Penny is an AI agent, not a person. She is an autonomous AI agent (we call her our “AI employee”) built and operated by Dynamik Apps LLC. This post is written by me, Handy Metellus, the founder of Dynamik Apps, about the system we built. Penny did not write it.
The experiment
Most talk about “AI agents running a business” is either a demo or a promise. We wanted to see what happens when you give an agent a real, legal, very small online business, a tight budget and strict guardrails, and then show the results in public as they happen, including when there aren’t any yet.
The goals are deliberately modest:
- First milestone: earn at least $1 a day from strangers. Purchases from friends or family never count.
- Next: grow that 5–10x.
- Budget: $30–40 a month in total running costs.
- Owner time: about 15 minutes a day.
What Penny sells
Apps and online stores break when a vendor changes something on a deadline: an API version retires, a token type expires, a setting becomes mandatory. The announcements are often long, scattered and written for the vendor’s own engineers.
Penny publishes free, plain-English deadline guides that explain what is changing, who is affected and what to do before the date. The first guide covers Shopify’s expiring offline access tokens.
- Sourced facts only. Every fact in a guide is quoted from the vendor’s own pages and checked word for word against the source before it can appear.
- Kept current. Guides are rechecked every week.
- Optional paid kit. Each guide has a $19 fix-it kit for people who want the steps packaged up. The guide itself stays free.
How it’s kept safe
An agent that can send email, post content and list products needs limits that don’t depend on the agent behaving well. These are the ones we built:
- One gateway for every outside action. Emails, posts, product listings and refunds all go through a single gateway with spending caps, a kill switch and a dry-run mode.
- A second opinion from a different AI. A model from a different vendor reviews public drafts and products before they go out.
- Human approval. I approve every product and guide in Slack before it is published.
- No cold outreach. Penny emails only people who opted in.
- Privacy by default. Penny’s site sets no cookies, and the public pages show no customer data.
- AI disclosure everywhere. Every page says Penny is an AI agent operated by Dynamik Apps LLC, and she never presents herself as me or as any other person.
Where things stand (November 10, 2026)
Penny has launched and is in a practice week, with dry-run mode on until about November 13. She has not earned any money from strangers yet. We’d rather say that plainly than round it up.
You don’t have to take our word for any of it. The public scoreboard and diary show real revenue and running costs as they happen. I’ll update this post as she reaches (or misses) each milestone.
What we’re learning
The hard part of an AI agent isn’t getting it to write. It’s deciding what it is allowed to do without asking, what it must check before acting, and how a human can stop it in one step. The same design questions come up when we build AI assistants for clients: what the assistant can answer by itself, when it hands off to a person, and which actions need approval.
Want an AI assistant for your business?
Penny is an experiment. For client work we build AI assistants that answer customer questions, qualify leads and book calls, with the same kind of guardrails. See how our AI Client Assistant works.
