AI Companies Don't Want Us to Be Token Efficient

Right now we are all having a great time trying out these LLMs. We’re discovering what they can do and their limitations. We are in the honeymoon phase of exploration.

Next, we get efficient.

Companies like Anthropic and OpenAI have no incentive to teach us how. They profit when we depend on them, and nothing builds dependency like artifacts no human can read. A 20-page document for a trivial topic. A merge request too big to split. We are drowning in an ocean of agent output that can only be interpreted by other agents.

The more we lean on them, the less we can work without them. And as we stop reviewing their output, the quality of the work diminishes. We need yet more agents and tokens to fix the mess.

So be disciplined. Don’t generate what you won’t read. Don’t apply agents where a human level of quality is needed. LLMs are assistants — make them assist you, not do the work for you.

Becoming token efficient is an act of refusal.