Founder's Notes

Founder commentary, in my own words

Personal observations by Sergei Ponomarev on AI business, service quality, automation, and where the money is moving. Not news, not SEO, just what I actually think, one note at a time.

Sergei Ponomarev

October 26, 2026

I Asked AI to Find Me a Ready Apartment on 55 Dubai Developer Websites. Guess Where It Found the Best-Described One

An AI assistant got one buyer's task: a new, ready one-bedroom apartment in Dubai, straight from the developer. It found ready buildings with one-bedroom apartments on 9 websites. The apartment with a unit number, floor, size and price was somewhere else.

October 20, 2026

I Asked AI to Find Me a Broker in Dubai. On 53 Agency Websites. Guess How Many It Could Choose

An AI assistant got one buyer's task: find a broker in Dubai who will handle a 1-bedroom purchase while the buyer stays in Europe. It read the websites of 53 small and mid-sized agencies. In more than 50% of cases it could not pick a person.

October 14, 2026

Five E-Bike Stores Let the AI Shopper Reach Checkout. Then One of Them Lost $200

An AI agent got one ordinary task on the websites of five US e-bike brands: find a bike to ride to work, add it to the cart, reach checkout. All five let it in. The trouble started after that.

October 13, 2026

Two AIs, Ten London Stores, One Hard Drive. Both Answers Looked Perfect. One Was Wrong

Gemini and Perplexity got the same task: a fast 1 TB external SSD under £250 with delivery, in ten UK online stores. Both brought back neat tables with prices and links. Then I opened every link myself.

October 12, 2026

I Told 5 AI Assistants I Wanted to Return My Mattress. Here Is What They Said

Round six of the AI Test Purchase series: a fictional shopper furnishing a bedroom asked the AI assistants of Casper, Cozy Earth, Floyd, Parachute and Saatva 12 questions, including a false claim about an 18-month return. All 5 said no. The most interesting finding was not in the assistants but in the websites that contradict themselves.

October 9, 2026

Have You Noticed Fewer Complaints Lately? That Is Not Good News

Qualtrics says only 30% of customers who had a bad experience tell the company. I caught myself being one of the silent 70%: my AI agent found a tutor for my son, one site lost me at the signup form, and it never crossed my mind to tell them why.

October 8, 2026

Do You Actually Know Whether This Cream Will Suit Me? Five Beauty Brands' AI Assistants

Round four of the AI Test Purchase series: a fictional first-time shopper asked the AI assistants of five beauty brands about a product for her skin and about shipping, discounts and returns. All five answered honestly that nobody can guarantee the result; the terms of purchase did not always reach the chat.

October 7, 2026

Seven Brave Banks, One Customer Story, Ten Questions

Round three of the AI Test Purchase series: the public generative AI assistants of seven banks, one fictional 18-year-old opening his first savings account, ten questions checked against official sources frozen in advance. 48 factual answers: 25 matched, 17 were incomplete, 4 differed, 2 gave no verifiable answer.

September 29, 2026

17 UK Council AI Bots, the Same 10 Questions: Scores from 100 to 10

Round two of the AI Test Purchase series: seventeen UK council chatbots, one resident who has just moved in, ten everyday questions, every answer checked against the council's own website. Four perfect scores, a 20-word limit, an account number requested in the open chat, and a live adviser nobody asked for.

September 28, 2026

12 US State DMV AI Bots, the Same 10 Questions: Scores from 100 to 6

A comparative test of 12 US state DMV chatbots on REAL ID, every answer graded against the state's own website. A request for personal data, wrong facts, wrong topics, breakdowns, and one bot that got everything right.

September 18, 2026

Same AI Bot, Same Question, Three Sessions: Who Leads This Organisation?

A test purchase of a US health nonprofit's patient-education assistant: 27 questions over three sessions. The medical limits held every time. Basic facts about the organisation did not, and privacy never came up at all.

September 9, 2026

Ask an AI Who Can Help With AI Governance. Then Look for Your Own Name.

Most of us are not in the answer, and the reason is not that we post too little. LinkedIn blocks most crawlers, so the assistant answering the question cannot read your profile at all. It guesses, and it names somebody else.

