Summary
David Ondrej, who claims over 2,000 hours of AI coding experience and has interviewed leading agentic engineers, shares his complete 2026 agentic engineering setup in detail. The video covers a layered stack of open-source and custom tools: BB (a unified interface supporting Claude Code, Codex, Cursor CLI, Grok, and Hermes in a single UI), CMAX for multi-pane terminal agent management, Ghost Ty as a native terminal with Herder as a lightweight agent runtime, and Coral — a custom priority-queue manager Ondrej built himself to ensure high-priority agents surface before low-priority ones.
Ondrej also walks through agent harnesses including Pi Agent (the minimal four-tool harness that OpenClaw is built upon), Cursi, and several others, explaining the tradeoffs of each in terms of stability, multi-agent scale, and subscription flexibility. He demonstrates live VPS setup entirely through natural-language agent commands, showing how one bootstrapped agent can install and configure subsequent ones.
The video goes deeper into concepts like work trees, sub-agent spawning, and skill libraries — reusable prompt modules that trigger multi-agent review pipelines. For developers looking to move beyond single-agent chat and into orchestrated, parallel agentic workflows, this is a dense practical reference covering real tooling choices and the reasoning behind them.
📺 Source: David Ondrej · Published September 02, 2026
🏷️ Format: Workflow Case Study
