Claude Did the Work — Opus vs Codex | Real Minds AI

At 5.30am this morning Anthropic (Claude) dropped Opus 4.6. Also this week OpenAI dropped their Codex desktop app — equivalent to Cowork by Anthropic.

I ran a messy early-morning prompt in each about JOT, our internally built Practice Operating System — a markdown vault that runs our entire consultancy. Pipeline, clients, objectives, daily rituals, dashboards. 313 files across six layers.

The prompt: review JOT, tell me what’s broken, what could become autonomous agents, and how it could become a product.


What each tool delivered

OpenAI gave me a solid markdown document. Analysis, mermaid diagrams, a phased roadmap, a next-10-tasks list. Competent. Useful. A good report.

Claude didn’t write a report. It did the work.

It read 50+ core files in parallel, produced an interactive HTML analysis with six SVG architecture diagrams, then — unprompted — offered to anonymise the entire vault so we could share it. It replaced 30+ organisations, 16 contacts, emails, phone numbers, locations, and revenue figures across 125 files. Renamed 26 files. Fixed every cross-reference and wiki-link. Ran three verification passes until zero stragglers remained.

Then it regenerated our v3 dashboard with the updated figures and rewired the launcher script.

The takeaway

The future is a very different place.

JOT system integration comparison between Claude and OpenAI Codex
JOT codebase analysis by Claude Opus showing real development work

Got questions about this?

How We Work Proof Talk to us
How We Work Proof Talk to us
Ask us anything