Zach Gramstad

opus 5 is no fable. more like opus 4.8 pretending to try to be fable

pretty kewl

Opus is so dumb. I miss Fable 🥲

How was $SPCX able to just buy Cursor? I haven't heard anything about possible anti-trust of a space company buying an artificial intelligence company that is also a social media and now an AI-software building company. Is there not a risk of any of this being unwound?

I have some ideas that I'm just not building right now because I know they will be subpar, and I will throw them away and start over once I have Fable again. So sad. Please come back soon. API-based pricing will also make me sad again. Anthropic did well with this one.

The biggest improvement to my claude code workflow for the last ~6mos. has not been new models dropped by @claudeai , it's been the /codex plugin by @OpenAI . Telling opus to create a plan, codex review, implement, then codex review gets so much closer to hands-off development.

It’s crazy that a “cancellation save flow” often used to be a multi-person, multi-week project, and now I can just knock that out with CC in a day as one thing on my todo list. Fully researched, designed, built, tested, QA’d and client-approved. Such a fun time to build things.

One thing claude code has taught me is how much I should've been able to get done with CLIs all this time

Codex might be smarter than Claude, but Claude is so much better at communicating than Codex. The difference is stark.

Starting to get on my nerves how @ChatGPTapp always ends something with a question. "one more clarification...", "could you give me more details about...", etc It's sometimes appropriate, but a lot of times just transparently an engagement hack. Starting to lean @claudeai more

So I feel like Claude code is actually much better at doing things than @openclaw And the UX is so much nicer too. I feel like if you actually want to build stuff, use CC. Then just use openclaw to interact with it.

I don't understand all the people fawning over Opus 4.5. It has been like marginally better than Sonnet 4.5 for my workflow, if that. It's way worse at research and spec-writing than GPT-5.1 high and often doesn't follow a spec that has clear tasks and acceptance criteria.

So annoying that this used to actually map to the top 3 posts when I clicked it, and now it's just "algorithmic variable reward" system learning that I'm a @sama fan

Actually valuable @Apple Intelligence feature request: Blocking spam texts and calls. The volume seems to have increased dramatically in the last few years/months, but nothing has evolved to handle it. My grandfather was scammed out of $5k last month by scammers.

Seems like @Google should buy @cursor_ai now. They don't have a competing IDE like Microsoft does (VSCode), and they'd get so much coding data across all LLMs to improve their models further. Would only cost ~10% of cash on hand, assuming all-cash deal.

Just updated my Arc browser and my uBlock Origin has been disabled. All my content blocking settings no longer work with uBlock lite. I thought this wasn't going to be an issue @browsercompany @joshm?

This sort of thing will work for a while. But eventually, people will get angry that they are being lied to by AI mimicking personal testimony. Trust in information will continue to fall, and authenticity will become even more important in advertising.