Series · 03 — Strategic synthesis
What Codex missed: a worldview
Source-grounded models give you data; synthesis-strong models give you a worldview — and the honest answer to "which should I use?" is both, in parallel, on the same prompt.
Codex gave me 23 items, neatly ranked, every claim cited. It did not tell me what the ranking meant.
Claude did. The closing sentence of its answer was: "Stop thinking platform-launch, start thinking single-feature artifacts that pull qualified people."
That sentence reordered every other line in my GTM plan. It said: don't run a launch event for "Cerver." Run a launch event for one painful problem you fixed — like, "I built a session_id that survives swapping Claude → GPT-5 mid-conversation. Here's the curl." Codex's #5 item (Show HN) became a different thing after I read Claude's closing line.
Codex was correct. Claude was useful. Different jobs.
Source-grounded models give you data. Synthesis-strong models give you a worldview. You usually need one to make the call and the other to keep you honest. The honest answer to "which one should I use?" is "both, in parallel, on the same prompt."
See it for yourself.
Run the same prompt on Claude and Codex in one session and keep whichever won — that's cerver compare.