<aside> 🎛️

I wanted to see if I could use ChatGPT from ideation and design all the way through to complete engineering implementation, and measure how much effort the full run took.

I found my available Pro Max usage allotment had dropped by half, so I tapped in Qwen3-Coder for a small amount of unit test updates.

</aside>

• Summary

At a glance

• Work: I created an interface for visualizing people, sessions, handoffs, quality, change, and investment as one transparent plan. I packed UX/UI research and design libraries into a single-page prototype, and used it as multi-modal prompting with ChatGPT and Qwen3-Coder to build it with AI end-to-end.
  • Scheduled update November 12, 2026: See the full Investment Studio on my personal website. Use it to plan out a project or as a visual when conducting experience research.
• My direct time: About 20 hours by my estimate, from the original calculator prototype through the integrated Studio; not reconstructed from commits.
  • The foundation merged on October 28, 2026; the completed version merged on November 10, 2026. [2][1]
  • Some individual AI turns ran 11–13 elapsed hours while I did other things.
• Team: Only me designing and building. Five colleagues and friends graciously took the time to test it and give feedback.
• Result: A working, tested piece of software that is being extended to include embedded AI via Cloudflare AI workers and SystemOne decision models like Clef.
  • Reuse in practice: On November 11, 2026 morning, I used the same planning documents used to build the Investment Studio in a single prompt to create the interactive lab here Studio Rack — Visual Fidelity Contract.

Kapture 2026-11-11 at 14.11.02.mp4

Solution Architecture

Debrief

I wanted teams to see how their work fits together: who participates, when people meet, what needs review, and what the plan asks of them. The Investment Studio carries that idea into a tactile planning experience that other teams can use to visualize cross-departmental partnerships on the way to making good software that helps people.

I make music as a hobby, and I practically live in Jira. So of course I wanted to arrange a project a little like a track: separate parts, measures, changes, and the occasional breakdown that needs another pass. Cassettes, reels, faders, and a sprint board make the work something a team can explore together.

I decided to use generative AI end-to-end, and to measure the effort. Main models used are ChatGPT-5.6 Sol and 6 Astra.

The Bottleneck

• The challenge was to keep a rich, consistent experience together across many sessions and decisions. A useful planning tool needs both an inviting surface and a clear account of what changes when a person edits a session, moves between views, or revisits a draft.
• The calculations needed precision, and the whole experience needed to hold together as a person moved through it. The central questions were whether someone could complete the work they came to do, and if ChatGPT could understand the project structure every new session.

The Intervention

I used my Notion and Obsidian wikis as working memory, with Notion as the main source for this project.