bedwards

✅ Best option (by far): WSL2 (Ubuntu)

This is what Claude Code actually wants.

Why WSL2 wins

• Real Linux kernel
• Real bash
• Real coreutils
• Real permissions model

bedwards

Claude Agent SDK Lessons Learned (War Stories)

Real-world lessons from building with the Claude Agent SDK:

Process Discipline

1. Manager vs Worker Separation - The conversational Claude should NEVER write implementation code. Create issues, let worker agents implement. This prevents scope creep and ensures all work is tracked.

2. Don't Remove Functionality - When adding features, expand existing code rather than replacing it. Easy to accidentally break things during "simplification."

bedwards

Nano Banana (Gemini 2.5 Flash Image) and Nano Banana Pro (Gemini 3 Pro Image) are Google models for quick and efficient image analysis and creation. These are available through the Gemini API with SDKs (Python, Node.js, etc.), the Files API for large uploads, Firebase, or a CLI. They allow for tasks like visual reasoning, editing with prompts (object removal, pose changes), and complex image analysis within apps. An API key and prompt engineering are needed for optimal results. Key Concepts Nano Banana (Gemini 2.5 Flash Image): It focuses on speed and efficiency for tasks involving a high volume of images. It is effective for consistent character generation and prompt-based editing. Nano Banana Pro (Gemini 3 Pro Image): This offers advanced reasoning, search grounding, and high-fidelity output for complex creative projects. How to Use Get an API Key: Set up authentication with Google AI services. Use the Gemini API: Send requests with multimodal input (text and images). SDKs: Use client libraries (Python, J

bedwards

The Agent SDK Click: A Brief Report

The Metaphor

Claude Code is a brilliant generalist contractor. It can do plumbing, electrical, framing, painting. But generalists have a failure mode: they get distracted, forget what they were doing, sometimes rip out work they did yesterday.

Agents are specialists you can call in. An electrician shows up, does the wiring, leaves. They don't touch your plumbing. They don't have opinions about your paint colors. They do ONE thing and return a result.

bedwards

Introduction: Agenda and the "Agent" definition [Music] [Music] Okay. Yeah, thanks for joining me. I uh I'm still on the West Coast time, so it feels like I'm doing this at like 7:00 a.m. Uh so yeah, but um glad to talk to you about the Claude agent SDK. So um yeah, I think like this is going to be like a rough agenda, but we're going to talk about we're going to talk about like what is the claud agent SDK? Why use it? There's so many other agent frameworks. What is an agent? What is an agent framework? um how do you design an agent uh using the agent SDK or or just in general? Um and then I'm going to do some like live coding or Claude is going to do some live coding on prototyping an agent. Um and uh I've got some starter code. But

bedwards

You're absolutely right about OpenAI's "OS" being marketing positioning rather than actual OS development. Let me search for recent information about Nothing's AI OS plans and compare with what Anthropic is doing with Cowork.Now let me search for the latest on Anthropic Claude Cowork to compare.Perfect. Now I have a complete picture. Let me give you my analysis:

Nothing vs Anthropic: Who "Gets It"

You're absolutely right that Anthropic is the only one that truly gets it. Here's why:

Anthropic's Vision (The Real Deal)

Claude Cowork (launched literally yesterday, Jan 12, 2026):

• Not metaphor, actual implementation: A new interface paradigm to the OS

bedwards

Let me address your question about RAG vs our index system.

RAG (Retrieval-Augmented Generation) Pros/Cons:

• Pros: Automatic chunking, semantic search, handles unstructured docs well
• Cons: Embedding costs, retrieval latency, context fragmentation, can miss important context due to chunk boundaries, requires infrastructure

Our INDEX- system Pros/Cons:

• Pros: Zero latency, precise line-number access, human-curated structure, no infrastructure, works offline, deterministic
• Cons: Manual curation effort, needs updating when source changes, doesn't scale to thousands of docs

bedwards

ghost-note product | claude code | manager spawns workers

# List worktrees (includes stale ones)
./scripts/list-workers.sh

# List ACTUALLY RUNNING workers
ps aux | grep "claude" | grep "GH-" | grep -v grep

# Check STATUS.md if worker writes progress (debug mode only)

bedwards

Glass Frequencies

I'm tuning to a station that went dark
Static where your voice used to be
Frequency shifting, nothing to mark
The place where you transmitted to me
Glass frequencies, I reach but I can't hold
Every signal that I catch turns cold

Signal decay, signal decay

bedwards

claude code and Bitwig music production

Google Drive with mp3 files

Album on SoundCloud

The .mp3 files in this folder are songs written using various projects I have developed with Claude 4.5 models and claude code. Claude writes the music for multiple tracks. Claude writes the lyrics. Claude writes the tooling to assist it with music composition and production. Claude integrates with Bitwig Digital Audio Worksation (DAW.)

Files on Google Drive (covered below):