You think everyone knows about AI and then you go out into the world. World with ppl happily struggling for a few hours with positioning text boxes on PowerPoint (witnessed it myself). And no curiosity about AI either. Compounding for us in this space better be epic.

Is Anthropic throwing free compute at us or is Opus 5.5 really this efficient? For tasks with lots of image processing, I’m finding Opus 5.5 to have amazing token efficiency. I should do proper evals but I’m just getting so much done. Astra/6.1 Sol has a serious competitor.

Being able to see Citi and HSBC logos is worth it. I might be biased but I would pay more if it was JP Morgan. 😅

It’s so strange that Opus 5.5 is now getting more done and **quickly** than Sol 6.1 (or Astra). Incredible change. I was almost ready to cancel Claude. Now I’m wondering if I should get 20x and cut Codex in a month or so to 5x.

$2500 worth of credits from OpenAI after changing the Pro x20. Can’t help think it’s related to $250 from Anthropic. I wonder why…😀

Other sessions work so just got it to send a message to another thread. 😅

What did they do to Astra + codex? It was happily setting up things for me on Cloudflare and suddenly getting blocked today. Clearly additional guardrails brought in which is fine but it has degraded the DX without any improvement in security.

Exciting stuff incoming…

Either Astra is nerfed or Opus is just that much better on launch. I’m just not getting the same juice from Astra xhigh today. Not a clean comparison as complexity of the work has gone up. But it’s like team member who is missing things they shouldn’t.

What the fuck? All those billions spent joining their Iraq fiasco and this is what the UK gets in return. Those 100Bn+ sure could have come handy now. I guess time to choose either we stay subservient or go build it ourselves.

Well, bus company knocked off my BMW’s mirror and their claim handlers are rude. Fair enough but then to withhold CCTV moment it becomes clear they might be liable…naughty. So, like any self respecting physics, maths and CS guy. I will now personally win this w Opus and Astra😀

Fine. Opus 5.5 for video editing tasks at least is superb. I’m really enjoying its judgement.

What if the learning platform could create personalised teaching content for you exactly when you need it. Opus + our content library + math components + H3. Thanks @fal for the credits to experiment.

Gemini 3.8 Live has inconsistent British accent. I just can’t seem to make it work. It’s a good model but this is frustrating.

How are ppl managing context on Claude Code? Compaction is solved on Codex. I can run threads for weeks. On CC anything beyond 300K and I start seeing issues with Opus’s work. Then I have to do some version of handoff dance. Any useful pattern I’m missing?

Competition is a good thing. I hope this is a good model because I am close to replacing Claude with another Codex sub.

I’m at 0% now. Waiting on Tibo. 😂

A few months ago, it was difficult to get wife to try Codex. She was a Claude fan. Today she told me she prefers Codex. This is someone non technical and focused on getting things done efficiently, fitting it around school/club runs. No time for BS - Codex is delivering.

Isn’t that exactly the type of challenge that would make ppl want to build it? 😀