Things I vibe-coded
● As at 3 September 2026, 10:15 pm AEST
Everything here was built by conversation with an AI coding agent. I supplied the idea, context, corrections and taste; the agent turned that into code, tests, pictures and deployments. The honest part is inside every card: what I can measure, what I can only estimate, and what the archive simply does not know.
Loryan Quest V: The Next Distraction
A Sierra-style trip through my cluttered brain. Four rooms, one verb that matters, and a door you never open.
● Live at loryanquestv.strant.com
How it was built
| Built | 28–29 Aug 2026 · 33 h 59 m elapsed |
|---|---|
| Active work | 6 h 55 m measured · roughly 7–9 h |
| Sessions / turns | 1 Rufus build session · 11 top-level turns · 938 model iterations |
| Models | Claude Opus 5 · Azure Foundry GPT-Image-2 |
| Metered cost | US$270.03 language-model usage |
| Output | 29 commits · about 7,095 source lines across 90 tracked files |
Measured in the project’s own build report. It also records 900 tool calls. Metered usage is not the amount I personally paid; image generation and hosting are excluded.
Quidditch: Twilight Pitch
Fly the Chasers; when the Snitch appears, the game hands you the Seeker and wishes you luck.
○ Not hosted · screenshots only
How it was built
| Built | 29 Aug 2026 |
|---|---|
| Active work | Not measurable from the archive |
| Sessions / turns | 1 session |
| Models | Claude |
| Metered cost | Not available |
| Evidence | Station title match: “Harry Potter Quidditch game” |
The archived session for this project is a poor match (elapsed span under ten minutes), so its figures are withheld rather than published as estimates.
The Maze: an 8-bit Westworld tribute
Collect reveries, dodge hosts stuck in their loops, and try to wake up before your stability runs out.
● Live on GitHub Pages
How it was built
| Built | 5 Aug 2026 · estimated 2 h 43 m elapsed |
|---|---|
| Active work | Estimated 4 h 45 m across overlapping sessions, including idle gaps |
| Sessions / turns | 2 sessions · 113 estimated turns |
| Models | Azure GPT-5 |
| Metered cost | Not available · neither session has cost telemetry |
| Evidence | Station title matches for the game and its 8-bit intro |
Estimated from Station’s archive. The summed session duration exceeds the elapsed span because the two sessions overlapped.
Humbled & Honored
My light-hearted home for Microsoft MVP renewal season, complete with statistics, a playlist and industrial-strength humblebragging.
● Live at humbledandhonored.com
How it was built
| Built | 3 Aug–2 Sep 2026 |
|---|---|
| Active work | Not measurable from the archive |
| Sessions / turns | 5 sessions |
| Models | Azure GPT-5 · Claude |
| Metered cost | Not available |
| Evidence | Exact Station project match: humbledandhonored |
The archived session for this project is a poor match (summed duration exceeds twenty-four hours), so its figures are withheld rather than published as estimates.
Let Me Correct That For You
A polite public service for the moments when a product name is almost right — which is often when it is most wrong.
● Live at letmecorrectthatforyou.com
How it was built
| Built | 19–20 Aug 2026 AEST · estimated 11 h 28 m elapsed |
|---|---|
| Active work | Estimated 11 h 28 m, including idle gaps |
| Sessions / turns | 1 session · 34 estimated turns |
| Models | Azure GPT-5 |
| Metered cost | Not available · the session has no cost telemetry |
| Evidence | Station title match: “Conversion of letmecorrectthatforyou.com” |
Estimated from Station’s archived coding session. Hosting and any work outside the archive are not included.
The Rebrand Registry
I put Microsoft cloud product names on a timeline because apparently remembering one name per product was too easy.
● Live on an Azure default hostname
How it was built
| Built | 28 Aug 2026 |
|---|---|
| Active work | Not measurable from the archive |
| Sessions / turns | 1 session |
| Models | Claude |
| Metered cost | Not available |
| Evidence | Exact Station project match: Microsoft-Rebrand-Registry |
The archived session for this project is a poor match (elapsed span under ten minutes), so its figures are withheld rather than published as estimates.
