Intriguing concept, especially because I've thought of dabbling into this area.
Can this tool solve these problems with models:
- insisting on using sed and shell redirection instead of File Edit Tools?
- trying to use npx to run commands instead of npm run that's right there in package.json
- forgetting to produce the docs that was asked for in AGENTS.md
- checking and using latest package versions instead of deciding somehow that years old versions are good enough
To my mind these are context problems, where somehow model chooses other information it has over what's in the repository, and what tools it has on call.
The other side of Lat.md, checking diffs models make among other changes, that's hard for me to grasp. I'd need to see it in action. Perhaps a coding session as a stream?
As an aside, there's now Lance data format support in DuckDB through their extension. It has Lance's vector search support available among other things:
I just noticed this, and your post, and haven't yet checked neither (sorry). I'm however doing some vector search benchmarking soon, with DuckDB's options alongside others. So your work caught my attention here.
Well, this looks very nice. I have so far avoided runbooks, preferring to use Ansible or such. Installed, and will see if I'll change my habits. Containers and such have made Ansible usage more cumbersome.
Also, I noticed there's only 60+ Atuin sponsors (at Github), so added myself. Been using Atuin for some while now. Hopefully their work is sustainable.
I wonder what is the difference between efficiency of MacBook display vs Framework laptop. Whilst CPU and GPU take considerable power they aren't usually working at 100% utilization. Display however has to be using power all the time, possibly at high brightness in daytime. MacBooks have (all?) high resolution displays which should be much power hungrier than Framework 13 IPS. Pro models use mini LED, which needs even more power.
I did ask LLM for some stats about this. According to Claude Sonnet 4 through VS Code (for what that's worth), my Macbook's display can consume same or even more power than CPU does for "office work". Yet my M1 Max 16" seems to last a good while longer than whatever it was I got from work this year. I'd like to know how those stats are produced (or are they hallucinated...). There doesn't seem to be a way to get display's power usage in M series Macs. So, you'd need to devise a testing regime for display off and display on 100% brightness to get some indication of its effect on power use.
Can this tool solve these problems with models:
- insisting on using sed and shell redirection instead of File Edit Tools?
- trying to use npx to run commands instead of npm run that's right there in package.json
- forgetting to produce the docs that was asked for in AGENTS.md
- checking and using latest package versions instead of deciding somehow that years old versions are good enough
To my mind these are context problems, where somehow model chooses other information it has over what's in the repository, and what tools it has on call.
The other side of Lat.md, checking diffs models make among other changes, that's hard for me to grasp. I'd need to see it in action. Perhaps a coding session as a stream?