All articles
Filtered by AI Agents — 9 articles
Jun 10, 2026Artificial Intelligence
Fable 5 Cannot Save Your Bad Harness
A smarter model helps, but weak prompts, context, tools, workflows, and verification loops still turn good intelligence into bad output.
Jun 4, 2026Artificial Intelligence
I Gave My Agent A Real Browser And It Got Useful
Browser automation became useful once I gave the agent a permission-aware routine for using a real browser without relearning the dance every time.
May 26, 2026Artificial Intelligence
Why I Am Moving My Agentic OS Substrate From Hermes To Pi.dev
Hermes helped me get an agentic OS prototype moving. Pi.dev gives me a smaller harness I can shape instead of inheriting a whole runtime.
May 21, 2026Artificial Intelligence
25 Commits in 30 Hours: What AI Autonomy Mode Actually Ships
A real 30-hour AI autonomy mode push: 25 commits, 12 arcs, zero API spend, what shipped, what broke, and what still is not real yet.
May 21, 2026Artificial Intelligence
How to Stop AI From Being a Yes Man With an LLM Council
Stop AI from agreeing with bad ideas. Use an LLM Council with 5 role-based lenses, honest framing, and production-tested gates for Codex and Claude Code.
May 20, 2026Artificial Intelligence
The Council Picks The Move. The Triad Ships It.
The LLM Council helps pick the right move before code exists. The Triad makes sure Planner, Executor, and Verifier actually ship what they claimed.
May 12, 2026Artificial Intelligence
Stop Building Agents For Workflows The Tools Are Absorbing
Codex and Claude Code are turning agent features into platform primitives. The durable work is designing workflows, files, tools, and verification.
May 7, 2026Artificial Intelligence
Most Coding Agents Should Be Skills
Agent count is not capability. A practical decision rule for when a coding workflow needs a real agent, and when it should be a skill.
Apr 26, 2026Artificial Intelligence
Karpathy's CLAUDE.md Rules Are a Starting Point, Not a Rulebook
A practical way to think about Karpathy's CLAUDE.md rules: keep the baseline, then add only the rules your own AI coding workflow has earned.
Get in touch