Proof / The work, day by day
How does the work actually get done?
I direct AI agents in loops. That adds up to about 1,300 hours, 1,257+ prompts and 2,020+ files shipped.
~1,300
Hours working with AI
~100
Days active
1,257+
Prompts logged since Aug
2,020+
Files shipped since Aug
41
Live domains
Logged work, day by day
121 prompts
01 / What the numbers mean
What do these numbers mean?
They show someone steering, not someone pressing a button.
Anyone can type one prompt and wait. That is not how I work. I stay with every build through rounds of building, testing and fixing, and I read what the agents make. For you, that means a finished product that has been checked, not a first draft.
02 / Why it matters
Why does that matter to you?
It means what you buy has been checked, fixed and checked again.
It isn’t prompt-once-and-wait.
Each project goes around the loop many times.
A human is in the loop.
Agents do the heavy lifting. I read their work and send it back when it is wrong.
You can see the work.
Every day is charted above, and the case studies below show real products.
03 / The loop
What is the loop?
Research, plan, build, test, review four times, then iterate.
- 01PlanResearch what exists. A top-tier model plans. Cheaper models build.
- 02BuildAgents write the code, pages and art in small pieces.
- 03TestScripts and real browsers check the work.
- 04Review×4Four reviews in a row before I sign off.
- 05IterateEvery finding goes back into the build.
→ back into the build (02)
The four reviews, in order: a helper agent, a blind agent that has not seen the work before, a top-tier reviewer, and me. Scripts and real browsers catch broken links, clipped text and bad forms in the test step. Every finding goes back into the build, and the loop runs again until the bar is met.
04 / Case study
Case study: ProspectorBuddy
The biggest build so far: 335 prompts and 187 supervised hours, Aug 16 – Sep 26 2026.
It was built across 18 active days. The longest single run held 102 prompts. This is what a real product looks like: the same files reworked again and again until they are right.
- Directed prompts
- 335
- Supervised hours
- 187
- Active days
- 18
- Edits, to 130 files
- 450
05 / Case study
Case study: MECH Stranded
One person, three agents, twelve days.
GAMEPLAY
MECH Stranded is a browser twin-stick shooter. I built it with three agents in twelve days, and you can play it now.
- Days, Jul 16–27
- 12
- Commits
- 170
- Generated art files
- 389
- Specs, reports and reviews
- 193
06 / Timeline
Month by month
Prompts and hours at the wheel for each month since Aug 16.
August
421 prompts
255 hours at the wheel
September
516 prompts
171 hours at the wheel
October
320 prompts
62 hours at the wheel
07 / By project
Numbers by project
| Project | Prompts | Hours at the wheel | Status |
|---|---|---|---|
| ProspectorBuddy | 335 | 187 | Live |
| AI-SEA | 117 | 26 | Live |
| Auto-content portfolio: 40+ sites | 11 | 26 | Live |
| DiscordImageBot | 9 | 25 | Private tool |
| Outbound and CRM system | 152 | 21 | Private tool |
| Deep Claim | 63 | 8.3 | Prototype |
| Boomshaft | 17 | 5.6 | Prototype |
Want this loop on your project?
Start a project