Seven Questions to Ask Before Buying an "AI One-Person Company" System
Table of Contents
- What these products actually are
- Question one: what exactly is delivered
- Question two: what does it run on, and what do I supply
- Question three: where is the refund written, and do both documents say the same thing
- Question four: is there visible operating evidence
- Question five: does the seller use it
- Question six: can the items in the value stack be bought separately
- Question seven: what do updates and community actually promise
- The advanced question: where does your data go
- Running the same checklist on ourselves
- Closing
"AI employees", "one-person company systems", "hire a whole AI team": products in this category have multiplied over the past few months, priced anywhere from a few dollars to several hundred. The sales pages are usually well made. Well made and substantial are two different things.
This post names nobody. It is a checklist to run before you pay. Sellers who can answer all seven get to talk about price. At the end I run the same checklist against what we sell, including the boxes we fail.
What these products actually are
Whatever the name, the substance is the same: a pre-written set of rule files, prompts, knowledge-base structure and tool configuration that you install into your own AI coding assistant. Claude Code and Codex are the common hosts.
It usually does not include the AI. You bring your own platform subscription; model costs and third-party API costs are extra. What you are buying is the how-to-use layer.
There is nothing wrong with that. We sell something in the same layer. The problem is that this layer is hard to judge from outside, which is why the buyer needs a checklist. As for what actually separates an "AI employee" from an AI tool, we took that apart into four differences separately; this post is only about what to ask before paying.
Question one: what exactly is delivered
The most basic question, and the one most often left unanswered.
Ask for the file list, file formats, sizes, where they live, and how to install. "After payment you receive a delivery email with all files and setup instructions" is not an answer. It describes a process, not the contents.
Follow up with one more: can I see the table of contents first. A seller who shows the contents and a seller who offers adjectives are two different sellers.
Question two: what does it run on, and what do I supply
Three sub-questions:
- Which operating systems? Windows, macOS and Linux, or only one? If no operating system name appears anywhere on the site, be careful.
- Which subscriptions are required? Claude, ChatGPT, Codex: which is mandatory and which optional.
- How much do the API costs add? "Third-party costs not included" is honest, but honest is not the same as cheap. Ask for a range.
If the seller says you need to be "willing to follow the setup guide and handle the necessary permissions", translate it: you will be using a terminal. Not a bad thing, but know it going in.
Question three: where is the refund written, and do both documents say the same thing
Open two pages together: the product page and the terms of service.
A "seven-day refund guarantee" badge on the product page is common. A terms clause saying "digital content delivered online is exempt from the cooling-off period once you have consented to the service and begun using it" is also common. Both appearing on the same domain at the same time is not rare.
One thing first, so nobody reads this as "the badge means nothing": in Taiwan the promise on the product page carries weight. Article 22 of the Consumer Protection Act says a business must ensure its advertising is truthful, that its obligations to the consumer may not fall below the content of that advertising, and that after a contract is formed the advertised content must be honoured. A purchase page that says seven-day refund sets a floor the terms cannot lower.
So why read the terms at all? Because two documents contradicting each other is a signal in itself: this seller's paperwork has not been reconciled, and getting a refund from them will probably mean citing Article 22 yourself. Buying a product should not require knowing the statute. A seller whose terms and product page say the same thing is saving you that effort.
Question four: is there visible operating evidence
Three tiers:
| Tier | What it looks like | Worth |
|---|---|---|
| Illustrated interface | A beautifully typeset "to-do card", "quote", "+XX% growth" on the sales page | Zero. That is a design mockup, not a product |
| Real screenshots | An actual application window, a cursor, real dates, imperfect fonts | Some reference value |
| Screen recording plus outputs | The full path from input to output, plus the actual posts, slides or reports it produced | This is evidence |
One simple check: does the same seller have real recordings on their other product pages. If the video-editing tool has before-and-after footage and the flagship product has none, that is not a capability gap. It is a choice.
Question five: does the seller use it
"I use it every day" is a claim, not evidence.
What you want is publicly verifiable output. For example:
- Merged pull requests you can open, where the link goes to github.com rather than the seller's own site.
- Articles with dates you can check one by one, rather than "three accounts over ten thousand followers" with no account names.
- Online tools you can actually operate.
The test: if this evidence were fake, would a third party be able to tell. A maintainer's merge record, a platform's follower count, a timestamped public article all pass. The seller's own screenshots do not.
Question six: can the items in the value stack be bought separately
Many sales pages carry a table: "total value $X,XXX, today only $XXX".
Pull it apart three ways:
- Does each item in the stack have its own product page? Items with no page cannot be bought alone. They exist to raise the total.
