Snapshot of Nov 08, 2026
Now + Setup
One dated snapshot of this page. The current one is always at /now.
Current focus
- This blog. It’s been my main focus lately, and I’m still improving it.
- I Love My Evening Television. My AI-made TV series, and a demo
of what Opus 5.5 can animate.
Wordhord (still an idea)
- A shared language core. One API and database any language-learning app can read and write to track a
learner’s level, streaks and vocabulary. No app has to relearn what you already know.
- Not a translator. It isn’t any-language-to-any-language translation.
- Vectors, not raw meaning. The core holds words as embeddings instead of raw semantics or prosody, and a
closest-vector check tells it which words you’ve learned. Learn one word in a neighborhood (big, large, giant,
huge, humongous) and apps can prioritize words outside it.
Current setup
Everything lives in the cloud. Each part is modular and replaceable, so the whole thing can keep expanding, and my
agents work inside it.
- Stack. Cloudflare R2 for storage, Infisical for API keys, GitHub for repos and skills, Notion as my
second brain, Runpod for GPUs and Cloudflare for hosting.
- Interface. Gages, my Grok bot, which dispatches Claude Code agents.
- Subscriptions. Two Anthropic Max 20x plans, ChatGPT Plus and a paid xAI plan.
- Paid APIs. ElevenLabs (AI voices) is the most important. Hugging Face is the biggest, about $57 a month, and
its Seedance video model is expensive, so it’s likely the first cut. DeepSeek runs the agents inside my own code,
like RAG Goobie, the Ask box on the ledger. I also have a Gemini key I probably don’t use.
- Game dev. Godot for now. Unity and Unreal are worth another look as their MCP support (how AI agents drive
them) improves, and Blender MCP for rendering deserves a second look too.
History
- Nov 08, 2026: Blog work and my AI-made TV series, the Wordhord idea, and the cloud setup my agents work in.