Ephes Blog

I'm sorry my attempts at freestyle rapping summoned a demon to your party. --Ricki Yasha Tarr

I'm on vacation right now. While it's lovely sitting by the ocean, the combination of wind, water, and sand isn't exactly laptop-friendly, so I had to leave my computer at home. Bit of a bummer. Haven't done much reading either - mostly just been mulling over the Home Assistant setup I'm planning to tackle when I get back.

Articles

• Welcome Our New Fellow - Jacob Tyler Walls | Cool 😎
• AI Is A Money Trap | The economics of it all is also completely beyond me.
• The Hidden Drivers of HRM's Performance on ARC-AGI | nice analysis
• AI in Home Assistant in 2025

Music

• David Bowie Nine Inch Nails Hurt
• Ecca Vandal - CRUISING TO SELF SOOTHE

Out of Context Images

Taking a summer break this week. Released a new version of django-cast with Wagtail 7.1 support, and just sharing a handful of links.

Articles

• The Future Of Software Is Small | This resonates with me - I think it applies to standard office tools like Excel too. What an exciting time to be building software!
• 🤖 VibeOps: Using Claude Code on Cheap VPS Servers | I've been doing something similar - giving Claude root SSH access when I'm working out Ansible deployment scripts. It's surprisingly effective.
• Python performance myths and fairy tales
• Enjoy better UX with Wagtail 7.1 | Cool!
• TabICL: Pretraining the best tabular learner
• Debian 13 "trixie" is released
• A contact just told me that my old "LLMs generate nonsense code" blog post from 2 years ago is now very outdated with GPT5 because it's so awesome and so helpful. So I asked him to give it a test... | Thoughtful thread that critiques LLM capabilities without being dismissive. I disagree with the conclusion though - these models don't need to be perfect to be valuable. Humans fail at reasoning too, yet we manage. I lean toward a functionalist perspective here. Still, the humanist/rationalist critique raises valid points. It reminds me of the Chinese room argument (which I also find flawed but thought-provoking).

In searching one does not know in advance how — Piet Mondrian

I've been experimenting with running coding models locally and managed to get GLM-4.5 Air up and running with mlx_lm.server and open-webui. It's pretty amazing what these local open-weight models can do these days. Next up, I want to try integrating local models with CLI-based coding tools like opencode or crush.

Articles

• New Claude Code feature: Subagents | I wasn't able to use this feature to generate any value for me yet
• A Friendly Introduction to SVG
• Frontend Performance Checklist
• It’s time for modern CSS to kill the SPA
• My 2.5 year old laptop can write Space Invaders in JavaScript now, using GLM-4.5 Air and MLX
• GLM-4.5: Reasoning, Coding, and Agentic Abililties
• Agentic Coding Things That Didn’t Work
• Do you validate your HTML?
• AI and Me: A Developer’s Perspective | Can relate, yes
• 6 Weeks of Claude Code | Very detailed and insightful

Software

• Peekaboo MCP: Lightning-fast macOS Screenshots & GUI Automation
• Crush - Your new coding bestie, now available in your favourite terminal
• Welcome to Anthropic's Prompt Engineering Interactive Tutorial
• Open WebUI 👋
• opencode - AI coding agent, built for the terminal.

Videos

• How To 10x Your Notes: Obsidian + Claude AI Agents | I'm just interested in all things obsidian
• MCPs are Boring (or: Why we are losing the Sparkle of LLMs) - Manuel Odendahl | Yes, important point.