Thariq@trq212
Using Claude Code: Spending your effort
His working loop: interview, then build on low/medium, then verify on high.
Sep 25, 2026Read on X
elvis@omarsar0
Jev-as-a-Judge for Agent Evaluations
Using the Jev model as an LLM judge for agent evals, with a refund-agent example that checks tool calls, not just the final reply.
Oct 6, 2026Read on X
EK@ekcheungAI
Skill Deep Dive: what every SKILL.md costs you in Claude Code and Codex
Advice: put the use case first in the description.
Oct 18, 2026Read on X
Tibor (Tee)@tibor_tee
Ship Code at Scale with Grok Bot and Cursor: people and Bots in one thread
People and bots in one thread, trust built step by step, and when to use the Bot Development Kit.
Oct 17, 2026Read on X
Santiago@svpino
Simple is not always better (and it’s often worse)
A warehouse note in the tool output caused an unwanted refund, so he splits responsibilities across agents with guardrails.
Sponsored: “thanks to Google Cloud for collaborating with me on this piece.” The demo is scripted, not a live failure.
Repo: github.com/svpino/returns-agent
Oct 16, 2026Read on X
Tomy@TomySpagnoletti
We stopped reviewing the code. Let's not stop reviewing the agent.
The human's job now is reviewing the agent's report (decisions, assumptions, asks), not the diff.
Oct 12, 2026Read on X
Eugeniu Ghelbur@eugeniu_ghelbur
Build an Obsidian Second Brain Every AI Agent Can Use
An open-source skill that gives Claude Code, Codex, Grok Bot and others one shared Obsidian memory.
Oct 14, 2026Read on X
beamnxw ./@beamnxw
Grok Bot Engineering: Set Up a Harness for Your One-Person Company
A defined job, a source-of-truth brief ('fill unknown fields with unknown') and an approval boundary.
Oct 5, 2026Read on X
Credit and copyright stay with each author. We never add affiliate links.