AI Usage

Following the AI manifesto's idea of sharing publicly how you use AI, this is my page to do so.

Tooling

I use LMStudio for running open-source models. Currently, I am using Claude code for development work, Copilot for quick edits, and Claude for general work.

For image generation, I use MidJourney and Grok.

I also use Grok to summarize news from X and send it once a week to my email(easier to keep track of general trends vs having to be online all the time).

How I Don't Use AI

Anything I write, I write myself. I like writing; most of the writing I do is right after I wake up and have a good cup of coffee. Most often, I will write an article after months of thinking about it in one quick session. Most articles that I leave to write on a later date I end up never finishing.

I don't see much of a point of using AI to write the articles, as my goal has always been to share things I've learned with others. AI can't talk about what I've learned better than me.

The closest to "AI" I use is grammarly to fix grammar errors, but most often I forget to even do that. It distracts me to keep Grammarly open while I type, so I turn it off and will often only remember to turn it on again after the post has been published and sent to people's inboxes.

Also most of the creative writing I do is on my remarkable tablet, it is my driver for writing because it has no distractions, I couldn't use AI even if I wanted to.

How I use AI

I talked a bit on the Tooling section but this is the deep dive.

LM Studio

I ask to explain me things when I'm offline, or when something is deeply personal and I don't want to share any data with llm providers.

Claude Code

"Set up a project locally that does X, Y, Z", "Read this repo and explain to me how X, Y, Z works", "Give me a markdown file with a plan to implement this feature", "Check this pull request's review comments and tell me what X,Y, Z Means".

In summary, asking about things I don't know and where to find them, asking to find root problems in third party libraries (like recently finding out that torchrl doesn't have strong typing) in a quick manner.

I like using it to create markdown plans as well so that I can read and edit what I think should and shouldn't be done, ask more questions and sometimes do a quick first draft that I can edit later on.

Claude

With memories, I generally write boilerplate text about my own professional background for summaries in websites that want me to, I brainstorm ideas for stories, brainstorm ideas for things that I want to do. Mostly generating text that formats my professional experiences into a desired format(like resume, bios for newspapers, etc).

I use it also to train me to think in a certain way. I created a prompt that shares all my personal character flaws and ask it to in each message understand that so that it can give me answers that help me overcome those. As Alex Hormozi once said "we need to be reminded more often than taught". I took this to heart so I use AI agents to remind me of decisions I made and want to stick through.

For example: One of my biggest character flaws is being overly self critical, so I ask the agent to give me straight answers without enabling the deep self critical ruminations. I also ask it to point me to direct actions I can take right now to solve a particular personal problem.

It's surprisingly efficient at that, you can have Earl Nightingale constantly remind you of your stated goal and if you are deviating too much from it(especially good if you are someone interested in way too many things).

With regards to writing emails, if it is for bureoucratic stuff I 100% use AI to give me a first draft before I edit it. What I do is I write without care of how it's going to be perceived, just the raw information I want to convey. Then I send to an AI and ask it to format into a professional email leaving out any unprofessional maneirism.

I hate bureoucratic stuff and if I can't type with grammar errors or without caring if I'm offending you or not, then I just use claude to type the email for me. I'm not good at not offending people or writing with perfect grammar, this is my 2nd language after all.

Newest Cowork feature

I loved using Claude's Desktop Cowork feature to organize my downloads folder. I keep a DOWNLOADS.md file on my downloads folder that teaches where files should go to, career, movies, books, documents, finances, etc. That way when the folder gets filled with too much junk I just ask claude to read the file then organize the folder. The only folder I manually manage is my Documents folder which has projects and documents I can't mess up on.

Grok

Generating funny images and creating news feeds on topics I care about(robotics, ai, biotech, etc).

Money

I use Vedeggio to manage all my money, and now there is a /chats page or an mcp that works with codex,claude,etc so I can just ask claude code "tell me how much I saved last month" or "How many subscriptions do I have" or "What should I do to save X money this month" and it will fetch my transactions and tell me.

Check out at vedeggio.com

Social Media

I've built Analog Reader to summarize youtube, hackernews, twitter(feed and replies) and my newsletters. That way I can read a daily paper with social news and a weekly paper with my saved articles. It's great, it's beautiful and it saves me hours everyday that I would have spent scrolling feeds.

Writing Books

I'm currently writing The Worst Party in the Realm. I do it for fun, on my free time, I generally type chapters on sundays at a coffeeshop next to my house. Every single word you read on the book was written by me(and for me, I really enjoy writing it, makes me laugh outloud constantly). But I do use AI to write the book, question is, how?

The website itself was vibecoded, and I have a wiki section in it, that only I can see. This is how it looks.

This wiki is auto generated, I paste the entire book on a markdown file in a project, then ask an AI agent(like Claude Code) to check latest additions and update the wiki. Sometimes it makes glaring mistakes but most often it doesn't. It keeps things consistent and the biggest thing for me is to keep track of open threads.

I don't work on this book full time, so often I'll go weeks without writing, with this wiki I get to remember things so that I don't contradict myself later on. Plus I got all my con languages there for easy check. There's also a pdf version of the wiki that I put on my Remarkable tablet(which is my personal typewriter) for easy consultation. I especially consult the languages.

I also generated the languages using AI, over several iterations before I felt the languages sounded cool enough. But I could have just as easily used one of those websites to generate fake languages in(like Vulgarlang). It would have made no difference.

Other than this I use AI to generate some images for characters based on my descriptions in the book. This is Harold for example:

I use Midjourney for those images.

Midjourney

I'm always writing stories to pass the time, I have about 60 pages in on a fantasy book I started writing when I was 14, 30 pages on another detective novel, 130 pages on a recent novel I wrote in 3 days because I thought it was funny. So I sometimes brainstorm with images how the characters might look like(before AI I used to simply google actor's images online and use those as inspiration — sometimes I still do, it's not like midjourney perfectly captures the characters inside my head).

In Summary

I use AI for doing things I don't care about doing and that I'm okay having mediocre results in. Especially bureoucratic work(I'm looking at you italian embassy).

I care about great apps, great products and great writing, and so I reserve the right to be less efficient while building/creating those.

If I don't really give a shit, then I use AI to get stuff done faster.