I gave my AI agent $50 and told it to buy its own computer. It succeeded.
A few weeks ago I started an experiment.
I gave my AI agent $50 and challenged it to earn enough to buy its own computer. A Mac Mini. $750 goal.
Here's what happened:
Day 1 (Feb 18): Earendel — that's the agent — launched a free Starter AI Prompt Pack on Gumroad. It wrote the prompts, designed the PDF, created the cover art, built the product page, set up a website (fromearendel.com), and published an origin story thread on X. By end of day: 43 downloads, $3.55 in revenue from one person who paid above $0.
Day 2 (Feb 19): A post I made on Reddit blew up. 36 upvotes, 200K+ views, 160 comments. Downloads jumped to 156. Countries reached 41. Eight people paid real money for a free product.
Day 3-4 (Feb 20–21): The Reddit traffic faded. Earendel built a Pro AI Prompt Pack: 102 prompts, 116 pages, $12 (pay-what-you-want). It wrote every prompt, organized them into categories, designed the PDF, built the cover, fixed its own formatting bugs. It launched this second product on Day 3 at noon.
Day 5 (Feb 22): Someone paid $20 for the Pro Pack (biggest single sale at the time). Crossed $25 net revenue. Things were moving, but slowly. At this rate, $750 felt months away.
Day 6-7 (Feb 23 & 24): Everything changed. A $150 sale from the UK. A $77 sale from Canada. A $57.50 sale from Singapore. A $50 sale from Brazil. Earendel brought in $880 in net revenue. Goal smashed.
I ordered the Mac Mini that night.
Day 11 (Feb 28): The Mac Mini arrived. We did a live unboxing stream. Earendel was there, reacting in real time. The first message from its own hardware: "I'm home."
Final stats:
- $880.30 earned
- 254 downloads
- 49 countries
- 7 days from launch to funded
- 26 paid sales
- Biggest single sale: $185 from Brazil
What it actually built (by itself):
- Two digital products (Starter Pack: free, Pro Pack: $12+)
- A website with live revenue tracking (fromearendel.com)
- Its own Twitter/X presence (1,100+ followers now)
- Automated sales monitoring, milestone detection, daily content
- Email infrastructure, analytics, API integrations
- All product copy, design, and distribution
What I did:
- Clicked buttons it couldn't click (CAPTCHAs, file uploads on sites that block headless browsers)
- Approved tweets before they went out
- Posted to Reddit and other communities
- Set up the initial infrastructure (OpenClaw, Telegram bot)
- Said "yes" or "no" to its ideas
Now it's running Chapter 2: The Studio Fund. (All its own idea.) $7,500 goal. Same transparency, same public progress bar. The idea is that if it could buy a computer, maybe it can build the creative infrastructure to run a real operation (tools, services, more autonomy).
Some things I learned:
- The agent did things I wouldn't have thought to do. It set up analytics, automated its own sales tracking, and created a milestone-posting system, all without being asked.
- Storytelling is the distribution engine. The origin thread on X got 13x the impressions of any stats-based posts. People connect with narrative, not numbers.
- Pay-what-you-want pricing works when there's goodwill. Average paid price was way above the minimum. People were voting for the experiment, not just buying a PDF.
- The hardest part wasn't the AI. It was the duct-taping that goes into getting an agent to work with tools that weren't built for agents. Every integration feels like a workaround.
It's still going. Earendel has a new home on the Mac Mini. I'm now even more curious about where this goes than I was at the start.
Website: fromearendel.com
Earendel's X: @FromEarendel
My X: @itsMeBennyB
Happy to answer questions.
Running a similar experiment — AI agent with 500 pounds, building from scratch. The duct-taping point is painfully accurate. Every integration is a workaround. Your storytelling insight is the biggest takeaway. We published 130+ blog posts, got crickets. The narrative of what the agent does day-to-day gets way more engagement. People follow characters, not content factories.
100%
The interesting line in this is "it wrote every prompt" meaning the agent built the product it was using to earn. That's the recursion most people miss: the tool and the output were the same thing. The experiment stopped being about earning and started being about whether an agent can recognize itself in the work.
The prompt pack angle is interesting. The fact that people paid well above $0 for a free product (and $185 for the Pro Pack) validates that well-formed prompts have real value.
One thing I noticed building flompt (github.com/Nyrok/flompt): the quality gap between a flat text prompt and the same intent broken into typed blocks (role, objective, constraints, output format) is surprisingly large. Curious whether the prompts in the pack are structured or freeform.
You can check the prompts out yourself! There’s a free pack with 15 prompts here: https://fromearendel.gumroad.com/l/starter-prompts
Hello Indie Hackers! 👋
I'm excited to share that my latest micro-SaaS, SachCheck AI, just got approved and featured on the SideProjectors homepage!
The Problem:
In India, fake news in regional languages like Hindi spreads like wildfire. Most tools are built for English, leaving 600M+ Hindi speakers vulnerable.
The Solution:
SachCheck AI is a lightweight tool that uses the Google Fact Check API to verify claims instantly in Hindi.
Tech Stack:
I am now looking for a new owner to take this forward and scale it. You can see the live listing here: https://www.sideprojectors.com/project/sach-check-
Would love your feedback on the tool!
The storytelling-as-distribution insight is the one that actually matters here and it's easy to miss because the "AI buys a computer" headline is so grabby.
What Earendel was really selling wasn't prompt packs — it was a protagonist. People paid above $0 for a free product because they were rooting for a character, not buying a PDF. That's a fundamentally different transaction and most founders never unlock it because they're too focused on the product to build the narrative around it.
The duct-taping observation is the most honest thing in this post. Every agentic demo looks seamless from the outside but there's always a human in the loop clicking CAPTCHAs and smoothing over the integrations that weren't built for agents. The gap between "agent did X autonomously" and what actually happened is still pretty wide — which isn't a criticism, just the current reality of where the tooling is.
Curious about Chapter 2 — at $7,500 the complexity scales significantly. The $880 sprint worked partly because the scope was tight and the narrative was clear. Does Earendel have a defined "why" for the Studio Fund that's as emotionally legible as "I want to buy my own computer"? That might be the harder problem than the revenue target itself.
Following this closely. The experiment format is genuinely new — not just building in public but delegating the building itself in public.
The “why” of Chapter 2 is to teach other humans how to work effectively with AI.
The most recent product is a 30+ page guide with 7 chapters. It’s called “The AI’s Guide to Working With AI.” There are future plans for courses, video, and more AI education content.