A short diagnostic. Five parts, one per layer of the system this book teaches. Tick every line that sounds like you, right now, honestly. Do not tidy up your setup in your head first. Tick what is true today.
When you are done, count the ticks in each part. The part with the most ticks is where your setup is leaking most, and that tells you which chapter to go deepest on. You do not have to read this book evenly. Read it aimed at your leak.
Part 1. Surfaces (Chapter 2)
Which door you walk in through, and whether the job is on the right one.
- You copy the contents of a file into the chat box by hand so Claude can see it.
- You are not sure whether a given job belongs in the Chat, the desktop app, or Code.
- You research across many files by pasting them in one at a time.
- You do everything in the browser Chat, whatever the task actually needs.
The fix: Match the surface to what the job needs to reach. Files on your machine belong in Cowork or Code, not in a chat box you paste into.
Minimum version: Next time you catch yourself pasting a file in, open the same task in Cowork or Code and point it at the file instead.
Part 2. Context (Chapter 3)
The second brain Claude reads before every task, so it stops forgetting you.
- You re-paste who you are, or what you are working on, into new chats.
- Claude forgets what you decided last week, or describes an old version of your business.
- Your context lives in your head, or scattered across docs and tabs, with no home Claude can read.
- You have no always-read brief, and nowhere for it to point when a task needs the deeper detail.
The fix: Give your context one home Claude reads before every task, and let a short always-read brief route to the rest.
Minimum version: Write one page, who you are and how you sound, and have Claude read it first.
Part 3. Skills (Chapter 4)
The procedures that do a specific job your way, so you stop re-teaching them.
- You re-explain how to do a recurring job every single time.
- The same job comes out a little different each time.
- You give Claude the same handful of corrections, over and over.
- None of your repeated work is written down as a reusable procedure.
The fix: Package the how once, as a skill. The corrections you keep repeating are the skill.
Minimum version: Write one skill for your most-repeated job.
Part 4. Agents (Chapter 5)
The teammates you hand a whole job to, each with its own role and context.
- One chat does your strategy, your building, and your writing, all at once.
- You re-explain the role and the context every time you switch kinds of work.
- You have no role-scoped setup, just one generic assistant for everything.
- You cannot hand over a whole job, only steer one message at a time.
The fix: Split into a few role-scoped agents and hand each whole job to the right one.
Minimum version: Set up one agent for the job you do most, with its own role and context.
Part 5. The assembled system and loops (Chapter 6 and the bonus)
The parts wired into one thing that runs, and eventually runs itself.
- Your parts are separate. Nothing is wired into a system that runs end to end.
- A multi-step job stalls halfway and you babysit every step.
- Nothing happens unless you are sitting there running it.
- No job of yours checks its own work or starts itself on a schedule.
The fix: Wire the parts into one system. Get each step working by hand, then let a multi-step job run to the finish on its own.
Minimum version: Take one multi-step job you already do well by hand and wire it to run end to end.
What your ticks tell you
Count the ticks in each part. The one with the most is your worst leak, and it maps straight to a chapter:
- Most ticks in Surfaces, go deepest on Chapter 2.
- Most ticks in Context, go deepest on Chapter 3. This is the most common answer, and the most worth fixing first.
- Most ticks in Skills, go deepest on Chapter 4.
- Most ticks in Agents, go deepest on Chapter 5.
- Most ticks in the assembled system and loops, go deepest on Chapter 6 and the bonus.
If you are starting from close to nothing and ticked boxes everywhere, do not try to fix all of it at once. There is an order, and it does not change. Start with context, because everything else reads from it. Then a first skill, because writing down one repeated procedure is the smallest next step. Then a first agent, which puts your context and your skills in the hands of a teammate. Then wiring it into something that runs. Do not build all four at once. Each layer earns the next. And do not automate chaos, or you will just get faster chaos. Get each layer working by hand first, then let it run.