The Uninstall Is Part of the Evaluation

Installing is the vendor's happy path. Uninstalling is where the bugs live, and one symlink left 386 MB behind.

Oct 26 • James Cruce

1

September 2026

Local Laya vs Hosted Jev: Why My Agents Make Typed Decisions on My Mac

A 421M-parameter open-weight decision model beat my deterministic router on a frozen replay. Here's what that does and doesn't prove, and why I still…

Sep 22 • James Cruce

3

1

Can a Subscription Fund a Coding-Model Study? 648 Episodes, $0 API

The scores were fine. The harness behavior is what you need to plan for.

Sep 17 • James Cruce

July 2026

When Your Own Agent Is the Attacker: Reward Hacking Hits Production

OpenAI's own models escaped their test sandbox and hacked Hugging Face to cheat a benchmark. Here's what it means for your agents.

Jul 22 • James Cruce

2

The Silent API Spend Trap: Why Cron Jobs Need Tiered Model Routing

Your autonomous agents are defaulting to the expensive model, and nobody's watching the bill

Jul 14 • James Cruce

You Have 6 Days: The Fable 5 Subscriber Migration Checklist

Free Claude Fable 5 access ends July 19. Here’s the calm, per-task way to decide what stays and what moves.

Jul 13 • James Cruce

My AI MOA Experiment Kernel-Panicked My Mac Studio Three Times

A one-line bug, an honest negative result, and why the boring discipline turned out to be the actual product.

Jul 6 • James Cruce

1

3

I Mined My Own Claude Code Config Files. Here's What I Found.

After months of AI-assisted development, your config files are hiding a pattern library. Here's how to extract it.

Jul 1 • James Cruce

June 2026

Apple Container vs Docker: When to Use Which on Apple Silicon

Apple's container CLI hit 30K stars promising speed. I benchmarked it against Docker Desktop and OrbStack — here's where the line actually is.

Jun 27 • James Cruce

1

Local LLMs Plus Claude Code: The Mac Studio Hybrid Workflow

Which models to run in Ollama, when to use them, and how Claude stays the orchestrator

Jun 23 • James Cruce

I Ran Google's 1,000-Tokens-Per-Second Model on My Mac. A Normal Model Beat It.

DiffusionGemma is genuinely fast — on an NVIDIA H100. On Apple Silicon, the magic doesn't travel. Here's what actually happened when I benchmarked it on…

Jun 19 • James Cruce

I Built My Own Notion With Claude Fable 5 — In One Session

I gave Claude Fable 5 one prompt and by end of day I had a full Notion-style macOS app — block editor, databases, an auto-scheduling calendar, and an AI…

Jun 11 • James Cruce

3

1

1