Ari’s Space*   est. 2026
Skip to content
ari*s space[ a. ]
← Libraryoutcomes and systems / 2026-08-04

Hold the System

Systems thinking is how I hold a problem: parts, feedback, state outside the chat, and where judgement sits.

8 parts / Hover to explore. Click to read.

Systems thinking is how I hold a problem: parts, feedback, state outside the chat, and where judgement sits.

Inputs -> work -> feedback
            |         |
        saved state   judgement
            |         |
            +-> next session

Current files + settled decisions
Open questions + decision owner
A new session needs the project's state and decision boundaries as well as its finish line. Feedback must be capable of showing that the work failed.
Context Pack (for agents)

You've agreed what done looks like. Then you open a fresh chat, and you're explaining the project all over again.

Where did the decisions go? Which files are current? Does the new session know why you rejected the previous version?

This is the other half of how I work. I want the project to make sense outside the conversation. Otherwise, every new session has to rebuild the context before it can build anything useful.

Put the finish line inside the project

Define Done as Thinking gives you a result you can assess: what must exist, what evidence counts, and what you won't accept.

You also need to know how the work will get there.

A clear finish line won't tell Claude where the latest file lives, which decision is settled, or who can approve the next step. Put those answers somewhere it can read.

I want the model to work inside a structure I've thought about. I don't want every session to invent a different structure because I forgot to give it one.

Make a small map

You can start with four questions in a text file. You don't need to buy a diagramming tool for this.

What goes in? The brief, references, constraints, previous decisions, and existing files. If you want the result to match something, make that thing available.

What has to survive this chat? Save decisions, progress, and the current state in the project. A future session should be able to find them. A transcript may contain the answer, but making someone search a whole conversation is a fairly expensive filing system.

How will you know it is working? Use a test, screenshot, review, measurement, or an actual person trying the thing. There has to be a way for the work to fail the check. Otherwise, the check isn't telling you much.

Who says yes? Decide where judgement belongs. It might be you, a client, a reviewer, or a clearly defined check. Another model can review a narrow part of the work. Confidence in its own answer is not approval.

Inputs, saved state, feedback, judgement. Keep the map small enough that you will actually use it.

Give the next session a useful starting point

Suppose you are building a page with someone. You could hand them a folder containing the agreed brief, the current version, a reference, and a short note saying what needs attention.

They can start working.

Now imagine handing them the same folder with six versions called final, no explanation, and the instruction “you know what I mean”. They are going to spend quite a lot of time finding out.

Models have the same problem. Make the current state easy to identify. Keep settled decisions close to the work. Say what is still open.

Pack the context that earns its place

Context Is King is about bringing the material that makes the answer useful.

Your references show taste. Constraints explain the limits. Previous decisions stop the same argument happening again. The workspace tells the model where things belong.

That is what I mean by making context portable. A different session, or a different model, can pick it up and understand the job.

More text isn't automatically more useful. If the result is thin, check whether a crucial input is missing before you write a cleverer prompt. Then make the useful material easy to find, rather than burying it in another wall of chat.

Work out the system before the sequence

A workflow tells you which step comes next. These questions help you decide what those steps need to work with.

What feeds the job? What gets saved? What gives you feedback? Who decides the result is ready?

Once those answers are clear, a sequence becomes much easier to design. You'll find examples under workflows, and a practical workspace approach in Build Your Workspace Once.

You can use this way of thinking without adopting one particular tool or pipeline.

Where this fits

This is the second part of Context Is King. Define Done as Thinking helps you name the result. This essay helps you arrange the work around it.

The library lets you follow the question you have now. The product-building course gives you an ordered way to practise.

Leave it ready to be picked up

Before you end the session, check that the next person can find the inputs, understand the current state, see the checks, and know who has the decision.

That next person might be you tomorrow. Make her morning a bit easier.

[ comments ]

guests welcome. members show first in the list

no comments yet. start the thread_

1 / 10