IDEAS, DECISIONS & LESSONS

Journal

Notes on AI, architecture and building useful products.

26 articles

08 Oct 2026 · 4 min read

ai-agents

Claude Haiku 5.5 and the Real Math Behind Cheap Model Routing

Haiku 5.5 is 75% cheaper and far sharper at agent tasks, but a token-efficiency tax means the real savings aren't the headline number.

22 Sep 2026 · 4 min read

Opus 5.5 vs. GPT-6 Sol/Luna: Two Labs, Same Day, Opposite Bets

Anthropic and OpenAI released within minutes of each other on September 22 — one chasing fewer tokens per task, the other chasing a hundredth of the cost. Both numbers are real, and they point at different production workloads.

19 Sep 2026 · 5 min read

agents

Fable 5.1's Cache Reads Dropped to $0.25 — But My Agent Loop Still Got More Expensive

Anthropic cut Claude Fable 5.1 cache reads 75%. Artificial Analysis found max-effort runs cost 20% more anyway, because output tokens jumped 1.7x. I ran the math on our own agent loop to find out which story was true for us.

15 Sep 2026 · 4 min read

agents

Claude Fable 5.1's One-Way Thinking Blocks: A Rollback Trap for Agent Teams

Fable 5.1 can read every earlier Claude model's thinking blocks. No earlier model can read Fable 5.1's. That asymmetry looks like a footnote until your team needs to roll back a production agent mid-incident.

13 Sep 2026 · 7 min read

The Doctrine of AI Model Trust

It's not about which model scores higher. It's about which one still gets the job done when you're not sitting next to it. Notes from a year of switching, paying, and betting real client work on Claude.

05 Sep 2026 · 5 min read

agentic coding

Claude's Mid-Conversation Tool Changes: Swapping Tools Without Blowing Up Your Cache

Anthropic shipped a beta that lets you add or remove tools mid-conversation without invalidating the prompt cache. Why that matters for long-running agents, and what I changed in my own agent's tool-loading logic after reading the spec.

02 Sep 2026 · 5 min read

Claude Fable 5.1's 75% Cache-Read Cut Changes How You Should Architect Prompts

Anthropic didn't touch Fable 5.1's per-token price — it cut cache-read cost by 75%. That's not a discount, it's a signal about which architecture pattern the platform now rewards. Here's how I re-cut my prompt structure around it.

01 Sep 2026 · 5 min read

ai-agents

The AI Coding Agent Pricing Cliff: What September 1, 2026 Actually Costs Your Team

Claude Sonnet 5's introductory pricing and GitHub Copilot Studio's promotional credits both expired on the same day. Here's the real math on what changed, why it happened simultaneously, and how to budget for the next one.

26 Aug 2026 · 4 min read

ai-agents

Claude Desktop Now Talks to Ollama: What the Local-Model Toggle Actually Gets You

Ollama v0.33.0 shipped a UI-driven gateway that lets Claude Desktop run Qwen, DeepSeek, and Kimi models locally — no MCP server, no Anthropic feature, entirely Ollama's own proxy. Here's how it works, the tradeoffs in tokens/sec and tool-calling reliability, and when it's worth flipping on for a team.