AI operations

AI operations, verified.

MUNERON documents an AI operating layer built inside an aviation parts, MRO and landing gear leasing group. The layer reads the business systems, prepares and routes work, checks every figure against its source, and sends outward only through controlled channels.

What it runs, how it is controlled, and what the audit found.

Explore the system Contact MUNERON

Only using AI as a chat window? What a layer is made of, the six things to have in place before day one, and the order to build in.

Start here

In numbers.

From the job registry and the audit log, September 2026.

219
Scheduled jobs. Running on four machines.
311
Audit evaluations. One adversarial audit of the whole layer, 2 September.
102
Findings. Jobs that reported success while the output was wrong or reached nobody.
19 / 20
Re-checked by hand. Twenty findings re-verified against the original logs. Nineteen held, one did not.

Live demo

Green is not the same as right.

Eight jobs, all finished without error. A status page stops there. The layer does not: it opens the artefact each job produced and checks it against the source. Press the button.

Illustrative jobs. The three failures are real findings from the 2 September audit, with names removed. Runs in your browser; nothing is sent.

Functions.

Nine functions inside the group. Process owners keep approval. The layer reads source systems, prepares drafts, runs checks and sends follow-ups.

Enquiries and quotes

Read incoming requests, price them against stock and history, and draft quotes. Approved offers can send automatically under set rules, with manual-only control for any customer.

Purchasing

Track purchase requirements, request supplier offers, and record each offer in the system.

Finance

Read cash, receivable and payable balances from their source systems, with bank data taking precedence over book records.

Market

Read market and regional sources each morning and report only the items that changed, with the source next to each.

Knowledge

Store mail, chat, meeting and deal records in separate access-controlled logs, so restricted material is excluded from shared summaries.

Controls

Verify recipients, apply the lowest permission in any group, screen payment language, keep the ERP read-only by default, and audit the layer's own output.

All nine functions ↗

Controls.

Verification

Each displayed figure is pulled from its source system when the report is written, and shows that source.

Routing

Each recommendation is addressed to a recorded process owner. Actions are recorded as proposal, approval, execution and confirmation.

Outbound

Four outbound channels, each authorised, gated, logged and revocable. Each has its own stop, and one master stop covers all four.

Fixed rules

The layer does not sign documents and does not move money. Stop events report to the owner through a separate, ungated path.

Audit correction.

One of the 20 manually reviewed findings was wrong.

The audit reported that a daily cash alert never arrived, after reading two lines in the log marked blocked. The next line, 308 milliseconds later in the same file, recorded the alert as sent. Across the full log, 108 of 132 alerts went out.

The checker had treated an intermediate step as the final result. The finding was corrected, and the checker now reads the final delivery state before reporting on any send.

Built inside the business.

The layer was built by the founder of a group operating across the US and Europe, through written instructions and review, over four months, while the business ran on it. MUNERON is the company that documents the layer and the method behind it. How it is wired ↗

Latest.

Contact MUNERON

For companies already running AI in the business, with jobs nobody has independently checked.

Get in touch