Codex Mac App Subagents Fall Short in Real-World Tests

Users testing the Codex Mac App found that running complex tasks with subagents on the Ultra model consumed 80% of the context window without meaningful output, raising concerns about subagent efficiency.

2026-08-30 ~ 2026-08-30 · 2 related posts

• Codex Ultra struggles with long docs: burns 80% context with little code — lucasmeijer · 2026-08-30
• User Reports Codex Mac App Subagents Not Frugal on Main Thread — lucasmeijer · 2026-08-30