The agent ecosystem course
  • 0Before you start
  • 1Claude Code
  • 2The mental model
  • 3Map your ecosystem
  • 4The brief
  • 5The connectors
  • 6The runbooks
  • 7Working hours
  • 8The ecosystem
  • 9The backup
  • 10Sharing skills

The stages

  • 0Before you start
  • 1Claude Code
    • 1.01What Claude Code is
    • 1.02What a token is
    • 1.03Which Claude
    • 1.04How hard it thinks, the effort dial
    • 1.05What you have used, and how to use less
    • 1.06Chats, and starting fresh
    • 1.07Two chats side by side
    • 1.08The modes, and the two you use
  • 2The mental model
  • 3Map your ecosystem
  • 4The brief
  • 5The connectors
  • 6The runbooks
  • 7Working hours
  • 8The ecosystem
  • 9The backup
  • 10Sharing skills
BACK TO THE COURSE

Stage one, session one

Claude Code

Your bearings. What it is and what it can do, which Claude, what you have used, chats, and when it asks.

1.05

What you have used, and how to use less

/usage shows how much of your allowance you have used, as two bars, one for the five-hour window you are in and one for the week, with when each resets.

Under that it says what has been eating it.

Session (5 hours)37%

resets in 2 hours

Weekly (7 days)8%

resets in 6 days

Weekly, Fable6%

resets in 6 days

What is contributing, over the last day

  • 71%of the usage was at more than 150k context. Longer chats are more expensive even when cached. Compact mid-task, and start fresh when switching to a new task.
  • 49%of the usage came from chats that spawned subagents. Each subagent runs its own requests. Be deliberate about spawning them, and give the simpler ones a cheaper model.
  • 11%of the usage came from one kind of subagent, general-purpose. If one runs often, give it a cheaper model or tighten its brief.

Independent readings, not a breakdown, and approximate, from the chats on this machine only.

What it went on

  • Skills/update-config1%
  • Subagentsgeneral-purpose11%
  • ConnectorsComposio

What /usage showed on my own machine on a Monday afternoon, on the Max plan.

It is indicative, and a useful thing to look at now and again.

You have already set the model and the effort, so this is just a feel for how token hungry what you are doing is.

The way people run out does not tend to be steady daily work.

It tends to be one very long chat, because every message you send carries the whole chat with it, so the same question costs more the longer the chat gets.

Three habits keep it down:

  • Shorter chats. Start fresh when the task changes, because it costs less and the answers are better, which the next part explains.
  • Plain files. A page of plain text or markdown costs a fraction of the same page as a PDF or a Word document, because the formatting takes tokens too, which is why the files on this course are plain.
  • Only the context it needs. Name the file you want it to read rather than having it read the whole folder, and a file it reads once from the folder is cheaper than the same text pasted into every chat.

If you do hit the limit, you come back when it resets, and in the meantime every file is sitting on your own computer for you to read and browse as you please.

The agent ecosystem course

What you have used, and how to use less

This is part 1.05 of the course. Start it free and it opens straight away, with no password.

A few quick questions come next. You will also get my Thursday newsletter, and you can unsubscribe from it without losing the course. How your details are kept is in the privacy policy.

Already have access? .