@benjamingeer @andrewnez That's a good point

We're very much in the dirt path point of open source as infrastructure

My current LLM toolchain

https://chris.wensel.net/post/llm-toolchain-june/

#llm #ai

Haven't had time to blog about it, but I've released a set of skills I use every day to resolve bugs and feed my RDR process with complex features or debt. I pushed them to this repo, ymmv.

https://github.com/cwensel/kata-flight

#llm #ai #claude #codex

TBH, everyone has their own bespoke workflows, but I think we should all bias towards sharing and learning from them. So I don't expect anyone to adopt this, it probably only makes sense to me. but I would love to see other flows and learn from them.

gone are the days when legacy Cascading users pay me to fix decade old bugs they can't figure out.

funny, codex is writing perl scripts where claude probably would be writing python scripts to solve bulk find/edit issues

My wife has started a #dog grooming business here in the outskirts of #Berlin. Check out her website: https://ahrensfelle.de

Tell all your dog owning friends! #ahrensfelle

protip for arcaneum users

https://github.com/cwensel/arcaneum

> arc log tail --lines 50

to watch and ensure your skill are pulling in research materials (via search) during any design phases or when creating tests.

#llm #ai #rag

I've been using a spec first LLM dev model for about a year now, it's evolved quite a bit.

I recent got around to converting the process to live behind skills.

https://github.com/cwensel/rdr/blob/main/SKILLS.md

Find the whole concept dubious, same.

But I have tried to pull in as much research and actual empirical data as possible. Some of the research and journey is documented.

https://github.com/cwensel/rdr/blob/main/RESEARCH.md

Woah, diagramming in the terminal 🤯

📐 **pinstar** — A terminal-based editor for Obsidian Canvas, Mermaid, Graphviz, and PlantUML

💯 Create, edit, and navigate diagrams with pan/zoom, layouts, node editing & undo/redo

🦀 Written in Rust & built with @ratatui_rs

⭐ GitHub: https://github.com/reekta92/pinstar

#rustlang #ratatui #tui #terminal #diagrams #obsidian #opensource

Imagine you walk into a lab because you've generously agreed to participate in a psychology experiment.

It's pleasant: first you're shown a series of paintings and you indicate your preference for a certain artist, and you're told the paintings you saw were from two artists, and you mostly prefer Kandinsky over Klee. How nice!

But there's a twist.

fastmail's tech support seems to have degraded to "write a deflective answer so they are off my queue" :-( Maybe it is time to migrate all email elsewhere.

Visual Design for product designers and software architects.

Practice collaborative system design in a (remote) design studio setting, iterating across business and engineering strategy, and product and architectural design.

Oct 5-7 and Oct 12-14, 2026 at 11 am - 3:30 pm Eastern Time (US/Canada)
Remote (on Zoom)

More info/enroll: https://ti.to/bredemeyer/system-design-may-2026

Preview material:

Systems and System Design Overview
(pdf): https://www.ruthmalan.com/systems/2025SystemDesignOverview.pdf

Architecture Decisions (pdf): https://ruthmalan.com/Leadership/TechnicalDecisions.pdf

Arcaneum 0.7.0 is out

https://github.com/cwensel/arcaneum/releases/tag/v0.7.0

"We tried out the LLMs that everyone else was touting. Accuracy was about 10%, not 80%." Stonebraker on real enterprise text-to-SQL. https://youtube.com/shorts/iMKrev7ZKmQ

#TextToSQL #LLM #DataWarehouse

don't have the cycles, but I want to do an analysis of commits on very popular projects to see how often (or the delay until) an maintainer step on or rewrites a contributors feature/fix contribution.

is it common/easier to just rewrite the contribution than guide a new contributor through quality code. I've always rewritten on submission, but curious if there are built in delays.

I often wonder how much of Claude’s token usage and hence Anthropic revenue is pure waste (e.g. Claude not following instructions or doing things repetitively that it should have done once).

I expected it to be high but not this high. 😱

Of note, I've started indexing my LLM harness session histories in Arcaneum to help me identify and refine repeated workflows which I then mint as skills.

The result is starting to look like a state machine. At some point I'll document the whole research through ideation through design and then implementation phases.

This is a very high human touch model compared to what i'm reading about from other devs.

Posted some updates on my RDR spec first process. It's driving a new project I hope to talk more about soon.

https://chris.wensel.net/post/rdr-process/

Lot's of stability changes to Arcaneum queued up for the next release.

If a user, please pip install -e . main and give it a spin. also updated docker dependencies, so restart the containers (no dataloss)

https://github.com/cwensel/arcaneum

Claude-Mem lets developers search across Claude Code sessions by storing dated observations locally. The tradeoff: moving from a plugin to a local database stack with SQLite, Chroma, and provider calls to monitor. Open source under Apache 2 as of v13.0.0. https://www.implicator.ai/claude-mem-turns-claude-code-sessions-into-searchable-project-memory/ #AI #DevTools #OpenSource