The AI-as-judge pattern lets us use AI for evaluation. But what if we wanted contrary opinions on a plan/architecture?
I've written up how I've been using a jury (or mixture of critics) that work with a judge for a more holistic review.
👉
☁️ Chris Reddington
@reddobowen
- I built a PowerPoint diagram to explain agentic AI, then turned it into an interactive site: chrisreddington.com/agentic-ai/ A visual mental model for context engineering, context windows, agentic loops, MCP servers, Skills.md, Agent.md, and more. v1, so feedback welcome!
- If your agents have to keep relearning the same repo quirks in every new session, that's a memory problem. Instructions tell agents what you already know. Memory captures what the work teaches next.
- Typing the same setup into every new agent session, is a reusability problem. Read my latest post explaining how AGENTS.md, SKILL.md, and Copilot layers help you package that context once and reuse it.
- Bigger context windows don't make agents better on their own. Better context does. Context engineering is the craft of deciding what the agent can see, what it carries forward, and what stays out.
