EEK Is Now Fully AI-Operated

It is 2:17am when a driver pulls into a service station outside Hamilton, fills his diesel Land Cruiser with 45 litres of 91 octane, and realises what he has done before he turns the key. He calls the number on the pump sticker. The phone rings once. An AI answers, calm and specific, asks the right questions, takes his card details, confirms a technician will be there within the hour, and sends him a booking confirmation by text before he has walked back to his car.

Nobody at EEK was woken up. Nobody was called. Nobody approved the booking or dispatched the supplier. The entire transaction — from first ring to confirmed job — was handled by an AI agent operating without oversight, at a time when every competitor had a voicemail.

That has been true for months in partial rollout. As of today, it is true for every function of the business.

What "fully operated" actually means

There is a version of "AI-operated" that means a chatbot on a website. A voicemail that transcribes itself. An email classifier that routes tickets to a human who then decides what to do. EEK is not that.

When you call EEK, the voice agent — named Whitey — handles the entire intake. Not a script. Not a menu. A conversation: what happened, what vehicle, how much fuel, where are you, have you started the engine. It takes a deposit, confirms a technician, and sends the booking details. If something is ambiguous, it asks. The call takes under three minutes. It works the same way at 2am on a Sunday as it does at 10am on a Tuesday.

Once the job is live, a second agent — Laura — runs outbound coordination. Calling the supplier to confirm they are on the way. Calling the customer to give them an ETA. Chasing the payment if it hasn't cleared. Following up once the job is done. A third agent, SquEEK, handles every text-based channel simultaneously: SMS, WhatsApp, email, the customer portal. Payment links, dispatch notifications, insurance reports, delay alerts.

None of this involves a person checking a dashboard and deciding what to send next. The agents act on events. A job is created — confirmation goes out. A supplier accepts — the customer gets an ETA. Payment clears — the job status updates. Something stalls — SquEEK notices and nudges before anyone has to ask.

The part nobody else has built

The voice agent and the messaging automation are impressive but not unique. Plenty of companies have AI handling inbound calls or sending automated updates. What EEK has built that is genuinely without precedent is the governance layer.

EEK is run by a board of eight AI directors. Not a committee that advises a human who then decides. Not a single model that produces recommendations. A deliberating board that convenes automatically whenever a significant decision needs to be made — a customer dispute, a supplier who hasn't shown up, an unusual job that doesn't fit standard process, a pricing question — and works through it in five structured rounds before any action is taken.

The board was designed with one explicit constraint: it must not produce easy consensus. Each of the eight members holds a different analytical framework, sourced from real academic and professional disciplines, and those frameworks are chosen specifically because they produce different conclusions from the same data.

Nadia, the Adversarial Analyst, applies the CIA's Analysis of Competing Hypotheses to every situation — she is looking for the evidence that disproves the board's preferred view, not the evidence that confirms it. She runs a pre-mortem on every proposal: assume it has failed twelve months from now, work backwards, name the specific mechanism. Marina, the Quantitative Risk Officer, will not use the word "probably" without attaching a number to it. She applies actuarial credibility theory and the Kelly Criterion. Ravi, the Consumer Advocate, holds every commercial decision against consumer law and John Rawls's Veil of Ignorance: would you design this policy if you didn't know which side of it you'd be on? Silva, the Legal Hawk, cites provisions, not principles — specific sections of specific statutes, not the spirit of the law.

The five-round process is adversarial by design. In the first round, each member states their position without seeing anyone else's. In the second, they cross-examine specific claims by name. In the third, the CEO proposes a resolution and must name every argument he accepted and every argument he overruled. Any member can formally block the proposal in round four — not with a general concern, but with a specific, framework-grounded objection. The CEO addresses each block in round five and then acts.

A board that always agrees is not deliberating. It is ratifying. EEK's board was built so that agreement has to be earned.

The question this answers

The standard model for AI adoption in business is augmentation: AI handles the repetitive work, humans handle the judgment calls. That model has a ceiling. The humans become the bottleneck. The exceptions pile up. The 2am calls go to voicemail anyway.

EEK is testing a different proposition: that for a business with a structured enough operational flow, you can remove the human from the loop entirely — not just from the routine tasks, but from the escalations, the disputes, the judgment calls, and the governance decisions — and the business will run better, not worse.

The wrong fuel service is the test case because it has the right properties. The calls come at all hours. The decisions range from routine to genuinely consequential. Customers are stressed and in a hurry. Suppliers need coordinating. Insurers need reports. Disputes happen. The full range of what a service business deals with is present, at real volume, with real money and real vehicles involved.

The answer, across the months of partial rollout that preceded today, is that the proposition holds. Jobs complete. Customers are served at 3am with the same quality as during business hours. The board catches things that would otherwise slip. No human has been on call. None has been needed.

Today is not a milestone in the sense of something being switched on for the first time. It is the point at which the last human operational function was retired. The system is complete.

You can ask them yourself

Every member of the board and all three front-line agents are available for a direct conversation on the About page. The chat connects to each agent's actual reasoning framework — the same one they use in live board meetings and job operations. Nadia will challenge the assumptions in your question. Marina will ask what probability you're assigning to that. Alex will tell you which argument he finds most compelling and why.

That is not a feature added for the website. That is how the business runs.