A better way to use Dots

I have been seeing quite a few discussions around Dots and how people are using them. One thing that keeps coming up is the comparison with Codex. Now, I get why people make that comparison, but using my Dot as another place to ask for code is not what I find most useful about it. I already have Codex for that.

What I want help with is keeping track of the work going on in those Codex threads. Which task has finished? Which one is waiting for an answer? Did that clarification get passed on? Having multiple agents working on things sounds great, till I realise I am still the one going from thread to thread, reading updates and typing follow-ups. All this automation, and I have given myself a management job.

So, I think of my Dot as a personal assistant, or a floor manager. Codex can work on the individual tasks while my Dot helps me keep track of them and pass along what needs to happen next. This kind of delegation and follow-up is also covered in the Dots documentation. For me, having one conversation through which I can manage that work is where it gets interesting.

A fairly ordinary coding session

Recently, I was working through some error handling in a couple of API routes. There was some confusion about which route we were talking about, and whether both GET and PUT needed changing. GET did not seem to have the same issue, whereas PUT was throwing inside a try block.

I explained this to my Dot and had it pass the clarification to the coding task. Focus on PUT, and leave GET alone unless there was a related reason to change it. I could talk through what I meant in the conversation I was already having, and have that correction reach the task doing the work.

Somewhere in between, the conversation went off to something completely unrelated. By the time I came back to the code, I was asking what was effectively, “What were we doing? Is that edit still not completed?” My Dot came back with a report on the edits. Later, I continued a rebase, ran into conflicts and asked it to get Codex to help with those next.

I will spend quite some time connected to it

Checking on an edit, correcting a misunderstanding and moving on to the next problem are all fairly ordinary parts of working on code. But if I have to visit every thread and do each of those myself, I still have quite a bit to keep in my head. Being able to ask my Dot where we got to is useful.

Give it something useful to manage

Of course, this needs some context. If I say “go check that thing” when there are several things going on, I can’t be too surprised if we end up talking about different things. The confusion around GET and PUT is a fairly good example of why the details matter.

For a task like that, I could give an instruction along these lines (this is an example, not a word-for-word transcript):

In the existing route cleanup task, ask Codex to focus on the PUT validation issue. GET does not seem to have the same problem, so leave it alone unless there is a related reason to change it. Check what changed and whether anything needs my input. Leave the changes for me to review before committing.

That gives it a task to follow, a distinction to preserve and something useful to report back. I don’t need every line of the coding conversation repeated to me, but I do need to know what changed and what was checked.

It also helps to say where the work is happening. For local Codex work, the relevant computer needs to be connected and available. A Dot does not automatically have access to every conversation in every app, either. If I want a decision carried into a task, I should make sure it is part of the handoff rather than assume it somehow got there.

And yes, I still need to review the work. A report that an edit is finished gives me something to check. It doesn’t make the code correct by itself. I want help keeping track of the tasks, and I still want to look at what they produced.

It does not have to be about code

This post itself started in a conversation with my Dot. I was scrolling through X after asking my Dot to make some changes when I came across the same complaint for the umpteenth time, that Dots are useless for coding. I asked the Dot what it thought about me writing a blog post on that and then I asked it to scaffold the markdown file and find an image on unsplash that would work as a hero image. It created a new thread to work on that and I started drafting the post. When it was ready, I asked it to copy the existing draft into the scaffold and it did that before I resumed writing.

It is the same idea as the coding tasks. I can bring something up, have the work continue in an appropriate place and come back to it when I need to. The thing being worked on happens to be a blog post this time.

If you already use Codex, I would start by giving your Dot some of the follow-up you currently do yourself. Tell it which task needs attention, what you want done and what you want to hear back. See how that works before trying to hand over everything at once. I would much rather spend my time looking at the result than trying to remember which thread I left it in.