Home › Notes › Coding agents
What makes me confident about agentic engineering
By Rohan Kataria · Mar 30, 2026
Key points
Claude Code orchestrates, 12 Codex workers on GPT-5.4 research in parallel, and GPT-5.4-PRO reviews the result.
I saved my Opus hours for the work where they matter, because this was a volume research task.
One reviewer scores quality out of 100, and anything below the 90% gate loops back.
Single-pass search gives you a draft. Multi-pass agentic gives you a plan you can act on.
Updates and sources
I'm turning this workflow into a skill with Skill-Creator.
Sources I cited in the comments:
From the comments
One reader said the gain is running real-world constraints in parallel, which people track in their heads but can't execute at scale.