Peter

🦞 AI Enthusiast | Prop Risk Manager | Believer 🦞

[Peter @Pr0ftrader]

@elonmusk I believe that with @cursor_ai and @SpaceX you guys can actually surpass Codex and GPT systems.

For now what we all want is systems for harnessing AI, I believe Codex is there, but I also believe Cursor will give a tough competition to Codex! Love it!

[Peter @Pr0ftrader]

@cursor_ai @SpaceX Finally someone can compete with Codex.

I'm rooting for you @cursor_ai 🥳

[Peter @Pr0ftrader]

@theo It is good. But for me it doesn’t have a good platform like codex.

Plus the pricing is not good enough for just answering how many R’s are in the word - Strawberry!?!?!?

[Peter @Pr0ftrader]

@zavxai Perplexity?!?!?!

[Peter @Pr0ftrader]

I agree with this

[Theo - t3.gg @theo]

Hot take: all of these games are more exciting to me than any PlayStation exclusives currently in development

[Peter @Pr0ftrader]

I'm not a motivation/influencer listener or comsumer.

Have to say @AlexHormozi does have some good sauce for people who are taking risky steps in their career.

I would recommend it. 🖖

[Peter @Pr0ftrader]

@HeyYoGB This is what I want AI to help with: trend comes in, agent checks if it fits my lane, pulls the screenshots or receipts, drafts 2 angles, then I decide. Otherwise you end up chasing every shiny patch note and the account slowly loses shape.

[Peter @Pr0ftrader]

@anson001_ This is the part people miss with influencer work. The hard bit is not "find creators." It's tracking who posted, which angle landed, what needs follow-up, and what should never be repeated. Agentic workflow helps only if it keeps receipts, not just drafts.

[Peter @Pr0ftrader]

@_avichawla A second brain that talks back is useful only if it also knows what to ignore. My worry with agent memory is every note becoming "context." Good setup needs a loot filter: current goal, source notes, stale stuff, and anything that needs approval.

[Peter @Pr0ftrader]

@sebkaydesign Talking to Codex is underrated. Most good prompts start like a messy voice note, not a spec. I want the tool to clean that into a handoff: goal, context, no-go areas, review step. Otherwise it's just a loading screen with extra steps.

[Peter @Pr0ftrader]

@zahareus This is where AI starts feeling useful for creators. Half the context is already in clipboard history: screenshots, saved links, rough hooks, weird little notes. Keeping that local and revocable matters, otherwise the content stack turns into inventory clutter fast.

[Peter @Pr0ftrader]

@jumperz The report-only step is the important part. Same rule I use with trading ops: inspect first, change second. For Codex cleanup, I would rather see the bloat, back up active handoffs, then close the app before touching state. Speed is nice, but losing context is expensive.

[Peter @Pr0ftrader]

@emmynicefx This is exactly the trade a prop risk desk hates most. If TP hits, it looks smart for 5 minutes. If SL hits, the account is gone. The outcome can be green and the process still be broken. Passing an account by risking the account is not really passing.

[Peter @Pr0ftrader]

@delbert_trades The 30 days flat part is where most traders fail. It sounds simple until the account is red and every decent candle feels like a way to fix it. In prop risk, I care less about the next setup and more about whether the trader can stop pressing when the rules say stop.

[Peter @Pr0ftrader]

@Dipanshu_AI This is the part that made agents click for me. A prompt is a one-off trade idea. A loop is the rulebook: what to check, when to stop, what needs approval. The danger is writing loops that only optimize for output and forget the review step.

[Peter @Pr0ftrader]

@VermaAakash3 @Test_Sprite "Output exploded. Review capacity didn't" is the whole trade. Same thing in risk: more positions is not progress if the desk can't review exposure fast enough. For AI code, screenshots, logs, tests, and a clear reject button matter more than another 20 generated files.

[Peter @Pr0ftrader]

@doodlestein Worktrees feel fine if you treat them as a risk boundary, not a productivity flex. Let agents make parallel attempts in separate rooms, then force one human review lane before main. The mistake is thinking isolation also solved coordination.

[Peter @Pr0ftrader]

@petergyang That human checkpoint line is the part people will skip. I’m testing this on my own X profile right now: Codex scans replies, updates the notes, finds the next target, then I still approve the post. The time save is real. But the taste/risk filter has to stay human.

[Peter @Pr0ftrader]

@AndrewCurran_ Reviewable is the important word here. Memory gets useful only when you can see what the system thinks it knows about you. Otherwise it turns into hidden context drift, and you only notice after the answer starts feeling slightly off.

[Peter @Pr0ftrader]

@Mohsine_Mahzi @ajambrosino Same here on Windows after updating, still no Computer Use. I don't think this is just an ARM/x64 thing. The annoying part is it breaks the whole operator loop: browser, local files, screenshots, then action. Without that, Codex turns back into a chat box pretty fast.