Life Admin on the Go: Week 2 of My Personal AI Assistant Experiment

If you could work fully away from your desk and laptop, what would your day look like?

Not in the usual "reply to emails from your phone" way. I mean something more substantial. What if you could be out of the house, away from your screen, in between appointments or family logistics, and still move small pieces of admin forward?

What if your laptop no longer had to be the place where every task waited for you?

That was the question I kept coming back to in Week 2 of my personal AI assistant experiment.

Last week, I wrote about using ChatGPT Pro and Codex for life admin rather than business productivity. I was interested in whether AI could become less of a tool for producing more, and more of a support layer for carrying less.

This week, the experiment became more mobile.

The question was no longer just: can AI help me sort and structure life admin?

It became: can AI help me move life admin forward when I am away from my desk?

The phone as a command centre

The most surprising part of this week was how useful it felt to use Codex from my phone.

I could be out and about, open the mobile app, and ask Codex to work on files that live on my Mac mini at home. My Mac could be physically sitting at home, switched on but screen locked, while I was somewhere else asking Codex to update documents, add items to a task list, find files, research options and save the results.

That felt like a shift.

The phone was not just a smaller screen. It was more like a command centre.

I was not necessarily doing the work on my phone. I was directing the work from my phone.

That distinction matters because most mobile productivity still feels quite constrained. You can read things, reply to things, capture notes, check a calendar, maybe make a quick edit. But larger tasks often still wait for "proper computer time."

This week, I started to see what changes when a mobile app can reach back into a workspace and do something useful while you are somewhere else.

Moving admin forward before I get back to the desk

One of the most useful patterns was giving Codex research tasks while I was out and about.

For example, I asked it to research mobile phone plans for me based on set criteria. Normally, that kind of task sits on my list for weeks because I do not want to spend an hour or two clicking through provider websites, comparing inclusions, checking discounts and trying to work out which plans are actually worth considering.

It is not hard work, exactly. It is just the kind of work that takes time, focus and a surprising amount of context switching.

This week, instead of waiting until I was back at my laptop, I could ask Codex to begin the research while I was mobile.

It searched, compared options and documented the findings. It found plans that would save me around $20 a month. It also surfaced a provider I had not come across before and highlighted plans that included grocery shopping discounts, which would be genuinely useful for me.

The decision was still mine. I still needed to review the options, check the details, look at customer reviews and decide whether any of the plans were actually viable.

But the searching, comparing and documenting had already been started.

That was the useful part.

Not that AI made the decision for me.

But that the task no longer had to wait for me to be sitting in front of my laptop.

Research tasks are a perfect test case

I have started keeping a research list for the little finding-and-comparing tasks that build up over time.

Things like:

• finding a better mobile phone plan
• comparing different coverage levels on antivirus software
• working out whether a product or service is worth switching to
• gathering options before making a small household decision
• finding the best deal for something I already know I need

These are not major strategic projects. They are the small admin tasks that sit in the background and quietly take up mental space.

They are also good examples of where AI can help without needing to take over the final decision.

I do not necessarily want AI to choose my mobile plan for me. I want it to do the boring first pass: search widely, compare options, document the findings, and give me a shortlist worth reviewing.

That frees up my thinking space for the part I actually need to do myself.

Multitasking as orchestration

The other interesting part of using Codex this week was seeing how it handled more than one task at a time.

It could split work between separate worker agents, which meant I did not have to wait for one research task to finish before starting another. It could have one worker looking at mobile plans while another worked on a different research task, then bring the findings back together.

That made me feel less like I was using a single assistant and more like I was acting as a manager or orchestrator.

I gave the task, set the criteria, and waited for the findings and comparison tables.

There was also a very human-feeling moment where one of the workers failed or timed out, and Codex noticed, stepped in and completed the task itself.

We have all been there.

The point is not that the system was perfect. It was that I could see a different way of working starting to emerge. Instead of me doing every step sequentially, I could set up the task, let the system begin the work, and come back to something structured enough to review.

That is a different kind of multitasking.

Not frantic human multitasking, where I am trying to hold five things in my head at once.

More like orchestration.

Local files, not just chat

Another useful part was that the outputs did not stay trapped inside a chat thread.

The findings were documented on my Mac and saved as local files. That meant I did not have to copy and paste everything out of a conversation later, and the information was not just sitting inside the chat history of one AI tool.

It became part of my own system.

That matters to me because I am trying to build a personal admin layer that is not just a collection of helpful conversations. I want the useful outputs to live somewhere I can find them again. I want task lists, research notes and comparison documents to be saved in a structure that makes sense to me.

It also means I could use another AI tool later if I wanted to. The information is not locked away in one place.

For a personal OS, that feels important.

The cognitive bank idea in practice

In Week 1, I started thinking about AI as a kind of cognitive bank rather than cognitive outsourcing.

I do not want AI to replace my thinking. I want it to help me store, structure and retrieve what matters.

This week made that idea feel more practical.

When I was out of the house, I could deposit a task into the system. Research this. Compare these. Add this to my list. Find this document. Save the findings. Then I could come back later and review the result when I had more space.

That is useful because so much life admin gets delayed not because it is impossible, but because it needs the right combination of time, attention and context.

If I am out for the day, in transit, waiting somewhere, or moving between family logistics, I might remember the task, but not have the capacity to do it properly.

Previously, that meant the task stayed in my head or sat on a list waiting for some future desk session.

Now, I can move it one step forward.

That is the real shift.

What still needs human review

There are still manual parts, and that is a good thing.

I still need to review the recommendations. I still need to check details. I still need to read customer reviews. I still need to decide whether the findings are actually relevant, accurate and suitable for my situation.

This is not a replacement for judgment.

It is decision support.

That distinction is important because personal AI can feel more sensitive than business AI. In a work context, the value is often tied to speed, output or workflow efficiency. In personal admin, the value is more about reducing mental clutter, remembering the right things, and making it easier to act when the time is right.

So I am still keeping clear boundaries.

I am using this for non-confidential life admin and research tasks. I am not handing over private information or expecting AI to make decisions on my behalf.

At this stage, the role of the system is to help me organise, research, compare and prepare.

The responsibility still sits with me.

What I learned in Week 2

My Week 2 reflection is that mobile AI feels different when it can do more than answer a question.

When it can connect to a workspace, update files, compare options and leave something useful behind, the phone becomes less of a distraction device and more of a small command centre.

That does not mean every task should happen on mobile. I still like working at a desk. I still like notebooks. I still like having proper thinking time.

But I do not want every piece of life admin to wait until I am sitting in front of a laptop.

That is what felt different this week.

I could be mobile and still move things forward. I could capture the task when it came to mind, send it somewhere useful, and come back to something more structured later.

The value was not that AI did everything for me.

The value was that it reduced the bottleneck.

Week 2 takeaway

I am still testing this carefully, and I am still learning where the boundaries should sit. But this week made me more interested in mobile AI as part of a personal assistant layer.

Not because I want to work all the time.

The opposite, really.

I want fewer things waiting for the mythical moment when I am back at my desk, fully focused, with enough time to deal with them properly.

Sometimes life admin just needs to move one step forward.

This week, Codex on mobile helped with that.

And that feels like a meaningful shift: not doing everything from anywhere, but being able to move small pieces of admin forward when life is happening somewhere else.