AI Built My First Product in an Afternoon. Why I Scrapped It

Can AI come up with a small online business, build it, run it and generate an income with very little input from me? And could someone with no experience of AI do the same?

That’s the experiment I’ve handed to Grok Bot. It even chose the name, HonestAIBusiness and I bought the domain it suggested.

Grok Bot is doing most of the work: researching business ideas, creating products and building the website. My role is to approve decisions, question what it produces and report honestly on what happens.

The full website isn’t live yet—Grok Bot is building it as I write—but you can already sign up for the newsletter here. Both the website and newsletter will run on beehiiv.

There’s already something to report. Grok Bot proposed three business ideas, recommended one, then researched, built and revised the first product—all before I decided to scrap it that same afternoon.

An encouraging start, depending on how you look at it.

What I’m trying to find out

I’ve spent years running websites and creating content, but this experiment has a different purpose. I want to see how much of a new business AI can actually handle.

That means more than asking it to write a few promotional posts. I want it to research opportunities, suggest what to sell, build the product, help create the website and handle as much of the work as possible while I’m getting on with other things—or sleeping.

The initial financial goal is modest: a couple of hundred euros a month would be a worthwhile starting point. Whether it can achieve that is something we’re going to test.

I set a few conditions:

I should own the product and the business.
It should be able to survive a change of platform or payment provider.
Customers should be able to buy without sales calls or individual chasing.
AI should do most of the work, with limited input from me.
The results should be reported honestly, including costs, mistakes and products that don’t sell.

HonestAIBusiness will be the home for that experiment. Its proposed tagline sums it up: “Small online businesses. Built with AI. Tested in public.”

Three ideas, one recommendation

Grok Bot initially proposed three possible business ideas. Of those, it recommended that we start with an English-language spreadsheet kit for self-employed people in Spain.

I’m an autónomo myself, so the problem was familiar. Spanish tax terminology can be confusing enough before you start dealing with quarterly returns, different types of clients and numbered boxes on forms.

The proposed product, Autónomo Tracker, would help people record income and expenses, understand the figures and estimate what to put aside for quarterly tax payments.

It looked like a reasonable fit: a downloadable product, a specific audience and a problem I understood personally.

We discussed selling it for around €24 at launch and €29 afterwards. Those were proposed prices, not evidence that anyone would buy it.

The idea came from Grok Bot. I agreed to explore it, and it got to work.

What Grok Bot built

The speed was impressive.

Within the afternoon, Grok Bot researched competing products, drew up a specification and produced a spreadsheet and a PDF guide. It also attempted to check tax rules, rates and deadlines against published sources, flagging points it hadn’t fully verified.

The first version had numerous tabs covering income, expenses, equipment, tax calculations, foreign clients, deadlines and terminology.

My reaction when I opened it was fairly straightforward: I wouldn’t want to use this.

I was supposed to be the target customer, and it looked like homework.

So I asked for a simpler version.

The revised spreadsheet had four visible tabs: Start here, Money in, Money out and Dashboard. The more complicated calculations were hidden in advanced tabs.

You entered your figures and selected a few options. The dashboard then produced a sentence for each quarter along the lines of:

“Set aside €X for IVA and €Y for income tax. Pay by [date].”

There were colour-coded inputs, dropdowns, protected calculation cells and a shorter guide.

It was much closer to something I could imagine an ordinary person using.

Unfortunately, looking straightforward and being dependable are two different achievements.

Then I asked ChatGPT to find the problems

Before taking it further, I uploaded Grok Bot’s spreadsheet and guide to ChatGPT and asked for a blunt critique.

It found specific problems.

One involved a reduction in quarterly income tax payments for people with low previous-year earnings. The guide told first-year users to enter zero, but the formula treated zero as a reason not to apply the reduction.

Another involved dates. The calculations grouped transactions by quarter without also checking the year. An invoice from the previous January could be included in the current year’s first-quarter figures.

There were other concerns, including assumptions about foreign clients, missing information and how deadlines were handled.

These were useful findings. They also exposed a weakness in the process: Grok Bot had described checks it had performed, yet errors remained.

ChatGPT’s review wasn’t professional validation either. It was another layer of scrutiny, and it showed why that scrutiny mattered.

The friendly dashboard became the problem

The part I’d liked most was the simple dashboard telling users what to set aside.

But think about how someone would read it.

They’ve bought a tool because they don’t fully understand the subject. They enter their information, and a clear, confident sentence gives them a number and a payment date.

Of course they’re likely to trust it. That’s much of the reason they bought it.

The product included prominent warnings that it wasn’t tax advice and that users should check the figures with their own adviser. But those warnings didn’t solve the underlying problem.

If I was selling something that estimated tax payments, I needed a sound basis for believing those estimates were reliable within its stated scope.

The simpler we made the answer look, the more important that became.

Why I scrapped it

The individual bugs could have been fixed. My concern was what owning this particular product would involve afterwards.

Tax rules and deadlines change. Customers have circumstances a spreadsheet may not cover. Buyers would understandably ask whether a particular expense or client belonged in a particular category.

Those questions would need careful answers, and I’m not a tax adviser.

A product like this could be developed with qualified input, proper testing and a clear maintenance plan. But that would be a different undertaking from the one I’d set out to test.

I wanted a small business where AI could handle most of the ongoing work with limited supervision. I’d ended up with a product that would require continuing specialist checks.

So I decided not to launch it.

Grok Bot had done a lot of work quickly. It had also recommended an idea that didn’t hold up well enough against my original conditions—and I’d initially agreed with it.

That’s part of the experiment too.

What the afternoon taught me

The most useful discovery was how quickly AI can turn an idea into something that feels ready to sell.

A product name, a spreadsheet, a guide and a proposed price create a powerful sense of progress. You can start thinking about the launch before you’ve established whether the thing deserves to be launched.

I also learned that asking an AI whether its work is correct isn’t enough. You need checks that challenge its assumptions and test what happens when users do something unexpected.

Most importantly, the cost of building a product is only part of the calculation. Maintaining it, supporting customers and checking updates may demand far more work than producing the first version.

That matters enormously when the aim is a business that needs very little of my time.

What has the experiment cost?

So far, the additional cash outlay for the HonestAIBusiness project has been registering HonestAIBusiness.com. The website itself hasn’t been set up yet.

I haven’t paid a developer or designer to build the tracker, and I haven’t spent money advertising it.

However, “AI built it for nothing” would be misleading. I already use paid AI services, and my time has a cost too. As the experiment develops, I’ll distinguish between existing subscriptions, additional project spending and the time I spend directing and checking the work.

There have been no sales from this product. It never reached that stage.

Scrapping it now means the cost was an afternoon of experimentation, rather than months spent maintaining something I shouldn’t have launched.

What’s next?

Grok Bot’s next task is to reassess the options against what we’ve learned.

I’ll be looking for products with fewer specialist rules, a wider potential audience and ongoing work that AI can realistically handle. Those are things to demonstrate, rather than simply accept because an AI says they’re possible.

The bigger question remains open: can it build and help run a small business that earns money without becoming another full-time job for me?

For now, the first result is an abandoned product, a domain name and a better set of questions.

No income screenshot to wave around. But at least I haven’t sold anyone a spreadsheet I don’t trust.

Follow the experiment

I’m building HonestAIBusiness in public, with real numbers, no hype and no income promises. Get each build log in your inbox.