@thsottiaux Loving the 28 days of improvements, but as a UK Pro subscriber, watching all these Dots updates is like being invited to an all-you-can-eat buffet and being told I'm only allowed to smell the food. 😂
Any chance us Brits could get a usage reset instead, or are we just collecting disappointment points? 🇬🇧😂
@thsottiaux Been using Codex hard on a space RTS in Unreal — full fleet battles, hangars, original ships. Wild how much of the systems work an agent can actually carry now.
Perfect timing 😂
I’m currently seeing how far I can push Codex with long-running multi-agent dev sessions where the repo remembers the project, Terra builds, Luna tries to break it, and Sol integrates everything.
90% faster long-thread loading sounds very welcome right about now 👀
Challenge accepted 😅
Currently burning quota by making Codex build a guardrail system… using the guardrail system… while Terra agents build V3, Luna agents try to break it, and Sol integrates the survivors.
At this point I’m not testing Codex anymore, Codex is stress-testing Codex. 😂
I switched to Codex because it works at project scale, not just snippet scale.
My recent experiments ended up becoming an “Evidence-Gated Builder” because I kept hitting the same problem: an agent saying “done” isn't evidence.
I'd love a first-class supervisor/orchestrator mode that can split work across agents, keep durable mission state, require independent verification, detect repeated failed attempts, use bounded repair loops, and only promote code when acceptance criteria + tests + review agree.
Parallel agents are useful. Coordinated, recoverable, evidence-backed agents would be the bigger leap.
@thsottiaux Codex’s biggest unlock may be native evidence-gated workspaces: read-only sources, quarantined writes, durable checkpoints, process ownership, and retries only after a changed fingerprint. I’ve used this across 100+ work packages. Don’t eliminate failure—contain it.
@sama@sama I have been blessed with Sol access… and cursed with no quota 😭
Any chance of a tiny cheeky reset so I can play before the 11th?
I promise to use it responsibly. Mostly. 👀
Hey @thsottiaux,
Over the past month I built Draft to Take — a full local script-to-audio production studio (Script Canvas + Voice Studio + Timeline) using Codex as my main coding partner.
I have severe dyslexia + ASD, so Codex basically broke the syntax barrier for me and let me ship something I never could have otherwise.
Really proud of what we built: https://t.co/tYBeswLuUc
Would love to be considered 🔥
@sama 100% vibe coded with 5.5 Pro. Used massive tokens to direct AI agents to build this local AI audio DAW. Earlier models failed the architectural complexity of React + FastAPI + Docker all at once. 5.5 crushed it.https://t.co/Hhbml51bIe
@sama Built Draft to Take — local AI script → voiced timeline studio using the new wave of voice models. Write → generate takes → mix on timeline, all on your own GPU. Released the Docker beta yesterday. https://t.co/Hhbml51bIe
@sama Smarter models shouldn't do cheaper work. I built a model-router skill for Codex that keeps GPT-5.5 as the Mission Commander for strategy/escalation while routing mechanical tasks to 5.4-mini agents. Smarter models aren't just chatbots; they’re the OS for agentic collectives.