LLM (the model) is not the agent (ClaudeCode) that uses LLMs. LLMs improve slowly, but the agents are where the real value is produced: when should it write te

LLM (the model) is not the agent (ClaudeCode) that uses LLMs. LLMs improve slowly, but the agents are where the real value is produced: when should it write tests, when should it try to compile, how to move fwd from a compile error, can it click on your web app to test its own work, etc. etc.