Focus: LLM tools exploration (2026W43)

A closed laptop beside a Mac mini and a monitor showing agent tasks, with Mount Asama visible through the window at dusk.

(For the record, this was initially planned to be my week 42 focus, but a national holiday, two days at the Kaigi on Rails conference and being sick contributed to a short and unproductive week…)

Following my One Focus a Week post from last week, here’s my quick post describing this week’s focus (week 43 of 2026).

I’ve been mostly using OpenClaw and Pi in the past 6 months, backed at times by an Anthropic model, at others by an OpenAI one. I’ve recently gotten back to using the ChatGPT and Codex apps, on my phone and on my laptop, and realized there are powerful features I’m unfamiliar with, and that I would like to use more efficiently.

There are other tools that have caught my attention too, such as T3 Code and Solo, to which I haven’t been able to dedicate any time so far.

This week, my plan is to set up these tools, explore their features, and figure out what place they may have in future projects. The best opportunity to use them will be when I have concrete projects I want to work on, but spending this preliminary time exploring will allow me to understand better what’s available.

One specific set of features I’ll spend time exploring is everything around remote tasks: I don’t want to ever have to keep my laptop open so as not to stop an agent’s ongoing task. I’d rather have my Mac Mini always on at home, available for agents to use and run their work there.

I have several goals I hope to achieve this week:

  • Set up my always-on Mac Mini M4 as a remote platform for me to run remote agent tasks.
  • Set up an isolated VM for a new, clean and up-to-date instance of OpenClaw. There still is value in this open-source AI assistant, but my setup has kinda been rotting away on an old machine and lately has been feeling less useful than the ChatGPT apps.
  • Set up and play for a bit with these applications:
    • T3 Code
    • Solo
    • Claude Code as well, since I haven’t used it in quite a while
    • Pi
  • Experiment with those tools and their remote capabilities
  • Save the setup in a set of configuration files, which will allow me to recover that setup easily in the future.
  • Try out local models.

I’ll consider this week a success if by the end of it, my Mac Mini is set up, all tools are running, and I’ve spent some good time with each, identifying some things that I like or dislike about each. Things will need to be in a state that’ll allow me to tackle real work in the following weeks. I should be able to pick a tool, start a project, and get agents started on the work.