The agent bill

A chatbot answers once. An agent loops, and re-sends everything it has read at every step. Move the sliders, then switch on the fixes.

The task

Steps the agent takes

Tokens it reads per step (files, logs, search results)

Length of each reply

Chatbot (one answer)

Agent, no fixes

Agent, with your fixes

Agent vs. chatbottimes the cost

Your fixes saveof the agent bill

The fixes

Ask for shorter repliesReplies 70% shorter. Shorter replies also mean less to re-send. Send easy steps to a cheaper modelHalf the steps run on a model priced at one-fifth. Shrink what it readsFiles and logs compressed by half before they're sent. Stop paying full price to re-readThe repeated part of each step billed at one-tenth, as with prompt caching.

Illustration, not a quote. AI tools bill in tokens, chunks of text about ¾ of a word each. This uses example prices of $3 per million tokens read and $15 per million written, 2,000 tokens of starting instructions, and simplified versions of each fix. Real prices, discounts and savings vary by provider and task.