# there's a reason the computer was called Mother By [40pi](https://paragraph.com/@donniecast) · 2026-03-16 --- Oh goodness gracious. This is gonna be a short one, but it's a thing that's on my mind, and I don't want things on my mind right now. I want things out of my mind and processed. Hit a bit of vibe coding snag this morning. Gonna actually start forking over USDC because having a machine build software for me, it turns out is extremely I don't want to say addictive. But, uh, sticky? extremely sticky. 
We like sticky. The issue, though, is when I'm working with the agent and things are going well I don't keep track of project state. Everything feels fine. Until there's an outage. But the point is we have downtime. So now we're in quote unquote offline mode. 
And that's the moment where the lack of project tracking — or context tracking — really slaps you in the belly like a wet fuckin' fish. I need the means to say to my synthetic code jockey _here's where we left off_. That work of here's where the project is at... here's sort of the current branch that we're working on and the particular problem that we're trying to solve.
All that sort of status stuff. This is the case for a local agent. Or at least a different agent, right? Because the coding agent goes down for $REASONS and I need some mechanism for capturing the relevant information about where the project sits so that you can pick the ball up when the coding agent comes back online. That job, of taking the current state as it exists in short-term memory in my head and putting it down on quote unquote paper, is obviously a job for an agent. 
One that should not ever go down, which is to say, should be locally run on whatever metal I have access to. So that it can maintain situation reports and enable us to pick up where we leave off. So let's lay out the pieces on the board. There's me, as defined by my corporeal form. Then there's entities number, let's say, two through n. Entity two is this agent running on local metal. Which is to say an agent that incurs zero cost.
And we'll call this agent Alice. And then there's entity number two, which is Bob, which is an agent running on top of a paid model.
An external model. And there are tasks where spending dollars on them is worthwhile those that aren't. The challenge then is
I have to be able to direct work appropriately. I need Alice to keep track of progress against the overall plan (to say nothing of all the other plans). So when we've finished working with Bob for the day, and a particular piece of work has been completed, we want to make Alice aware of that and have Alice update the planning documents. So that means calling out to GitHub, right? That's a thing that shouldn't require spending dollars on. Whereas the heavier lifting of writing code and refactoring and blah, blah, blah, does feel worth the spend. And then there are instances where you would want to be able to instruct Alice, or we should say, to give Alice direction. to say: _go tell Bob to start working on this and let me know how it goes._ That level of separation is the next challenge. 
We're away from a keyboard and we want to be able to keep Bob working efficiently — to say nothing of Charlize and Daryl and...
Perhaps we can communicate directly to these agents from a telephone. But it's more likely the case where it's easier, from an interface perspective, to communicate to Alice and have Alice turn around and communicate to the team. Dialing in this dynamic is gonna be necessary for getting the most work done in the shortest length of time.
But now is not that time. Now it's time to get Bob back to work. --- *Originally published on [40pi](https://paragraph.com/@donniecast/computer-called-mother)*