How to use Claude to plan a transcript coaching app
Two engineers used Claude Code and other development tools to build Unpitched in a five-week co-founder trial, saving planning documents as implementation context.
news.ycombinator.com
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Help plan [transcript coaching application] for [target users]. Research and brainstorm before coding, define a walking skeleton, and document error-handling decisions in [planning file]. Use the agreed document as context when implementing [feature].
Reach people exploring AI agents
See placements →What happened
The authors report building a working mini-product in five weeks and learning how they approached AI development and collaboration together.
Reported by gkk on Hacker NewsHow the agent did it
- Structure project planning around walking skeleton and tracer bullet concepts.
- Plan error handling with the AI before coding.
- Save the error-handling plan as a document.
Where a person stepped in
The engineers intentionally guided the AI and documented plans; the post does not describe specific code-review or test approval steps.
A person approved each step before the agent continued.
Two experienced engineers built Unpitched during a five-week trial of working together as co-founders. The product analyzes customer interview transcripts, looking for moments when founders pitch instead of learning, and provides personalized advice based on principles from The Mom Test. The authors report that AI wrote approximately ninety percent of its twenty-four thousand lines of code.
Development used Cursor, Claude Code, V0, and briefly Grok. The application stack included Next.js, Supabase, Trigger.dev, and GPT-4.1 through the Vercel AI SDK. The authors emphasize intentional management of the assistants rather than assuming the tools share an experienced engineer's implicit priorities.
They structured planning with walking skeleton and tracer bullet concepts. Their working pattern involved thinking, researching, brainstorming, and planning before implementation. One concrete example was planning error handling with AI, saving that plan as a document, and then providing it as implementation context to keep behavior consistent across the codebase.
The trial produced a working mini-product and helped the pair learn about each other's working habits. The post does not describe a detailed testing or code-review process. Who it's for: experienced developers testing a collaboration while using AI to implement a tightly scoped product with explicit planning context.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · gkk
Frequently asked questions
How did they keep implementation consistent?
They planned error handling with AI, saved it as a document, and supplied that document as context for implementation.
What did the trial produce?
The authors report a working mini-product in five weeks and insights into their working relationship and use of AI.
More details
Remaining steps
- Use that document as context for implementation.
About this account
The authors say AI wrote approximately ninety percent of the code, with the engineers directing scope, planning, and implementation context.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=44605541. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: gkk on Hacker News
Source published 18 Jul 2025
Source checked 14 Oct 2026
Tools and topics
Claude Code · Cursor · V0 · Next.js · Supabase · Trigger.dev · Vercel AI SDK
AI development
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.