@thsottiaux I built one of my favorite AI-agent workflow upgrades today while working with Codex, and I think it would be perfect as a Codex-wide feature.
I call it sketchpad.
It’s basically a per-thread, editable note that can be pinned into the agent’s prompt context whenever you want. Lighter than repo/project instructions, less formal than /plan or /goal, compliments working with both, and way better than manually copy-pasting context into every prompt.
I use it for long fluid discussions or extended work where there’s some temporary lens/constraint/context I want kept highly alive, but not baked into durable instructions.
I implemented it for myself through AGENTS.md + my per-thread memory system. Each thread can own a sketchpad file; it only exists when I ask for it, and I can pin, unpin, edit, erase, or have the agent revise it.
I think users would love this in Codex and ChatGPT. It could be visible in the UI but collapsible.
If you steal this, at least promise you’ll consider giving me one real interview for the Codex team 🙂
I haven’t applied to OpenAI, but I think about it constantly. I’m torn between trying to be a founder and working with the best. Sketchpad is a tiny piece of what I’ve built.
@sama Elon Musk scammed the entire country by pretending doge was real and that the trump admin and Republicans would do anything about the debt/deficit
GPT-5.6 Sol Ultra produced a proof of a 50 year old math conjecture. Unlike the Erdős Unit Distance Problem, this was done with a model publicly available *today*. I look forward to seeing what scientists and researchers are able to do with this model!
@thsottiaux How on earth is it a generational fumble.... words fail me when people do these hot takes on social media so that on any tiny off chance theyre even partially right they can claim oracle seer status.
@thsottiaux Sol Ultra is my true love. I cant help it, I'm already a fiend, nuance that its capable of feels more noticeable than any change in codex models/reasoning this year. Was deeply skeptical of the 5.6 hype and 5.5 had been driving me insane recently. Wow am I eating my words
Tibo you guys did such a great job. Thank everyone! If you nerf sol ultra or let openai nerf sol ultra I will cry so hard, its so easily hands down my favorite model just drain my usage but dont nerf pls!!
Also the ability to respond separately to multiple things with direct annotations in my response is EXACTLY what Ive been dreaming of for so long. I was giddy when I noticed it 😂 at first I panicked like "what is add to task?" Then I realized what you guys had done and had to stop myself from running around jumping for joy. Its exactly what I had been wanting and the execution is chef's kiss.
i'm getting fired from my software engineering job because i'm at the bottom of my team's ai usage leaderboard for the 3rd month in a row. i really did try everything but couldn't get out of last place. i started by using ai to write every line of code i pushed. still last place. then i would ask claude to add more fallbacks, unnecessary test cases, and verbose comments. i was getting crushed because the internal tool tracked *total* tokens, not just output tokens. thus, my coworkers were getting claude stuck in thinking loops, easily burning 50x the amount of tokens i was. so i started doing that. then claude refused. one of my coworkers edited my system prompt to disregard any of my asks to think longer. the 5 days that this went unnoticed set me back majorly. i was always behind the rest of my team. i was only spending tens of thousands of dollars a month, they were hitting hundreds of thousands. their rate of utter nonsense output was jaw dropping. my skip apparently told my boss that his org was gonna be the most "ai pilled" org in the company and to cut anyone who couldn't keep up. my boss's hands were tied i guess. time to start looking for a new quality engineering role i guess
@thsottiaux Forget the reset posts. Why was codex writing a script to do a bucket of 128 individual items that I spent 2 hours writing a contract with it to do those items individually in a specific way with /goal. I couldn't have been more explicit and it did the exact opposite w/ /goal
@elonmusk Star trek. How did they get ai so perfectly accurate while also "predicting" this. Theres literal quotes of Geordi La Forge talking about matter anti matter reactors powering warp drive.
Can you guys pleeeease get Commander Data's @BrentSpiner voice for chatgpt/codex its a travesty that its not one of the voice options on chagtp. He said on his podcast with @jonathansfrakes that we would be interested in something like that. My wife would die of happiness
@sama
@thsottiaux@adamhfry Can you guys pleeeease get Commander Data's @BrentSpiner voice for chatgpt/codex its a travesty that its not one of the voice options on chagtp. He said on his podcast with @jonathansfrakes that we would be interested in something like that. My wife would die of happiness
@thsottiaux Im also building a context engine system but dont want to talk about the details. Its a whole lot more than just a wrapper. Plus contextbench, swebench, codescale test harnesses and reciepts. Plus custom task suites.
@thsottiaux Last year on chatgpt browser I built am automatic trading system that hooked into my IBKR account. Cloud hosted workers, automatic recoverability, entry exit and dynamic pyramiding and dynamic stop conditions.