Nathan @natfu.be · 12hI think these days models are so good you don't need many skills. If you have specific workflows or ways of programming a project that's fine and put in there but otherwise they do fine. What I usually do is a couple "researcher" sub-agents go through the main deps' docs and the language's docs. 000
Nathan @natfu.be · 03/10/2026You can use it to challenge you as well. I had some issues at work with a platform and limited access limitations. I wrote some scripts and tested some things in the UI as well and wrote down my assumptions. Then asked an LLM to poke holes at my setup and it managed to fix most of my problems. 010
Nathan @natfu.be · 30/09/2026You want www.manning.com/books/grokki...; good introduction is www.youtube.com/watch?v=bpBH... and then genuinely RTFM as a reference docs.python.org/3/library/as..., the docs are very complete imo.manning.comGrokking Concurrency - Kirill BobrovThis easy-to-read, hands-on guide demystifies concurrency concepts like threading, asynchronous programming, and parallel processing in any language. 010
Nathan @natfu.be · 30/09/2026First, we need to entirely convert the Netherlands to a field of dams and rivers, for energy creation, water cooling of data centers and building our own cloud infrastructure. The biggest benefit being the removal of the Netherlands. In this essay, I will 250
Nathan @natfu.be · 17/09/2026You can also use the Zotero API and check all your references right? 011
Nathan @natfu.be · 16/09/2026Ed, all your problems would be solved if you started using Emacs I swear, you'd get the GUI for free and like 5 years of tinkering tops! 000
Nathan @natfu.be · 13/09/2026this should sped up 2x and with techno music blasting, I don't make the rules 111
Reposted by NathanThorne 🌸 - Play Cozysters! @ens0.me · 10/09/2026if you wanna be a good trans ally, hire us to good jobs we are way more likely to face financial insecurity, make 60 cents on the dollar compared to our cis counterparts, and a lot of us are *insanely* talented 120744
Reposted by NathanThorne 🌸 - Play Cozysters! @ens0.me · 09/09/2026Looking for full-time, remote senior/staff or tech lead level work. TypeScript is my strongest suit, but I'm a flexible polyglot. Boosts for visibility appreciated! 2168109
Nathan @natfu.be · 09/09/2026I'd prefer if it helped me find edge cases, explore different architectures quickly, brainstorm solutions by looking at a ton of docs I won't myself, etc. I want those models to take me out of rabbit holes or go there for me but the incredibly talkative Claude or weird code golf isn't helping. 010
Nathan @natfu.be · 06/09/2026This will be a banger when someone steals that joke, I was here first 010
Nathan @natfu.be · 30/08/2026How do you feel about YAML generated with Jjinja because it's the only way to loop over complex items 220
Reposted by NathanThorne 🌸 - Play Cozysters! @ens0.me · 23/08/2026Still looking for a remote job. Senior/staff-level engineer or some sort of technical leadership role. TypeScript is my bread & butter but am a polyglot and willing to be flexible. 15312
Nathan @natfu.be · 18/08/2026But what about infrastructure as almost a subset of YAML but not quite? 140
Nathan @natfu.be · 15/08/2026Good stuff, I hadn't seen it in pi but seems like pi-subagents does have it 010
Nathan @natfu.be · 15/08/2026I kinda wish it was easier to have some kind of checkpointing, whatever the harness: being able to time box a task or limit the amount of tokens for a specific task after which the model must return an answer. 110
Nathan @natfu.be · 15/08/2026Block raw python call in tool calls and tell it if prefer `uv run`, it's been my go to lately though it'll sometimes try to circumvent with raw bash strings. Something else I want to try is block most tool calling and only interact with the codebase through `just`. 240
Nathan @natfu.be · 14/08/2026If they're training agents for coordination I'm expecting galaxy brain tier of claude speak between agents 010
Nathan @natfu.be · 10/08/2026I don't know if I have it in me to coordinate 60 Claude subagents. Maybe 59 070
Reposted by NathanAlexis King @lexi-lambda.bsky.social · 27/07/2026The recording of my talk from Software Should Work is now up. :) www.youtube.com/watch?v=0BXu...youtube.comThe Unreasonable Effectiveness of Constructive Data Modeling - Alexis King | SSW 2026YouTube video by Software Should Work 912435
Nathan @natfu.be · 01/08/2026Thanks Simon, I've been looking for something like this! Do you plan on adding how to write evals on your guides in the future? Figuring out what to test and how to grade is difficult and I don't see a lot of guidance there. 110
