Matt DuCharme @me.mattducharme.dev · 08/10/2026Because it's all lua as well, I can just have the llm write more lua for plugins or tools or whatever. Or have the agent write a lua tool for itself or its children on the fly. Whatever. 000
Matt DuCharme @me.mattducharme.dev · 07/10/2026My favorite thing is to tell it to use buffers for: checklists, diffs for plans, plans, communication with other agents, graphs (yes this is possible with lua) 100
Matt DuCharme @me.mattducharme.dev · 07/10/2026Is it novel? Not really, it just integrates weirdly into neovim. It has a malleable chat transcript (rip your cache) which is way more useful than anything I've experienced before, and all my vim motions work. Search, grep, find/replace, lua, oil.nvim, whatever. 100
Matt DuCharme @me.mattducharme.dev · 07/10/2026Been my daily driver for a bit and not that much different than most harnesses but here's mine github.com/ducharmemp/s... I really really hated the fact that other harnesses reinvent garbage UI when all I need is a buffer in neovim. Even for multiple agents. Everything else is information slop.github.comGitHub - ducharmemp/strapsContribute to ducharmemp/straps development by creating an account on GitHub. 130
Reposted by Matt DuCharmeAaron Allen @aaronmallen.dev · 30/09/2026Today I have published the first issue of "Hanami, why?". This is a new series I'm starting where we will dive in together to learn about #hanami. aaronmallen.me/writing/hanami-why-i… #ruby #hanakai 02114
Matt DuCharme @me.mattducharme.dev · 14/09/2026Why on earth would OpenAI call for regulatory capture when they've already shown their unreleased model is much smarter than Astra? Something doesn't add up with that possible explanation 000
Matt DuCharme @me.mattducharme.dev · 12/09/2026Try it out in your browser (QEMU WASM VM in your tab) trynix.dev?path=/nix/st...trynix.devtrynix — boot anything nixpkgs ever shipped, in your browserPick any package from 13 years of nixpkgs history and boot it in an x86_64 Linux VM in your tab. No server anywhere. 000
Matt DuCharme @me.mattducharme.dev · 12/09/2026Everyone else is making these to automate project setup on worktree/workspace init (probably because it's so trivial) but here's mine so far github.com/ducharmemp/f... Currently #jj only Definitely vibe coded but I'm pretty happy with how this turned out with elixir.github.comGitHub - ducharmemp/foyerContribute to ducharmemp/foyer development by creating an account on GitHub. 100
Matt DuCharme @me.mattducharme.dev · 12/09/2026Quick reflection on #ElixirConf 2026 mattducharme.dev/posts/elixir...mattducharme.devElixir Conf 2026Massive congratulations to the elixir conf folks this year, things were fantastic! I do wish a few more folks had leveraged LLMs a bit more tastefully when generating their talks, but to each their ow... 110
Matt DuCharme @me.mattducharme.dev · 12/09/2026Has he not heard of our Lord and Savior the Church of Nix? 010
Matt DuCharme @me.mattducharme.dev · 12/09/2026Obviously because the best distros come with the best dotfiles by default (this is sarcasm please don't flame me) 111
Matt DuCharme @me.mattducharme.dev · 04/09/2026I was hoping the blade runner AI ad targeting would at least come with the blade runner architecture 000
Matt DuCharme @me.mattducharme.dev · 25/08/2026I wonder if there's a venn diagram of how many folks call LLMs "he" and have unironically used the attack helicopter meme in the past 🤔 100
Matt DuCharme @me.mattducharme.dev · 25/08/2026I wonder if there's a tracker out there for the number of times "he" appears as a pronoun when describing LLMs of any kind by software devs. There should be to see if we're accelerating towards some sort of mass psychosis of what these models are 100
Matt DuCharme @me.mattducharme.dev · 17/08/2026It's still hyperpersonalized to my way of working so not (yet) something I'm looking at open sourcing, but I think the general idea of writing your own X in the AI era is way more powerful than a lot of folks are realizing. Why limit yourself to Anthropic or OpenAI to give you your tools? 111
Matt DuCharme @me.mattducharme.dev · 17/08/2026So plans, todos, data viz, communication between agents, whatever, it's all just another text buffer. 110
Matt DuCharme @me.mattducharme.dev · 17/08/2026Generally leaning into things that vim provides out of the box, or omitting useless features from other tools that wind up getting in the way. Also I've found a few cases where I can ask the tool to display something for me and because it's neovim, it can construct lua with splits/floats. 110
Matt DuCharme @me.mattducharme.dev · 14/08/2026"I wrote" here is too strong considering I delegated to the agents to write the harness for themselves 000
Matt DuCharme @me.mattducharme.dev · 14/08/2026I'm behind the curve but I don't care-- using Claude code instead of writing your own harness is something I shake my head at now. I recently wrote my own agent harness with neovim and I can't possibly imagine why I'd go use someone else's harness. 210
Matt DuCharme @me.mattducharme.dev · 02/08/2026The more I read into what folks at anthropic and openai are doing (ex: "loop engineering") the more it reminds me of the heydays of micro services and react-- over application of a useful thing to the point of absurdity. "Of course we have 1000 services, we're a team of 10 engineers! So much speed!" 000
Matt DuCharme @me.mattducharme.dev · 09/07/2026mattducharme.dev/posts/verbal... Musings on LLMs and productivity, as if the world needed more content in this space.mattducharme.devVerbal SpaceI had a discussion today about LLMs that helped me focus my attention on something I’ve been feeling for a while. I’ve been feeling fairly productive in vertical slices with LLMs when I’m having agent... 000
Matt DuCharme @me.mattducharme.dev · 09/07/2026github.com/ducharmemp/b... generated a bench script to comp current implementation with pgcat and pgbouncer using pgbench. Initial results here: github.com/ducharmemp/b... I wouldn't be very surprised if it's something stupid like a bad allocation somewhere to be honest.github.com 100
Matt DuCharme @me.mattducharme.dev · 07/07/2026I was mainly using pgbench for this with a trace/ptrace and some gut intuition to guide me, but I can whip up a script to automate this later 100
Matt DuCharme @me.mattducharme.dev · 07/07/2026This was interesting, like a reverse psychology approach to prodding the LLM. It did try and tell me to get rid of the IterReader I have in the code (I wanted a non-owning view, not sure if this is the right move) and spotted a couple more areas of improvement. Nice! 010
Matt DuCharme @me.mattducharme.dev · 06/07/2026I think I got a hit, nixos might have compiled the package on a machine with fancy instructions that my pitiful laptop can't utilize. Crash is an illegal instruction, so I just need to avoid the package cache here. Easy enough, might need to put up a PR for this on nixpkgs 000
Matt DuCharme @me.mattducharme.dev · 06/07/2026NixOS so the suffering is entirely self inflicted here. I haven't had a chance to run a tract on it yet but I'm sure that it's something trivial like a missing lib in the package. 100
Matt DuCharme @me.mattducharme.dev · 05/07/2026Thanks for the heads up, I'll rename it to something else. Probably "barnyard". 000
Matt DuCharme @me.mattducharme.dev · 05/07/2026Had a bit of fun with @ponylang.bsky.social recently. Working in a different actor paradigm than what I'm used to really stressed out the gray matter. github.com/ducharmemp/s... pretty neat little project.github.comGitHub - ducharmemp/stableContribute to ducharmemp/stable development by creating an account on GitHub. 430