Zero to Pete Weekly #1
Last week's signal.
A new experiment.
I read too much. Docs, release notes, research posts, Twitter threads from people who ship. Most of it’s noise. Some of it changes how I work.
This is the filter. Once a week, the stuff that’s actually worth your time, plus anything I published that you might’ve missed.
No fluff. No “10 AI tools that will BLOW YOUR MIND.” Just signal.
Let’s get into it.
Favorites from last week…
✦ Claude 4.5 Opus’ Soul Document
Turns out Claude might have a soul doc, an internal “character training” document that shapes its values, safety behavior, and priorities at the model level, not through runtime prompts. This LessWrong post reverse-engineers it through careful prompting and reproducibility tests. Anthropic has since confirmed it’s real. The extracted version is incomplete and noisy, but the implications are worth sitting with: this is how you bake personality into a model.
✦ OpenAI releases GPT-5.2 after "code red" Google threat
OpenAI launched GPT-5.2 in direct response to mounting pressure from Google, touting gains in long-context reasoning, coding, and professional task performance. The benchmarks look strong, but Ars rightly frames this as competitive, incremental progress — a useful snapshot of where the frontier is right now, not a decisive leap.
✦ Three new tools for precise image editing in Figma
Figma introduced three new AI image editing tools: Erase object, Isolate object, and Expand image, bringing image manipulation directly into the canvas. For Figma users, this closes a long-standing workflow gap by eliminating round-trips to external tools (*cough cough* Photoshop), making it faster to adapt assets across layouts, refine AI-generated imagery, and keep visual work consistent at scale.
✦ A visual editor for the Cursor Browser
Cursor’s new visual editor matters not because WYSIWYG is new, but because it pairs visual manipulation with agent-driven code updates in the same loop. For anyone vibe-coding with AI or lacking precise technical vocabulary, this makes fine-tuning and pixel-level polish accessible without breaking flow.
✦ Claude-mem
Claude-mem adds real, persistent memory to Claude Code by automatically capturing and compressing what happens in each session, then re-injecting the right context next time you start. It’s relevant because as AI coding workflows get longer and more agentic, continuity is quickly becoming the limiting factor.
✦ Training an LLM only on 1800s London texts
An open-source dev trained an LLM exclusively on 1800–1875 London texts, assembling a 90GB dataset to remove modern bias entirely. Even at an early stage, the model already captures the period’s distinctive voice, and the documented, constraint-driven build process makes this a compelling blueprint for novel LLM use-cases.
✦ Shopify Editions Winter ‘26
This edition is a sweeping recap of everything Shopify shipped over the last few quarters, presented through one of the most ambitious product microsites they’ve ever built. Beyond the sheer volume of updates (150+), what stands out is the design and implementation: an opinionated, immersive narrative that treats commerce as a cohesive system across AI, retail, checkout, marketing, and developer tooling.
Quick pulse check
Two questions each week so I can tune this to what you care about.
In cased you missed it…
Original Zero to Pete posts from the last seven days.
✦ Christmas Came Early: Async Subagents in Claude Code
Claude Code shipped async subagents, allowing multiple agents to run in parallel and wake the main thread only when attention is needed. We break down why this flips the workflow from “waiting on AI” to orchestrating a small agent team, with concrete examples you can use immediately.
✦ Stop babysitting Claude Code by setting up notifications
A missing notification system turned Claude Code into a silent productivity sink, where work stalls waiting for permission prompts you never see. We walked through a simple hook-based setup that adds sounds, speech, or native macOS notifications, eliminating the need to babysit the terminal and making parallel work viable.
That’s the week.
If something here changed how you work, reply and tell me. If I missed something good, same deal.
See you next Sunday.
— Pete
Love this new no-fluff series! Such a time-saver- straight to what matters and right into the week.