Chat

I subscribe to ChatGPT Plus and using 5.2 Thinking for most of my chat sessions. According to my 2025 stats, I had 1,884 chats and 7,661 messages this year. I use this for researching things, asking home and car repair questions, as a sounding board for ideas, workout ideas, creating recipes, and answering my 8-year-old’s many questions (“what was the longest war in history”).

At work, I have access to ChatGPT Pro through work and occasionally use the 5.2 Pro for deeper thinking (e.g., I recently asked “What were the best in class ai models at the end of 2025”).

Image Generation

Nano Banana Pro has become my only tool for image generation. It’s just so good. I don’t subscribe to Gemini personally, but I find the free tier gives me enough for things I want to generate (usually silly).

Code

A year ago, I had not used a coding agent. I mostly relied on copying and pasting code into ChatGPT and auto complete in Github Copilot to write code for me.

Early in the year, I started using Cursor at home and at work. By mid year, I had almost exclusively shifted to Claude Code which I exclusively use in --dangerously-skip-permissions mode. Opus 4.5 has made that an even better experience, and I find it reliably helps me refine what I’m asking for and then provide top tier implementations. I live in Claude Code these days.

A lesson I’ve learned over the past 6 months is that Claude Code isn’t just good for code automation but also computer automation. It’s far better at using CLI tools than I will ever be, and it’s ability to combine them enables automation of lots of work. For example, I rarely use git directly any more and let Claude handle it for me (even opening pull requests).

I subscribe to the $20/month pro plan and have extra usage enabled so I can pay for tokens after hitting the Pro limits.

Writing

As Robert Ghrist told me last year, Claude has the best writing style. I often rely on it to help me refine things I’m saying.

Learning

More and more I’m using LLMs to help me learn things better from understanding research to code bases to random topics of interest.

For random topics, this often starts with a ChatGPT Thinking session with web search enabled.

For code, I have found Claude Code can do excellent breakdowns and patiently fields my questions in discovering corners I don’t understand.

NotebookLM is amazing at helping me work through a variety of sources. Generating slides based on some documents or sources is a common starting point for my understanding these days and helps me refine the questions I’m asking. I also love generating a podcast with a prompt that guides them in the direction I’m interested in so I can learn on the go.

Transcription

I’ve never been a bit voice to text person, because I mumble with a choppy cadence. However, I’m learning the modern tools do not care how annoying I am. I can stumble through my words for 15 minutes and they carefully transcribe and punctuate.

I have been experimenting with Whispr Flow for personal uses; this has largely been using their iOS keyboard for vastly superior text-to-speech that the built in Apple Option.

I have Whisper Memos connect to my iPhone action button. This is a great one-trick-pony tool that starts recording on launch and then can email the transcript somewhere; mine goes to my Drafts inbox.

I just installed Sotto on my work machine this week which provides local model voice transcription. I’m interested in using this for interacting with Slack and Claude Code, but I haven’t made much progress yet.