Building AI agent products still means many manual decisions

The writer says they built two AI agent products. They say the core code was only part of the work, and tools like Claude Code or Codex helped with that when guided well. The harder parts they describe include login, deployment, API key handling, provider limits, and sandbox support for user requests.

Key points

  • The post is based on the writer’s experience building two AI agent products.
  • They say AI coding tools helped with core logic when given proper guidance.
  • They describe login choices as a source of extra work and cost.
  • They mention cloud deployment choices such as AWS or Vercel.
  • They raise questions about shared API key use, provider limits, and per-user sandbox support.

Quick term guide

deployment
The process of putting software changes into a running system.
provider
A company or service that supplies an AI model, such as OpenAI or Anthropic.
model calls
Requests sent to an AI model to get an answer or action.
model call
One request sent to an AI model to get an answer.
permissions
Settings that define what files or actions a system or user is allowed to access.
permission
The allowed range of actions a person or system can take.
AI coding tools
Programs like Claude, Cursor, or ChatGPT that write code for you when you describe what you want in plain language.
AI coding tool
Software that uses AI to help write, edit, or explain code.
Read original