Reposted by NathanBozhidar Batsov (a.k.a. Bug) @batsov.net · 29/07/2026Flycheck 38 is out with several major improvements (built-in Eglot support, native LSP client, inline diagnostics, quick-fixes, and project-wide diagnostics)! You can read all about it here emacsredux.com/blog/2026/07... #Emacsemacsredux.comFlycheck 38: Might & MagicNot all those who wander are lost. – J.R.R. Tolkien Today I’m happy to announce Flycheck 38 - easily the biggest and most ambitious Flycheck release since I took over the stewardship of the project. I... 394
Nathan @natfu.be · 29/07/2026Uh, I haven't gotten flymake ruff and eglot with ty to work together for Python, the diagnostics seem to be mangled and erase each other so I'll try this out, thanks! 010
Reposted by NathanHillel @hillelwayne.com · 29/07/2026I am delighted to announce that after 5 years of work, 15 early releases, 50,000 words written, and 100,000 more deleted, *Logic for Programmers* is now complete and available in print! Blog post announcement: www.hillelwayne.com/post/lfp/ Website: logicforprogrammers.comlogicforprogrammers.comLogic for ProgrammersPractical math for the working programmer. 1020348
Nathan @natfu.be · 29/07/2026I've been interested in something like this for a long time, bought it =) If you have other recommendations for practical usage I'm all ears! Thanks for writing this book 020
Nathan @natfu.be · 29/07/2026I'm curious about this www.seangoedecke.com/overtraining... and MoE where you'd just route and swap locally based on the user's line of questioning. It might make sense to have generalist models for broad research questions still. 000
Reposted by NathanLéαlinux 🐧 @lea-linux.org · 28/07/2026pagedout.institute Paged Out! is a free technical magazine about programming, hacking, security hacking, retro computers, modern computers, electronics, demoscene, and other similar topics.pagedout.institutePaged Out!Deeply technical zine. And it's free. 0169
Nathan @natfu.be · 27/07/2026I listened to it and it was super on point and interesting! Thanks 😊 001
Nathan @natfu.be · 23/07/2026I wish him good health, plenty of Finnish rye bread and not too many NVIDIA drivers. 010
Nathan @natfu.be · 23/07/2026That's just as a person though, what I really want to see is better dev experience as teams where you can actually get better output AND understanding as a team, haven't find something great to keep everyone in sync though and I don't believe a ton of markdown files is the solution :/ 010
Nathan @natfu.be · 23/07/2026by @grimalkina.bsky.social. I think having agents do TDD doesn't really make sense, they just generate the all the code at once anyway. But building the tests and types yourself after discussing the architecture, that I can see and I've seen @mariozechner.at mention it as well. 110
Nathan @natfu.be · 23/07/2026Thanks! I think I've arrived to about the same conclusions. It's difficult to stay engaged all the time though so I think we should have our harnesses reinforce that instead of having "discipline", that'd be bad UX/DX anyways. I've tried to use something like "grill-me" github.com/DrCatHicks/l...github.comGitHub - DrCatHicks/learning-opportunities: A Claude or Codex skill for deliberate skill development during AI-assisted codingA Claude or Codex skill for deliberate skill development during AI-assisted coding - DrCatHicks/learning-opportunities 210
Reposted by NathanThorne 🌸 - Play Cozysters! @ens0.me · 22/07/2026Still interviewing, very hopeful something works out soon (though happy to hear of any leads). As always, any extra help is appreciated. 2307
Nathan @natfu.be · 22/07/2026You'd need a BDFL with a vision, tight knit core and lots of money. The bar for good languages is also incredibly high these days, you'd need a great linter, LSP, debugger,... Can't these AI labs with infinite money give it a go? 😁 041
Nathan @natfu.be · 21/07/2026How did you surface the design decisions and have the agents "clash" or ask you, and not just implement without oversight if you haven't read the code? 000
Nathan @natfu.be · 21/07/2026We all know @besttrousers.bsky.social is too busy planning fake charts at the evil bald men meeting with Lex Luthor to know what an Xbox is! 000
Nathan @natfu.be · 20/07/2026Using writing docs to prepare my transition to a sci-fi writer career and inflicting incomprehensible metaphors on my colleagues. 000
Nathan @natfu.be · 19/07/2026Yeah that makes sense, I just don't see the return right now. Ofc the usual answer is to do it all from inside Emacs and use it as a multiplexer. I might get away with Emacs running as a systemd service and saving its sessions itself... 010
Nathan @natfu.be · 19/07/2026I wanted to like Zellij but it walks all over my Emacs keybindings and habits; tmux does this as well tbh. Maybe I need to learn vim keybindings... 210