- For items that do have a page, does the stack price match the page price? A small difference per item makes the total look much better.
- Does the page itself carry small print like "value stack uses estimated pricing"? If so, the table is marketing, not a quote.
Question seven: what do updates and community actually promise
- How long do updates run? "Ongoing updates" means how long? Does it still work after that?
- What does year two cost? Many pages only show year one.
- What is the community called, on which platform, and how many people are in it? "One year of community access" contains no information.
A community with no name and no headcount should be treated as not existing yet.
The advanced question: where does your data go
One selling point of these systems is "Gmail, calendar, IG and Threads already connected". Every connected service is another place your data flows.
Ask four things: whether your content goes into training by default, how much is collected, how long it is retained, and whether you can withdraw it. Put those four to the seller directly. We wrote them for people building products; they work just as well coming from a buyer.
Running the same checklist on ourselves
We do not sell an AI employee system. We sell a handbook: the Claude Code handbook, with a free 26-page starter and a NT$490 production-discipline edition of 102 pages plus a toolkit.
Seven questions, our own answers, failures included.
| Question | Our answer |
|---|---|
| Deliverable | PDF plus toolkit, download email sent immediately after payment. Page counts are on the purchase page. The table of contents is visible before buying |
| Runs on | The handbook is a PDF, any machine can read it. To actually follow it you need your own Claude Code subscription; that sentence now sits directly under the buy button (it was missing when this post first went up, added the same day) |
| Refund | Email within seven days and you get a refund, stated on the purchase page, and section 11 of the terms of service now carries the same promise. When this post first went up there was no such clause and the promise lived on one line of the purchase page; added the same day |
| Operating evidence | The verification script in the book was blocked by our own website the first time we ran it, and that case is in the book unedited. There is no screen recording. That box we fail |
| Seller uses it | The workflow in the handbook is the one we run every day. Verifiable output: 28 pull requests merged into 15 organisations, 277 dated articles (as of 21 September 2026), 5 live products. None of it is our own screenshots |
| Value stack | None. One price, no "total value" table |
| Updates and community | One-time purchase includes all future updates. No community is promised, because there is none |
Three boxes failed when this first went up: the purchase page did not make "you need your own Claude Code" clear, the refund promise had no formal terms behind it, and there is no screen recording. The first two were fixed the same day and the record stays here; the recording is still missing.
Closing
The point of the list is not to catch anyone out. It is to separate "looks substantial" from "is substantial".
None of the seven questions needs a technical background. A seller who can answer them will be glad to, because those answers belonged on the page in the first place. A seller who cannot has already given you your answer.
Running a company alone really can go further with AI. We are one person plus AI; we have written about how we split the work between models and a human and about the decisions we do not delegate to models. Having walked it is exactly how you learn which promises can be kept and which cannot.
Seven questions before you buy beat seven days of regret after.
FAQ
What is an "AI one-person company system"?
A pre-written bundle of rules, prompts, knowledge-base structure and tool settings that you install into your own AI coding assistant, typically Claude Code or Codex, so it can play several roles: copywriter, editor, researcher, support. It usually does not include the AI itself. You bring your own platform subscription; what you buy is the how-to-use layer.
What is the difference between an AI employee and an AI tool?
"AI employee" is a product narrative. Underneath it is still a model plus a set of rule files. The real differences are shared memory, continuity across tasks, and a review step. You cannot judge that from the name. You judge it by whether the seller can show real screen recordings and real outputs rather than illustrated interfaces.
What is the single most important question to ask before buying?
What exactly is delivered. Ask for the file list, formats, sizes, the operating systems supported, and which subscriptions and API costs you must supply yourself. A seller who cannot answer this one has answered all the others.
Prices range from a few dollars to hundreds. How do I judge what is reasonable?
Check three things: whether each item in the value stack can be bought on its own, whether the stack price matches the standalone product page, and how much the AI platform and API costs add on top. Pulling the total value apart is more useful than comparing totals.
Where should I look for the refund guarantee?
In two places at once: the product page and the terms of service. A product page that says seven-day refund next to terms that say digital content is exempt from the cooling-off period is not rare. Under Article 22 of Taiwan's Consumer Protection Act a business's obligations may not fall below its advertising, so the page promise is not void; but two documents that disagree are a signal that you may have to cite the statute yourself to get the refund.
How can I tell whether the seller actually uses the product?
Ask for publicly verifiable output rather than a claim. Merged pull requests you can open on github.com, articles with dates you can check one by one, online tools you can actually run. "I use it every day" is a sentence, not evidence.