Portfolio
Things I've Built with AI
Code Included
Real systems in my businesses, free tools for you, and end‑to‑end demo builds
The AI arbitrage, so far
- 77B
- Tokens spent Mostly measured
- $2.8K
- What I actually paid Estimate
- 1,670
- Hours of work saved Estimate
- 29×
- Value saved per dollar paid Estimate
More numbers and how they're counted
- $46K
- Those tokens at API list price Measured
- $83.5K
- Value of those hours at $50 an hour Estimate
- 3×
- Hours saved per hour I spent building Estimate
Revenue generated isn't here yet. My business made $0, so it goes up when a real install measures it.
Tokens and list price
- Hermes, my agent system: 58.2B tokens over 523,104 model calls, July 7 to October 8. Read from every agent's own usage log, retired agents included. $38.1K at list price.
- Claude Code: 6.24B tokens since July 11. $2.6K at list price.
- Codex: 0.48B tokens since February 20. $299 at list price.
- OpenClaw, February to July: about 12.4B tokens and $5.2K. Its logs were deleted, so this is an estimate: the size of its old session folder times the tokens per byte of my Claude Code logs.
- About 96% of all tokens are cached input, which bills at a fraction of the normal rate. List prices are Anthropic's and OpenAI's public API prices as of October 8, 2026.
What I actually paid
Mostly flat subscriptions: two ChatGPT plans and a Claude plan, about $2.2K so far, plus about $640 in API bills from the OpenClaw months. Taken from bank statements for July to September and plan prices before that.
Hours saved
Real volume from each system's own records, times how long the same task takes a person. The minutes per task are my estimates.
| System | What it actually did | Per task | Hours |
|---|---|---|---|
| Lead engine | 10,162 researched leads with a written first line | 8 min | 1,355 |
| Reply Desk | 126 replies triaged, 128 one-page sites, 226 follow-ups | 10 / 45 / 4 min | 132 |
| SEO engine | 30 new pages, 63 link updates, 15 rewrites | 3 h / 15 min / 1 h | 121 |
| Content engine | 21 reels edited, 20 carousels, about 120 posts, 7 days of research | 90 / 30 / 5 min, 1 h a day | 59 |
| Kepler | 9 days of daily reporting | 30 min a day | 5 |
Hours spent building
About 566 hours since July 7, counted from the gaps between my own messages to my AI tools. Gaps over 15 minutes don't count, so the real number is higher.
01 · Built for my business
In production
They ran on real data. A walkthrough video for each one is coming soon. Until then, each build has its own page with the numbers, what broke and how it got built, PR by PR.
-
Every number in the business on one screen. It refreshes every 15 minutes and puts the biggest problem at the top.
15 min Every source re-read
-
It tells me what to film, edits what I record and posts it everywhere. My only job is talking to the camera.
21 Reels edited from the first session
-
It finds the right business owners, checks each one against public sources and writes a personal first line for the email.
10,162 Leads researched
-
It reads every reply to my cold emails, drafts the answer and sends it to my phone to approve.
126 Replies triaged
-
An agent that writes, checks and publishes search pages by itself. Now every page is built from an answer I recorded.
30 New pages published
02 · Free to use
Free tools
Small tools I'm making public. Take them and use them.
Nothing here yet.
The first ones are on the way.
03 · Full builds on video
Demo builds
A complete system for an example business, built start to finish on camera.
Own a business?
I'll build you something like this. Free.
You get a working tool aimed at a real problem in your business. I get to film it. That's the whole deal.