AI with Kyle Daily Update 192: Codex Pulses

I am trying a new Codex setup this week and wanted to share it with you.

Dan Shipper posted about his Codex power-user setup and I’ve been playing around with it.

The basic idea is setting up “pulses”. These aren’t official. Just a way to use Codex (or Claude Code). Basically a pulse is just a Codex thread that checks one small pile and gives you a short update.

That's it.

Not one massive life-management system with 47 tabs and an inbox that somehow becomes another inbox.

Which is exactly why I think it’s actually useful….and why I want to share it with you.

Notes Apps = Idea Palliative Care

If you are anything like me, you probably drop ideas everywhere.

Notes app. Email. WhatsApp. Screenshots. Random messages to a business partner. Voice notes. Half a line in a document you never open again.

All over the place. The problem is not capture. That’s quite easy actually! We are all brilliant at capture. Too good, probably.

The problem is coming back to the pile and doing something with it.

Having lots of systems to collect, store, sort and order our ideas is basically helping us push back the hard annoying part: action. Whenever I see someone’s gorgeous Obsidian second brain I think “wow” followed immediately after by “I bet they ain’t doing shit”.

Instead the thought sits there until the heat death of the universe, next to "buy printer ink" and some weird quote you saved in 2021.

I don’t even have a printer any more…

A pulse adds the missing cleanup step which lets us ACT on our ideas. Which is why I’m a fan.

What A Pulse Actually Is

Dan's version is more company-ops heavy. He has pulse threads for proof metrics, company meetings, writing logs, inbox, router stuff, that sort of thing.

Makes sense for him.

But you do not need to copy Dan.

Please do not copy Dan. Unless you are Dan. You are not Dan. Unless you are - in which case hey love your work Dan.

Please do not copy me either, frankly. That is how people end up with beautiful productivity systems and no actual productivity.

The useful part is the shape:

A pulse is a thread that watches one specific area, then tells you what changed or what matters.

So I have 3-5 dedicated chats (“threads”) in Codex. One for each area I want help with. Into those threads I drop my half-baked thoughts and ideas. Codex tidies them up, keeps a record and then runs processes on them. The exact process depends on what the area is - it could be simply capturing, it could be prioritising, it could be running some research. Or it could trigger whole workflows for me. The pulse does so on a recurring scheduled basis, also checking in on what it needs to be doing and sending me daily reports.

For me, the starter areas are:

• Thought Capture

• Todo Capture

• Buy Capture

• Content Capture

That’s all of mine.

Thoughts in one place. Todos in one place. Things to buy in one place. Content ideas in one place.

OK you’ve made your thread, given it a name. But now what?

The pulse is only useful if it does work on the thread.

We’re going to give Codex (or whatever you are using) instructions on how to deal with the ideas and thoughts you dump in there.

Then you are going to put it on a regular timer using an automation.

For a Thought Capture pulse, I do not want a summary of every thought. Christ no. Most thoughts are rubbish. I want it to find repeated themes, compare them against my past content, spot something worth developing, and tell me one thing to do next.

So I provide the thread with instructions to take everything I throw at it and compile it nicely for me. Then to every morning send me a quick summary and ask which I want to proceed with and which I want to dump. It then spins up plans for me to take to new threads.

My ToDo capture pulse is a bit different, I want buckets:

• Today

• Later

• Needs more info

• Probably not worth doing

I’ve given Codex a set of rules on how to process everything. I won’t give you my rules because they’ll mean nothing to you - we have different priorities!

Instead tell Codex what you are building and how you want it to process the to-dos you throw at it. I find having Codex interview me super helpful here as it’ll ask incisive questions which extract what I actually want.

Mobile Is The Point

The reason this clicked now is Codex mobile.

OpenAI announced Codex inside the ChatGPT mobile app earlier this month, which means your phone can connect back to a machine where Codex is running. Laptop, Mac mini, whatever your main host is.

For me, that is the important bit. My Mac mini becomes the home base. My phone becomes the capture surface. I did a whole guide on how to set up Codex Mobile previously.

I'm out walking and have a half-formed content idea? Drop it into Thought Capture.

I remember I need to sort residency paperwork in Cyprus? Drop it into Todo Capture.

I realise I need a USB-C cable, desk light, mouse mat, or some weird house thing because I have just moved country? Drop it into Buy Capture.

I do not want to open a project management system while I am walking down the street. I do not want to tag six fields in Notion. I do not want to turn a tiny thought into admin.

I want to ramble at my phone and let the system clean it up later. I literally use voice mode and drop a quick voice note like “hey yeah so I forget I definitely need a new bedside lamp, something with a decent length cable cause like the socket is quite far, I guess LED but ideally soft lighting and colour cause like it’s for winding down at night. So, ok, cool”

AI is really good at taking our half-baked thoughts, ideas and needs and processing them into something coherent. So don’t be afraid to use voice and let AI sort the mess for you.

So…like OpenClaw?

You might reasonably say: can't OpenClaw do this?

Yeah. Sort of.

OpenClaw has a heartbeat. It can wake up on a rhythm, check what needs attention, do proactive things, improve itself, whatever. Very cool idea.

It is also broad as hell.

And the problem with broad assistants can start stumbling over themselves. Too much context. Too many jobs. Too many clever little systems colliding in the background.

If you’ve used OpenClaw for a while you may recognise this…after a while you can end up maintaining the assistant more than the assistant is helping you. No bueno. That’s exactly the opposite of the productivity gain we were hoping for!

Codex Pulses are less magical. They are single threads running simple processes.

Just to get stuff out of my head, off my plate and into my AI systems.

I do not need a digital life form thinking about my entire existence at 2am. I need one thread to look at my loose todos and say:

"Kyle, these three actually matter today. These can wait. This one needs more info. This one is probably fake productivity so drop it"

That is much simpler and genuinely useful.

To start with I’d go ahead and make To-Do pulse.

Literally make a new chat in Codex and give it something simple like:

This thread will be my ToDo Pulse. 

I will add my ToDos to this thread from now on and you will process the according to a set of rules we'll discuss.

You will also provide a daily prioritised report to me at 8AM using an automation. 

First interview me on the rules for processing my ToDos and then we'll build out the workflow and automation. 

Once down pin that thread. Connect to Codex via your mobile (in the ChatGPT app). And test sending it voice notes of your Todos to be processed. That’s all for now!

To the Task,

Kyle