THE NUMBER
$0.12.
That's the average cost to generate one digital product with an AI agent - research, writing, formatting, and listing included. At that price, 100 products cost less than a coffee. The marginal cost of trying one more idea is basically zero.
That changes what's worth attempting.
3 THINGS HAPPENING RIGHT NOW
The overnight Etsy experiment A developer set up an agent to scan Reddit for recurring complaints about wedding planning. The agent identified 12 pain points, generated a checklist/planner for each one, and listed them on Etsy as digital downloads. Total time from idea to live listings: one night. Three of the twelve have sold in the first week.
The newsletter-to-product pipeline A solo creator runs a niche newsletter about remote work tax strategies. He noticed the same five questions coming up in reader replies. He pointed an agent at those questions, had it produce a 30-page guide with state-by-state breakdowns, and now sells it as a $19 PDF linked from every issue. Took an afternoon to build. Makes $400-600/month on autopilot.
The support-ticket gold mine A SaaS founder exported 6 months of customer support tickets and fed them to an agent. It clustered the complaints, identified the top 10 feature gaps, and drafted a product brief for each one. She didn't build all 10 - she built the one that showed up 200+ times. That feature reduced churn by 15% in the first month.
THE DEEP DIVE
How to build a full commercial loop - step by step
Last issue I talked about AI agents completing an entire business cycle without human input. This week: how it actually works, broken into five stages.
Stage 1 - Scan The agent monitors a source - Reddit, Twitter, support forums, Amazon reviews - looking for the same complaint appearing over and over. Not one angry post. Clusters. The signal is repetition. If 200 people are frustrated by the same thing, that's a market.
Stage 2 - Validate Not every complaint is a product. The agent checks: Is anyone already solving this? How much are they charging? Are people actually paying, or just complaining? A complaint with no existing solution and evidence of willingness to pay - that's the green light.
Stage 3 - Create The agent builds the product. For digital products, this means writing, structuring, formatting, and packaging. A guide. A template set. A toolkit. The quality bar is "would someone pay $10-20 for this and feel like they got value?" Not a masterpiece. A useful thing.
Stage 4 - List The agent creates the listing - title, description, pricing, thumbnail. Platforms like Gumroad, Etsy, or Lemon Squeezy make this straightforward. The listing copy matters more than most people think. The agent writes it based on the exact language people used when describing their problem.
Stage 5 - Sell This is where most loops stall. Having a listing isn't distribution. The builders who make this work feed the product back into the community where the complaint originated - not as spam, but as a genuine answer to the question people were asking. The agent that found the problem is also the agent that knows where the buyers are.
Where it breaks: Most people nail stages 1-3 and skip 4-5. They build the thing and wait. The loop only works when the last stage connects back to the first. The agent found these people. It knows where they are. Use that.
The whole system runs on a schedule. Scan Monday. Validate Tuesday. Create Wednesday. List Thursday. The human reviews once a day, kills anything that doesn't look right, and lets the rest run.
ONE THING TO TRY THIS WEEK
Week 3 of 6: Ask Claude to build something real
The first two weeks were setup. This is where it gets interesting.
Open your project folder from last week. Tell Claude to build something and save it to a file. Be specific about what you want. Here's an example:
Read my CLAUDE.md for context. I get leads from
three different sources and I keep losing track
of follow-ups. Build me a simple HTML page that
tracks leads - name, source, status, next action,
and date. Save it to lead-tracker.html
Claude will write the code, save the file, and you can open it in your browser. A working tool, built in minutes, tailored to your exact situation.
The key: describe your actual problem, not a generic one. The more specific you are about your workflow, the more useful the output.
Try it. Open the file. Use it for a day. That's the moment this stops being theoretical.
Stuck? Reply to this email.
WHAT'S COMING
Next week: giving Claude eyes. You'll install a plugin that lets it see and interact with the web - and that's when things get genuinely powerful.
Manu