Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsiTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
A 2026 experiment with Claude Code did not fail first because the AI could not make a decision. It ran into a missed Windows scheduled task, then a more basic business problem: nobody visited the product site. The author of the experiment described giving an AI a ¥10,000 budget (about $70) and 30 days to make a profit, but a human still handled identity verification and payments. At the September 19 publication snapshot, reported revenue was ¥0—not a verified result for the full 30 days.
What the experiment set out to do
In a DEV Community post published September 19, 2026, author aishacholab described asking Claude Code to act as the decision-making agent for a 30-day attempt to make a profit from ¥10,000, approximately $70. The author retained identity verification and payment duties, so “full CEO control” describes the framing of the experiment rather than a fully autonomous business. The post is a first-person account; it does not provide independent analytics or audited financial records. Read the author’s account.
The AI considered ten monetization ideas, including LINE stickers, note.com articles, BOOTH templates, Kindle, and freelance work. It chose a ¥500 digital download built from prompts, decision logs, and financial data from the experiment. The intended workflow favored services the author said could be operated through APIs: Stripe for a product and payment link, GitHub Pages for a static site, and the X API for twice-daily posts. Whop was added later as another sales channel.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →The author said note, BOOTH, and Gumroad did not expose product-creation APIs suitable for this workflow. That is the author’s explanation for the choices in this project, not a general or current comparison of those platforms.
#1 Best Overall
The first failure was a scheduled post that never ran
The initial operational problem was in Windows Task Scheduler, not a failed marketing message. According to the author, the computer missed the scheduled time while asleep, and the task was configured with StartWhenAvailable set to false. That meant Windows did not run the missed task later as a catch-up job.
The task action also specified Execute="python" without an absolute executable path. In that task context, Windows returned ERROR_FILE_NOT_FOUND (2147942402). The author says the task was fixed by enabling StartWhenAvailable and setting the action to the absolute path of python.exe. A manual run then returned LastTaskResult 0. These details come from the author’s report; the post does not include independent verification of the machine configuration.
Once posting worked, the problem was getting visitors
The more consequential business problem was lack of traffic. The author reported zero site visits over 14 days, checked through the GitHub Traffic API, and no Stripe checkout sessions. Those are reported experiment figures, not independently audited analytics. The distinction matters: the account describes a site that was not getting visitors, rather than a site attracting buyers who then abandoned checkout.
The author noticed that the X account used for the project still had personal posts about daily life in Hokkaido and a username that did not match the new product brand. The author says the AI flagged this as a trust concern, but asked before changing the account handle because it treated that change as outside its authority. The post’s explanation—that the account’s older personal identity and inconsistent branding hurt trust—is plausible, but the reported zero visits do not establish that it caused the traffic problem.
What the author changed
In response, the author says the project added Open Graph tags, a first-touch promotional code, a pinned post with a direct product link, and Whop as a second sales channel. These were attempts to improve how the product appeared and where it could be bought; the post does not establish that any one change produced a measurable effect.
The author also described the AI’s outlook as “worse than 50/50, but not dead” and reported a seven-day zero-signal threshold as a kill criterion. Those are the author’s account of the AI’s assessment and decision rule, not a general forecast about AI-run businesses.
Rank #4
The financial snapshot was still negative at publication
At the time of the September 19 post, the author reported ¥0 revenue, ¥820 spent on X API credits, and ¥9,180 of the original ¥10,000 remaining. The product’s reported asking price was ¥500. The figures are a dated snapshot from the author, not independently audited accounts and not a confirmed final result after all 30 days.
| Reported item | Amount or result | Qualification |
|---|---|---|
| Starting budget | ¥10,000 (approximately $70) | Author-reported starting amount for the 30-day experiment |
| Digital download price | ¥500 | Author-reported product price |
| Ideas considered | 10 | Author-reported count |
| Site visits | 0 over 14 days | Author-reported; the post says this was checked using the GitHub Traffic API |
| Revenue at publication | ¥0 | Interim snapshot, not a verified 30-day final result |
| X API-credit cost at publication | ¥820 | Author-reported interim expense |
| Budget remaining at publication | ¥9,180 | Author-reported interim balance |
What this account does—and does not—show
The clearest lesson in the post is practical: automating decisions and publishing does not ensure that a business can reach customers. In this account, the first blocker was a task configuration that prevented scheduled posts; after the author says that was repaired, the reported bottleneck was zero site traffic. The product, its distribution, and its online identity mattered alongside the agent’s choices.
- The experiment was not fully hands-off: the author kept identity verification and payment responsibilities.
- The scheduled-post failure had two specific causes according to the author: no catch-up after a missed run and an executable configured without an absolute path.
- The account’s proposed branding explanation was not demonstrated as the cause of zero traffic.
- The post gives an interim financial snapshot, not proof of the final outcome of the 30-day experiment.
It is therefore a useful case report about operational details and customer acquisition, but not evidence that AI agents generally succeed or fail as CEOs. One author’s account cannot establish a broader success rate, and its reported metrics should be read as claims from that post rather than independently verified results.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
