Skills > Agents: Why Powerful LLMs Make Frameworks Obsolete (2026-04-10) — Single-agent systems match multi-agent at equal token budgets. 3.2x cost, 17x error amplification, 4.8s latency overhead. Claude Code's skills + subagents pattern shows a better way.
Your Orchestrator Prompt Is a God Class (2026-02-27) — Why agent orchestrator prompts spiral into unmaintainable god classes, and how a prompt-lab CLI with regression testing closes the self-learning loop.
Context Rot Is Real, and Subagents Are the Fix (2026-02-10) — Why agents get dumber over time, and how subagent-driven context isolation fixes it. Lessons from building 20+ agent systems at Epiminds.
The Architecture of Creative Machines (2026-01-11) — How to simulate human creative teamwork in agentic systems, reverse-engineering brainstorming into multi-agent pipelines with structured knowledge bases.