No more posts about Claude Code. Please. There's no point in building workflows or skills around it.
Developers are too dependent on this one tool, when so many other possibilities exist. I've seen developers share scripts that delegate to Claude Code. I've seen tutorial websites targeting Claude Code. And embarrassingly, I've seen developers brag about running out of tokens. It gives the impression of a domesticated animal.
But you don't need Claude Code. You can use other agent tools, or you can build your own. Anthropic's models might be "good at X" and their tools might have optimized tool calls, but that doesn't mean Claude Code is irreplaceable. You can be lazy and use the Claude Code, that's fine, but then don't pretend you're building on a solid foundation.
We should know better than to build our skills and projects on top of proprietary, VC-backed tools. How long will Claude Code remain a product? How long will it keep it's current pricing? How long until their prompting breaks your workflow? How long until Anthropic swaps out the model to save costs?
I would rather see developers focus on building coding agents of their own. Instead of asking questions about Claude Code, consider investigating the questions here:
- What are the best ways to prompt for tool calls with different models?
- What is an appropriate number of tokens for when an agent edits a single file?
- Where can we use smaller, faster models in our workflows?
- What independent scripts can we build that will perform better than a wrapped Claude?
- What open weight models can we use so that our workflows aren't tied to VC-backed companies?