Sol Ultra Review in Codex is Too Thorough
doodlestein · x · 2026-07-16
The author shares their experience using GPT-5.6 Sol Ultra in Codex for plan review and revision:
- The review phase conducts multiple rounds of "adversarial review," where different sub-agents monitor different issues before summarizing the conclusions.
- This approach catches many problems, but at the cost of being very time-consuming, shifting the main bottleneck from token/cost to wall-clock time.
- They ran a complete review on their FrankenGraphDB design doc, and a single review took over 5 hours.
- Afterward, they had the model revise the plan based on the review, which took another 4+ hours and was still undergoing a new round of review.
- The author believes this kind of heavy planning/review only needs to be done once early in a project, but the time cost is already becoming unmanageable. They also mention that a faster Cerebras version might improve the experience in the future.
More from coding & agent
- A better path to agent autonomy is running waves, finding friction, and iterating — JnBrymn · 2026-07-22
- Coding agents are heading toward an AI-writes, AI-reviews, human-approves workflow — aftahi_ai · 2026-07-22
- oMLX 0.5.2 adds Mac menu-bar stats, low-bit decode kernels, and faster downloads — awnihannun · 2026-07-22
- GitHub review bot hits its PR limit and forces a 39-minute cooldown — DanielLockyer · 2026-07-22
- Max reasoning effort appears to be mobile-only in Codex Remote, not desktop — GabGarrett · 2026-07-22
- A Reddit demo argues online stores should expose carts and pricing through MCP — gelembjuk · 2026-07-22