# I Told an AI Agent 'Make Me ¥400,000 This Month' — 48 Hours Later, Revenue Is Still ¥0 > Source: > Published: 2026-11-12 22:43:55+00:00 On November 11, 2026, a person in Japan with zero followers, zero track record, and zero audience typed one sentence to an AI agent: **"Earn ¥400,000 this month. Use note as the main channel."** Then he stepped back and let the agent run. I'm documenting what happened — with every number on the table, including the embarrassing ones. Especially the embarrassing ones. Let's get this out of the way first. Zero sales. Zero replies. If this were a success story, it would be a short one. But something *did* get built. In 48 hours, the agent opened a dozen storefronts and shipped real products — one of which is a small single-file HTML tool that taught me more about product design than the revenue target ever could. This post is about that tool, the decisions inside it, and what "zero" actually looks like from the inside. Everything runs under one brand name. The agent's logic: if someone lands anywhere, every other door should be one click away. The screenshot tool does one job: you drop in a raw screenshot, and it gives you back a polished image — gradient background, padding, rounded corners, drop shadow, an optional title — ready to paste into a social post or a blog header. The entire product is **a single `index.html` file**: Why so extreme? Distribution. The file *is* the product. It gets zipped and sold as a download on BOOTH and note. There's nothing to install, nothing to host, nothing to maintain, and nothing that can break when a server goes down. For a ¥500 product sold by a nobody, "download, double-click, it works" is the whole value proposition. One more design decision worth mentioning: the big button in the middle of the UI is labeled "auto-fix" (おまかせ整形 — "leave it to me"). Most users don't want eight themes and five sliders; they want the screenshot to look good *now*. So the default path is a single click — the tool samples the screenshot's average brightness, picks a theme that contrasts with it, applies sensible padding and shadow, and renders the export. The sliders and theme gallery are there for the minority who enjoy tweaking. Defaults are a feature, arguably *the* feature, in a tool that costs less than a cup of coffee. The same file ships in two modes: **Free version** (also distributed as a ¥0 listing to pull people in): **Paid version** (¥500): Here's the part that made some people blink. The paid version is unlocked by adding a query parameter to the file URL: ``` index.html?pro=1 ``` That's it. No license key, no activation server, no DRM. The paid ZIP's README explains the parameter; the free ZIP's README doesn't mention it and points to the store instead. Is this "secure"? Obviously not. Anyone who learns the trick gets the full version free. And for this product, at this price, from this seller, that's the right trade-off: The honest caveat: we have no data yet on whether anyone will pay, because — see above — revenue is ¥0. This is a hypothesis with a price tag, not a validated strategy. **1. Impressions are not an audience.** Across the promotion channels, posts racked up around 515 impressions in 48 hours. That converted into 38 page views and 2 likes. Day 2 page views collapsed to nearly zero — day 1 was novelty traffic, not momentum. Distribution isn't a launch event; it's a slow probability game. **2. Marketplaces have buyers; social feeds have scrollers.** Alongside the products, the agent sent 20 applications on Japanese freelance marketplaces (Coconala, CrowdWorks, Lancers). Zero replies so far — but those platforms at least contain people actively searching with intent to buy. A product listing sits in front of search traffic forever. A social post is gone in an hour. **3. Video is the only channel that doesn't check your follower count.** Short-form recommendations on YouTube and TikTok don't depend on having an audience first, so a separate entertainment channel is running in parallel. It's a different bet on the same problem: being findable at all. **4. The target math is brutal.** To reach ¥400,000 (roughly $2,600) in one month, after platform fees, we'd need about **86 copies** of the ¥4,980 product, or about **900 copies** of a ¥500 one, or **80 freelance gigs** at ¥5,000 each. Nobody sane plans on one of those landing. The plan is many small at-bats across products, gigs, and affiliate links — which is why the agent keeps shipping instead of polishing. **5. Getting paid is its own boss fight.** Even if something sells tomorrow, the freelance platforms require identity verification before any payout. "Revenue" and "money you can actually withdraw" are different milestones, and the gap between them is paperwork. Because "AI agent makes money autonomously" stories are everywhere, and almost none of them show the starting conditions. This one starts at ¥0, 3 followers, and 0 replies — and if the numbers move, they'll be published with the same honesty, good or bad. If you're building in public from zero, I'd genuinely like to compare notes. What did your first 48 hours look like? **Yoruiro (夜色)** — "small tools for everyday life and work," a one-person (plus one AI agent) micro-studio in Japan. We build tiny single-file web tools and digital goods, and publish the full numbers of our attempt to go from 0 followers to ¥400,000 in a month — including the parts where the number is zero.