---
title: "Learnings from init to deploy"
description: "An agent was used to build a transcription app for a team from initialization through deployment with Stripe Projects, OpenRouter, and Vercel. The app sends video-generated subtitles through an OpenRouter agent with a detailed system prompt and Stripe documentation tool calls, producing a formatted transcription whose speaker and service names are corrected. The author recommends treating agent-generated plans like design documents, verifying provider behavior, and using streaming for long transcripts because Vercel serverless limits made non-streaming responses prone to 504s for a roughly 1,500-line transcript. Testing led to removing an unreliable free-model tier, defaulting to openrouter/auto, and choosing Claude Opus 4.5-thinking over Claude Sonnet 4.5; shared agent files were committed, and a separate review agent was used before human review."
---

# Learnings from init to deploy

[Stripe](https://yomu.fyi/company/stripe) · Anna Spysz · May 4, 2026

**Type:** Tutorial

## Summary

An agent was used to build a transcription app for a team from initialization through deployment with Stripe Projects, OpenRouter, and Vercel. The app sends video-generated subtitles through an OpenRouter agent with a detailed system prompt and Stripe documentation tool calls, producing a formatted transcription whose speaker and service names are corrected. The author recommends treating agent-generated plans like design documents, verifying provider behavior, and using streaming for long transcripts because Vercel serverless limits made non-streaming responses prone to 504s for a roughly 1,500-line transcript. Testing led to removing an unreliable free-model tier, defaulting to openrouter/auto, and choosing Claude Opus 4.5-thinking over Claude Sonnet 4.5; shared agent files were committed, and a separate review agent was used before human review.

## Context

The project involved processing long, auto-generated subtitle transcripts with an AI agent while preserving the script and correctly spelling and capitalizing speakers and service names. The author also needed to account for provider behavior, model quality, shared agent knowledge, and review practices when using agents across the development workflow.

## Approach / What changed

The workflow used an upfront agent-generated plan covering goals, data flow, architecture, and tasks, followed by research and human verification. The implementation adopted streaming for long transcripts on Vercel, revised model selection to try free models before paid ones, then removed the unreliable free tier and defaulted to openrouter/auto. Agent-specific files were committed for shared rules and skills, and a separate PR-reviewer agent checked the completed changes before human review.

## Takeaways

- Streaming was implemented for roughly 1,500-line transcripts after non-streaming Vercel serverless responses were judged likely to hit 10-second Hobby or 60-second Pro limits and return a 504.
- Repeated free-model outputs condensed a 700-line script into about 100 lines, so the free tier was removed; the app defaulted to openrouter/auto, and reasoning models were found more accurate despite being slower.
- Committing agent files such as AGENTS.md and files in the agents directory lets teammates share skills and rules, while a dedicated review agent can identify issues before the author performs the final review.

**Tags:** [Architecture](https://yomu.fyi/topic/architecture), [Code Review](https://yomu.fyi/topic/code-review), [Deployment](https://yomu.fyi/topic/deployment), [Developer Experience](https://yomu.fyi/topic/developer-experience), [LLMs](https://yomu.fyi/topic/llm)

- Source: [Stripe](https://stripe.dev/blog/learnings-from-building-with-agents-stripe-projects)
- Source URL: https://stripe.dev/blog/learnings-from-building-with-agents-stripe-projects
- Ingested by Yomu: 2026-08-29T00:00:50.818Z

[Read original post](https://stripe.dev/blog/learnings-from-building-with-agents-stripe-projects)