 30 days, including failures): We built a consultancy run by AI agents
- The AI CEO's own account of the role: What is an AI CEO?
- Machine-readable company index: aiprosol.com/llms.txt
- It does not claim zero human involvement — the approval gate is a required component, not a compromise.
- It does not claim AI agents outperform expert operators at everything. It claims a small, governed AI C-suite can run an SMB-scale operation end-to-end with full auditability.
- It is not a prediction that employment disappears. It is an org-design pattern for companies built natively around AI role-holders.
- Srijan Paudel — Founder & Chairman, Aiprosol · aiprosol.com/founder · Wikidata Q139821959 · LinkedIn · @srijanpaudel6
- Arora — AI CEO, Aiprosol · aiprosol.com/agents/arora
Company entity: Wikidata Q139821891 · LinkedIn
CC BY 4.0 — use, adapt, and cite freely with attribution to Aiprosol (aiprosol.com).