In practice, when building agentic systems, AI models are rarely the bottleneck anymore. The harness around them is. Anthropic spent two years building that harness for Claude, the orchestration code that picks the right tools and grades its work before declaring success. Claude itself is built around 62 carefully composed components spanning machine learning patterns like compute optimal allocation, deliberative alignment, bi temporal memory, alongside agentic patterns like the OODA loop, plan and execute, architect editor splits, and many others.
Those 62 components that define Claude’s thinking approach are distributed across 4 main principles: Cognition, Orchestration, Reliability, and Grounding and Trust. — Read More
Daily Archives: May 9, 2026
Ben’s Builds #3 – an email app
What did I build this week?
An email app…
I use Gmail. I’ve used Superhuman for years. I like it a lot. It is fast, keyboard-first, clean, and is good software. But like many saas products, it keeps adding features that I don’t need and more importantly, I don’t need to be paying for email.
I wanted a split inbox and rules to organize my emails.
Kicking off with Codex: — Read More
The Neural Shortcut to Language
Speech is often viewed as a massive leap in brain complexity, but new research suggests that evolving complex vocalizations might be much simpler than we thought. By comparing the brains of ordinary lab mice with Alston’s singing mice, a Central American species famous for its rapid-fire vocal duets, researchers discovered that the difference isn’t a bigger brain or new regions.
Instead, evolution simply tripled the number of neurons connecting the mouth-movement center to just two key areas. This “minimalist” neural adaptation may mirror the same evolutionary trick that eventually gave humans the gift of language. — Read More
The April every AI plan broke
April was a strange month for anyone who’s been tracking AI pricing. I keep a running file of the meaningful packaging and pricing moves from the major labs. By the third week of April my notes for the month had outgrown the page and started spilling into a separate document. Five major announcements, three of the four biggest providers, all in three weeks, all pointing in roughly the same direction.
… Five panicked moves in three weeks, from three of the four biggest commercial AI providers in the world, with one common thread:
The original design of their subscription plans is being challenged by evolving product capabilities and usage patterns. — Read More
The Roadmap to Mastering Tool Calling in AI Agents
Most AI agent failures do not trace back to bad reasoning. The model understands the task, then calls the wrong tool, passes malformed arguments, gets back an unhandled error, and produces a wrong answer anyway. The reasoning layer gets the attention; the tool layer is where production incidents actually happen.
Tool calling — also called function calling — is what bridges a language model’s reasoning to real-world action. Without it, agents are capped by training data: no live queries, no external systems, no side effects. With it, an agent can search the web, call APIs, run code, retrieve documents, and trigger transactions in any system that exposes an interface.
Getting this right means understanding the full stack, not just the happy path. — Read More
Rick's Cafe AI 