I am building a system I already need.
That may be the most honest thing I can say about Substrate right now.
Starting over, again
Here is what working on this project actually looks like some days.
I open a conversation. I have a task. But before I can do the task, I have to reconstruct where I am: what was decided in the last session, what the coding agent is and is not authorized to do, what questions are still open, and how today's work connects to the larger intention.
That reconstruction takes time. It fills the working memory I was planning to use for the actual work. And it is never fully complete. There is always something I am carrying in my head that has not been recorded anywhere, something that will either survive the session or quietly disappear.
This is a continuity problem.
It is not unique to software development. It is the shape of any work that exceeds the span of a single sitting: any project that requires, across sessions, a kind of memory about itself.
I am building Substrate to address this problem. The difficulty is that I do not yet have Substrate to help me build it.
What the gap reveals
That difficulty is not a failure of planning. It is the founding pressure of the project.
When you are inside the absence you are trying to repair, the absence stops being abstract. You can feel exactly where it is. Every working session without Substrate demonstrates what Substrate is for.
The specific problems are these: knowing what has been decided and what remains open; knowing what authority I have granted a coding agent and what falls outside that grant; knowing how today's task relates to the larger design; knowing when a question I am about to answer has already been settled, and where.
Without a layer that holds those things, I hold them myself, imperfectly, across tools that were not designed for this, with a persistent low-level awareness that something is probably being lost.
A capable coding agent makes this more visible, not less. The agent can implement quickly. But it cannot inherit the project's living context, authority boundaries, and unresolved questions unless those things have been made explicit. I have to tell it, again and again, what has already been decided, what constraints apply, and what is out of scope.
The cognitive cost of that telling is precisely the cost Substrate is designed to carry.
This is the development lesson I did not anticipate: the need becomes legible because I am working inside it.
What Substrate is meant to do
At the level of design (not yet fully implemented, but clearly in view), Substrate is a governed layer beneath the visible work. Its intention is to hold what has been decided, mark the boundaries of AI authority explicitly, and let a human remain oriented across a developing project without reconstructing context from scratch at every session.
This is not a description of finished software. It is a design aim that has become legible through the pressure of working without it.
Earlier posts, The Missing Substrate and The Child Machine and the Child Institution, make the philosophical case for that aim in depth. This post is not making that case again. It is describing the present moment: the point at which design clarity has become implementation pressure, and where building the thing is revealing, concretely, what the thing must do.
Why this is worth saying now
There is a temptation to wait until there is something finished to show. It would be cleaner to publish a system than to publish the need for one.
But the moment when a need becomes clearly visible is itself a form of progress, and it is not always available at the start.
I did not know the precise shape of what was missing until I was inside the work long enough to feel where it was absent. That legibility took time to develop. It is not nothing.
Recovery taught me to recognize this pattern.
The point at which you can name clearly what has been lacking is not a low point. It is a kind of orientation. You know where you are. That is more than you had before. You can begin from there.
That is where this project is today.
Today I am not publishing the finished system. I am publishing the need that is teaching me how to build it.
This post is a draft. It has not been through adversarial review. It is offered as a working statement of the current moment in the project's development.