AI Everything
Tired of the “AI” boom? Me too, but I love using it so much. Just with a different approach than “it will suppress us and we as developers are gonna be jobless”.
Make it yours!
Take some time and set up your workflow for agentic cooperation. I mean, if you set constraints for everything, then all would be smooth and shiny.
A few examples of what may go wrong:
- Anthropic itself notes that “Excessive token consumption from tools makes agents less efficient”. What does it mean? Well, too much of anything is not good. MCPs can pollute a lot of your precious context. Be savvy about that. If you want to see how much your Claude context is polluted
- Code examples - what do you like and what you do not like. Its simple, as you were instructing some junior dev
- Create commands, agents and leverage everything you need. Automate everything you can, so you just know what works and what does not. And put on repeat
- Always present a plan before implementing - this is my opinion, but I see it quite useful. Agent itself does now know what vague tasks like “Add button to submit a form” means and it will try to do most probable thing - just a boring button
Work in smaller chunks
It's better to break down complex things into smaller chunks. Not only because of context, but also due to testability and maintainability. Oh, and you can shuffle between multiple tasks in the meantime. It’s as simple as working with work-trees, even without a complex IDE developed especially for complex agentic workflows
Keep learning
And remember to always learn and improve new things. Especially in this environment, there is a new thing almost every week.