I’m increasingly writing my own tools with AI rather than reaching for existing open-source libraries. For smaller applications — glue apps, the $5 appstore kind, MCP servers — I can iterate to exactly what I need faster than reading through someone else’s codebase and adapting to their decisions.
I know pretty well what I want, or can iterate fast to get there, versus being dependent on other people’s tools and their priorities.
The maintenance question
Yes, maintenance is a concern. But even that becomes more manageable when AI can help you understand and update your own code. The cost equation has shifted.
What AI coding agents don’t do yet
One thing I’ve noticed: Claude Code doesn’t encourage reuse. It doesn’t suggest “hey, there’s an existing tool that does this.” I have to manually point it at existing projects. The default behavior is to build from scratch.
Where does this lead?
I’m not sure. A few perspectives from conversations around this:
- KISS beats DRY — When code is cheap and disposable, you’re better off using basics that models have been trained on than clever abstractions
- Token economics may drive reuse back — Reimplementing everything wastes tokens; LLM-optimized library descriptions could help agents discover existing tools
- Build vs. adopt economics flip — The maintenance cost of refactoring around external dependencies often exceeds the cost of just building it yourself
- Hybrid approach — Contributing AI-generated patches to existing projects when learning their codebase from scratch is prohibitive
The pattern I keep seeing: the cost of understanding someone else’s code hasn’t changed, but the cost of writing your own has dropped dramatically. That changes which tradeoffs make sense.
Originally posted on LinkedIn