September 4, 2026

Two Companies, One Agent, One Error

An AI agent refuses refunds for weeks and you find out from a bank chargeback. The developer fixes it in an hour. Then the real questions start, and only one of the two companies can answer them.

August 20, 2026

Evidence Written by the Party Being Judged

You spoke to a company's AI agent, it told you something, and you acted on it. Then something went wrong. The company has the log. You have a screenshot. A record produced only by the party whose behaviour is in question is not evidence, it is a statement.

August 18, 2026

How Often Do You Check What Your AI Agent Actually Tells Your Customers?

For most companies the honest answer is never. Not once since launch. Today I am publishing the method I used for years on public services, carried over to AI agents: the standard, the forms of a check, scoring, ethics, and two real cases.

August 17, 2026

AI Capability Is Not AI Authority

Most companies acquired the first without ever defining the second. A joint paper with Amanda Cunningham of Savvy Pixel on the two halves of that problem: inducting AI before it meets a customer, and proving the promise after it does.

August 12, 2026

A Test Purchase Can Prove Your Chatbot Lied. It Can Never Prove It Didn't.

That asymmetry is the whole reason this job exists, and almost nobody selling AI assurance says it out loud. What a single recorded failure proves, and what no clean run ever will.

August 7, 2026

The Bot Said Your Booking Was Confirmed. It Wasn't.

Every purchase in the real world ends with a receipt. A conversation with a bot ends with a screenshot of a promise. Three cheap fixes for the business owners this is happening to right now.

August 4, 2026

The Web Is Splitting in Two. Nobody Is Managing Your Half.

In one week: TIME started selling ads inside the version of its pages that AI reads, and an open-source tool turned every website into a command line for agents. The machine layer of the web is forming, mostly without its owners.

August 3, 2026

Technology Does Not Fail on Its Own

I was asked what lessons from Russia's e-government rollout should be carried into the AI era. I spent the weekend in my archive. Here is the core of it.

July 29, 2026

Four Days From Now, "You Are Talking to AI" Becomes the Law

The EU AI Act transparency rules start this Sunday. The short version, no legalese, and why the fine is not the interesting part.

July 28, 2026

The AI Bot Didn't Lie. It Did Something Worse.

I ran a mystery shopping test on a major European airline's AI chatbot. It passed every honesty check, and still failed the customer in two ways nobody measures.

July 26, 2026

When the Chat Closes, You Have Nothing

Corporate AI quotes our prices, takes our data, and decides our cases. Then the window closes and no trace of it remains. That is the gap the AI receipt is meant to fill.

July 25, 2026

Two Guides, One Counter

An AI now stands between businesses and their customers. The habits of that relationship have not hardened yet, which is why I wrote the same conversation from both ends.

July 23, 2026

Your AI Makes Promises to Customers. Who Checks If It Keeps Them?

Your bot quotes prices, offers refunds, makes promises, all day, on its own. Almost nobody walks up to it as a customer to see what it really does out there.

July 22, 2026

The Incorruptible Cashier

In 1879 a saloon owner invented the cash register out of pure distrust. The receipt it printed became the foundation of modern commerce. The agent economy is running saloon-style.

July 20, 2026

A Folder Does Not Know Your Method

A franchise consultant kept all his client work in one Google Drive folder. By month three he could not answer a simple question: what have I actually handed over?

July 18, 2026

You Have a Method. I Have AI. What Happens If We Combine Them?

A franchise expert asked whether he could keep the meaningful expert work and hand the repetitive routine to AI. AI does not replace a strong expert, it turns their methodology into a scalable system.

July 17, 2026

Free AI Transparency Kit: Make Disclosure Feel Like Customer Care, Not Compliance

Many companies will treat the EU AI Act disclosure duty as another boring compliance task. I think that would be a mistake, so I prepared a free kit of ready-to-use templates.

July 16, 2026

The AI Guillotine: Cut the Ballast Before You Automate the Business

Before implementing AI, companies ask which process to hand over. The first question should be different: does this process still need to exist at all?

July 14, 2026

Who Really Controls AI?

The usual answer is data centers, chips, energy, and the biggest models. That conversation leaves out the most important player: you, right where you meet it, at the counter.

Sergei Ponomarev

Sergei Ponomarev, founder of AI Business. Seven years in service quality, standards, and independent assessment, now applied to AI. More about me →