Shawn Morel

@charliermarsh The swift evolution string design proposals are real gems. The newer Apache arrow StringView with the 12 byte embedded cutoff shows some interesting pragmatic trade offs at scale

Considering becoming a guy who's obsessed with string design

@zeddotdev this is cool. any thoughts on file pane filters?

Open multiple files in a row from Zed's file finder by pressing the right arrow key. Dead simple, but a fabulous quality-of-life addition. Landing this Zednesday. Thanks init05!

For all the talk about context engineering, it's surprising that there don't seem to be any agentic software engineering tools that let me... engineer mey context

Another day, another socratic dialog with my stochastic parrot

@DiscountedTr If you’re going to be a self-proclaimed shit-post account, you could at least aim to not be a C-tier shit-poster

@dexhorthy You’re absolutely right!

me trying to prompt claude better

@mitsuhiko @AmpCode

Alright I was wrong, @AmpCode folks were probably right. The traditional slash commands I used (/commit and /update-changelog) work just fine as skills. I'm a convert

@design_proteins Wha do you like about it?

I think the coolest looking structure/sequence viewer for protein design right now is the one by BioGeometry, good UI/UX, great visuals Cooking one of my own up right now, we'll see how it turns out

@dexhorthy Hot take but I think it’s here to stay. my prediction is shell capability and container systems. Waaay lower overhead than oci containers, more like BSD jails or plan9 fs protocol

By the end of 2026 the bash tool will be considered harmful. Some people will get popped but more broadly people will realize that a deterministic set of ~20-30 pre defined commands (think - make tasks) is much better for anything but the most greenfield of projects

@thdxr Yes-ish but if it moves to something like eg SQLite and there’s a way to grep the contents that’s fine (eg similar to how I can use the git cli to find stuff in the packed objects)

we're starting to hit the limits with keeping all the data for OpenCode as a bunch of flat json files these are nice for easy readability but obviously doesn't scale super well is anyone particularly depending on this functionality? it'll likely all get moved into sqlite

MoE models MoE problems

@thdxr @jacobmparis What about a short onboarding screen just like when you plug in a new usb keyboard on macOS - “press shift enter. Now press cmd enter. You’re all set!”

@jacobmparis so many different terminals and multiplexers all handle this differently - we tried our best without fucking with your config it's also possible some other app fucked with your config and broke it for us ctrl+j always works as a falback

trying opencode and it sends messages even if I shift+enter so I can't do multiline

@LukeParkerDev @aspiredotdev Love it!

OpenCode is now OTEL Gen AI semantic conventions compliant... so we get really, really cool stuff out of the box from @aspiredotdev because it ALSO looks for that spec

@charliermarsh touch CLAUDE.md

I need several more commenters to suggest Claude's /init, is anyone else available to comment this?

@LukeParkerDev @opencode @aspiredotdev @davidfowl I'd be curious your thoughts about some of the comments I left ;) I included an example plugin that, while no perfect, re-using the ai-sdk instrumentation emits otel gen-ai sem-conv spans that eval vendors like langfuse, braintrust, etc can present rich views for.

@strangemonad @opencode @aspiredotdev @davidfowl Just a PR for now, it’s a PoC that the team can evaluate :) It’s a wider technical decision not just an easy merge

I added full OTEL to @opencode There is no UI that comes close to this. Is this the year of @aspiredotdev ? OpenCode performance issues will be easy to view here, especially with metrics, like memory usage that you can correlate with spans cc: @davidfowl

@HannesStaerk How do you think you might used this to update your training set sampling (if at all)?

They provide a neat server visualizing all sorts of useful data about non-canonical amino acids and how they occur in PDB 👌 🤗 Paper: biorxiv.org/content/10.648… Server: mondet.tuebingen.mpg.de Very nice to easily see, e.g., how much training data of non-canonicals we have for each type and to quickly browse them!

Use this one trick to instantly turn all your existing software repos into agent skills: ln -s README.md SKILLS.md Subscribe for more agentic engineering tips!

@thdxr With a whole week to spare too. My guess is that my Dec usage was actually lower than Nov but I have no great way to compare.

@thdxr Hold my drink. Dec isn’t done yet

we analyzed 3400 opencode zen users and their anthropic usage last 30 days median spend: $13.52 average spend: $44.36 only 75 people above $200 (3%) i know there's a psychological impact of a "subscription" but maybe you don't really need that $200 plan

@jeffreyhuber @0xblacklight That’s more or less the OG take on ai from the 1950s. What do you think is different now?

@0xblacklight AI is reasoning + search everything else will be bitter lessoned

here’s a hot take: compaction is the wrong way to view the problem the correct tool is to give the agent search over its current trace