Field notes · L-14Weekly Tinkering #5
Weekly Tinkering by Junod.dev for Feb 11th
Filed underMaking things
The one where we talk about AI eating SaaS
SaaS is how the world consumes software, there’s a thing running somewhere and we pop in and out as we need the value produced by that system. Some team, in the beforetime, envisioned a problem, maybe raised some funds, and built a software package to solve your problem.
Most of the time it works pretty well, but sometimes you hit edges. The widget can’t change the data just how I want it, or the knob doesn’t turn quite enough to get exactly what I need. Software, in broad strokes, is designed to solve everyone’s problems, not yours. Your problem looks a lot like everyones, and close enough that we’re all willing to deal with it.
A couple posts this week really got me thinking about that.
Things to read/watch
- Geoffrey Huntley writes about the future belonging to idea guys that do. This whole post is a kick in the ass to build things with AI. He writes about accelorating the time to “Oh Fuck” moments. This is a short piece that everyone should read. I’ve been through a number of these “Oh Fuck” moments with AI, and it’s hard to describe to others that haven’t yet.
- Lee Robinson writes about personal software, which is a newish concept to me, but it makes perfect sense. We’re approaching a time when anyone can craft software to solve their problem. Complexity in software appears as it grows, and many problems, at the scale of 1 to a few users, can be solved with very simple software. The idea that people will create small apps to solve their personal problems is actually quite realistic. Is there a huge opportunity for the myspace of personal apps?
Things to try
- MCPs continue to rule much of my workflows. Most of the time I’m using either the Claude app, Goose or Cline, all of them work really well with MCPs. I’m constantly on the hunt for new MCP clients. Sage is a recently released MacOS Claude client with full support for MCPs, and it won a recent MCP hackathon.
- LLMs are really good at logic as long as you can provide the best context. Tools like Cline and Goose are good at dynamically creating context over a folder or git repo. If you have more context than a simple folder, but less than a full fledged rag solution, check out rag-retreiver. Also, even if you don’t need this tool, looking through how it gets integrated into AI pair programmers is super valuable.
Things I’m tinkering with
This week I’ve been working through a long video that’s been making the rounds. Shared by one of my coworkers, it’s a 3.5 hour video that dives deep into how LLM’s work. About 15 minutes into it I sent it to my son and told him he had to make the time to learn this. This video is a must see for everyone, and it falls into a thing I’m tinkering with because of it’s overall legnth and quality. If you watch 1 movie on LLM’s this year, make it this one!
Deep Dive into LLMs like ChatGPT
I’m also going to be going through Huggingface’s new Agent course, the first week is available. If you’re on the discord say hi to me, I’m _tranqy.
I hope you enjoyed this week’s post. Please let me know if you have any feedback, and thank you for making it to the end.