A self-experiment in the cost of building software with an AI assistant

As an experiment, I tracked the full API equivalent cost of building a real software product using Claude Code, end to end.

Over one month of nights and weekends, I used Claude Code to help build a multi-device usage analytics app (this), spending approximately 3952 USD in API equivalent usage (but actually I only paid 200$ for max plan). I used Claude throughout the process for writing code, debugging, refining architecture, and shipping production features.

I tracked the API usage in detail to better understand the true resource cost of developing with AI assistance. The graph shows how usage accumulated over time.

While 3952 USD (or 200$?) is not a small number, it is significantly lower than the cost of hiring even one additional software engineer, let alone a team. AI assistance allowed me to independently deliver something that would normally require multiple roles - backend, frontend, DevOps, CLI tooling, and infrastructure.

This made me seriously reflect on how AI is changing software engineering.

With tools like Claude Code enabling solo developers to build complex systems that once needed full teams, are we entering a new paradigm? Will AI shrink teams, shift required skills, or create new roles entirely? What happens to software engineering jobs over the next five to ten years?

I’d love to hear your thoughts.