Post #1839185

2026-04-22 20:58 UTC

I am not an AI doomer. I enjoy using Claude Code. I do see it as a force multiplier for me, even if it&#39;s &quot;just&quot; being able to find the edge cases, parse large logs to narrow in on a root cause, or just damn well type faster for me. But I&#39;ve always felt the rug pull was inevitable. Compute costs + capex are unsustainable. I&#39;ve always focused on getting useful, long-lasting artefacts (e.g. plans, scripts, tools, macros).

Replies (2)

@rhysmorgan@mastodon.social 2026-04-22 21:02

That rug pull felt inevitable – and I think we're truly at the start of it. People have been complaining over the last couple of weeks (correctly, IMO) that Claude's top tier Opus models have felt stupider, taking much longer to respond. Anthropic added an effort slider to make maximum "thinking" not a default too, decreasing average request resource usage. Opus 4.7 hasn't felt like any kind of meaningful update, and requests just take longer to get back to you.

Open ##2578746
@dom@iosdev.space 2026-04-22 21:04

@rhysmorgan@mastodon.social I’ve had similar feelings, although I’m expecting local models to be good enough for a lot of things by the end of the year (with the right harness / context). I’m tempted to try Qwen 3.6 for plan implementation only. I’m also hoping that Apple might release a local swift / objective c model at some point. It feels like it’s a more tractable problem for them especially if they can distill from future versions of Gemini.

Open ##2578753