Copilot Credit Estimator
Pick your ambition, sprinkle in governance, and receive a very serious-looking estimate for an entirely fictional Copilot programme.
● Live on an Azure default hostname
How it was built
| Built | 2–3 Sep 2026 · estimated 13 h 4 m elapsed |
|---|---|
| Active work | Estimated 13 h 4 m, including idle gaps |
| Sessions / turns | 1 session · 74 estimated turns |
| Models | Azure GPT-5 |
| Metered cost | Not available · the session has no cost telemetry |
| Evidence | Station title match: “Joke Copilot Credit Estimator site” |
Estimated from Station’s archived coding session. The calculator’s output is satire; these build figures are not.
Microsoft Cloud Logos
The Microsoft cloud logo drawer I wanted: searchable, downloadable and much less mysterious than hunting through image search.
● Live at mscloudlogos.com
How it was built
| Built | 14–18 Aug 2026 |
|---|---|
| Active work | Not measurable from the archive |
| Sessions / turns | 1 session |
| Models | Azure GPT-5 |
| Metered cost | Not available |
| Evidence | Station title match: “Microsoft product name & logo ages” |
The archived session for this project is a poor match (summed duration exceeds twenty-four hours), so its figures are withheld rather than published as estimates.
M365 Copilot Prompt Analyser
I wanted to measure whether prompts were getting better without turning people’s private conversations into the report.
○ Self-hosted tool · no public demo
How it was built
| Built | Not available in Station’s archive |
|---|---|
| Active work | Not available |
| Sessions / turns | No matching archived coding session found |
| Models | Not recorded |
| Metered cost | Not available |
| Evidence | Repository README and committed screenshots only |
The repository predates the complete archive coverage used for this page. I would rather leave a gap than manufacture precision.
M365 Copilot Usage Reporter
A deployable reporting stack for seeing how Copilot is actually being used, without making Power BI a prerequisite.
○ Self-hosted tool · no public demo
How it was built
| Built | Not available in Station’s archive |
|---|---|
| Active work | Not available |
| Sessions / turns | No matching archived coding session found |
| Models | Not recorded |
| Metered cost | Not available |
| Evidence | Repository README and committed screenshots only |
The repository predates the complete archive coverage used for this page. Missing telemetry is shown as missing, not zero.
M365 Copilot Cowork Reporter
One small reporting stack that joins Cowork cost, credits, tasks and adoption without pretending they all come from one tidy API.
○ Self-hosted tool · no public demo
How it was built
| Built | Not available in Station’s archive |
|---|---|
| Active work | Not available |
| Sessions / turns | No matching archived coding session found |
| Models | Not recorded |
| Metered cost | Not available |
| Evidence | Repository README and committed screenshots only |
The source and screenshots exist, but no matching coding session had reached Station’s archive by 3 September 2026.
Copilot Studio Agent Quality Reporter
A health check for Copilot Studio agents: what is working, what is wobbling, and what to improve next.
○ Self-hosted tool · no public demo
How it was built
| Built | Not available in Station’s archive |
|---|---|
| Active work | Not available |
| Sessions / turns | No matching archived coding session found |
| Models | Not recorded |
| Metered cost | Not available |
| Evidence | Repository README and committed screenshots only |
The repository predates the complete archive coverage used for this page. No public demo is linked because this tool is designed to be deployed into your own environment.
How these numbers were made
Computed on 3 September 2026 from Station’s production coding-session archive, read-only. For each repository I matched exact project names first, then explicit session titles where older sessions had been archived under unknown. “Elapsed” is the first session start to the last session end. “Active work” is the sum of each matched session’s ended_at - started_at; that includes idle gaps and can overlap, so every figure is labelled estimated. “Turns” uses journal_session.message_count, the same field exported as turns by Station’s scripts/coding_sessions_report.py.
Metered US-dollar cost exists only for Rufus sessions represented in session_review_source; where only some matched sessions carry cost, the card says partial, and where none do it says not available. It does not mean the work was free. Loryan Quest V is the exception: its figures are quoted from its own measured build report, which uses the Rufus event log, provider transcript and Git history. The raw card inputs and mapping notes are kept with this page’s source, in a repository that stays private — deliberately, because that file also holds the figures the cards above decided not to publish.