Claude Code for me. I live in the terminal every single day / all day. I'm concurrently running two (or more) complex sessions across multiple projects.
Key tips that work for me are:
1. Use the AI model to create/refine your prompts. Take that enhanced prompt and start a new session with that enhanced, longer prompt
2. If you're capable, stub the project you're developing against. For example, if the frontend is React/Vue, use the scaffolding tools to stub out the project footprint and that's it! No changes other than what the scaffolding gives you. Once you have that project footprint, start your AI session in that directory
3. Encourage your model to follow strict TDD/BDD practices. Tests for the requirements are written first, they'll fail, and production code to make those tests pass comes second. By doing this, you'll incrementally move towards a cleaner, testable, and verifiable codebase
4. Write your AI-driven step-by-step (literally use those words) development plan to cap each iteration (aka step in the plan) at 100k tokens. Trust me, it works. I don't care how big the context window is - 100k is where the models start dropping things
If there's any point you take away from above it's number 1. Why are we writing prompts?! Let the thing that's good at text generation and has read the entire internet write the prompts! You can always refine what it gives you.
Addressing Claude Code vs local models: Claude Code is worth it but you don't have to pay for the Max Pro Plan. Start small and see how much you use. The local models will give mixed results and I encourage using those where applicable. Think about using a local model to take frontier-generated requirements and write integration tests against those requirements (TDD). See where that gets you? Swap the models out and try again.
cmux + Claude Code / Codex with a custom agents.md file.
With cmux your agents can access the browser, spawn full shells, and you aren't locked-in to a specific vendor. It's really powerful but you need to update your instruction file so your coding harness knows it can use the cmux tools to test.
You can also download Handy to use voice so you can just give your feedback via voice into the CLI during the planning phase of a project.
Claude Code for me. I live in the terminal every single day / all day. I'm concurrently running two (or more) complex sessions across multiple projects.
Key tips that work for me are:
1. Use the AI model to create/refine your prompts. Take that enhanced prompt and start a new session with that enhanced, longer prompt
2. If you're capable, stub the project you're developing against. For example, if the frontend is React/Vue, use the scaffolding tools to stub out the project footprint and that's it! No changes other than what the scaffolding gives you. Once you have that project footprint, start your AI session in that directory
3. Encourage your model to follow strict TDD/BDD practices. Tests for the requirements are written first, they'll fail, and production code to make those tests pass comes second. By doing this, you'll incrementally move towards a cleaner, testable, and verifiable codebase
4. Write your AI-driven step-by-step (literally use those words) development plan to cap each iteration (aka step in the plan) at 100k tokens. Trust me, it works. I don't care how big the context window is - 100k is where the models start dropping things
If there's any point you take away from above it's number 1. Why are we writing prompts?! Let the thing that's good at text generation and has read the entire internet write the prompts! You can always refine what it gives you.
Addressing Claude Code vs local models: Claude Code is worth it but you don't have to pay for the Max Pro Plan. Start small and see how much you use. The local models will give mixed results and I encourage using those where applicable. Think about using a local model to take frontier-generated requirements and write integration tests against those requirements (TDD). See where that gets you? Swap the models out and try again.
I really hope this helps!
cmux + Claude Code / Codex with a custom agents.md file.
With cmux your agents can access the browser, spawn full shells, and you aren't locked-in to a specific vendor. It's really powerful but you need to update your instruction file so your coding harness knows it can use the cmux tools to test.
You can also download Handy to use voice so you can just give your feedback via voice into the CLI during the planning phase of a project.