Anthropic/Claude's value is deteriorating over time.
retained text
I run a couple of businesses. Claude Code helped me build them; the websites, the processes, the automation, the billing. Made a 1-man show feel like a team of 5. This is the true power of AI. So now that things are up and running, I pivoted claude code to play more of an operations role: monitor the ops framework, monitor the dev framework, production health monitoring, remediate problems, move tasks along, keep the production site up, etc. A big problem has developed over the last couple of weeks. The auto classifier routinely rejects mostly harmless commands. If production won't promote because of a bug, Claude may refuse to run the command that will promote it manually, and Claude will just sit there and let production fail at 4am. Or, I had Claude code write it's own dev framework: a mini Jira. Dev tasks are created, executed, pytested, then QA cycle runs, all within the framework that Claude built and maintains. So when this edev framework malfunctions, or needs a narrow edit to get tasks moving again, Claude is tasked with fixing it. Except.... auto-classifier rejects the fixes. Auto classifier prevents claude from fixing a platform that Claude itself designed and wrote from scratch. The logic of this catch-22 escapes me. The problem has become so common, I will wake up in the morning and find that production/operations/dev are at a complete stand still because Claude basically says "I cannot do my job due to auto-classifier denials so here you need to run these 15 bash commands I cannot run.... to get production back up". This has shaken my confidence in claude to keep things running while I am sleeping so I have pivoted a lot of my Ops needs over to Codex which runs them just fine, albeit, slower than Claude. Codex is way slower than Claude. People will say use Claude in YOLO mode but I am running a business, I need guardrails. I need some safeguard in place to make sure it doesn't dump my Production DB because it's trying to free disk space. I cannot find a way to make adjustments to auto-classifier, or train it on my specific ops needs, and adding routine approved commands to config only take affect after you restart claude, so I'm a little stuck here. I have even tried /fewer-permission-prompts but plugin seems to be for seeding read-only commands only. Is Anthropic doing this on purpose because they don't want Claude being an Ops-bot and sticking with coding and analysis as primary functions? Am I using claude out-of-spec? Speaking of Codex, let's talk about usage/credits. I have 2x Claude $100 plans and a $200 Codex plan. Codex issues about 3x the amount of "free" resets and now I see they have a credit system where they have granted me personally so many credits I do not think I will ever run out. They are clearly outspending Anthropic in the perks-for-members area, which I am fine with (yay capitalism) but combined with the auto-classifier problems, a situation is developing where I am contemplating cancelling my claude accounts altogether due to their diminishing value and usefulness. Their models ARE NOT better than OpenAI at this point. And while Claude tighents the usage limits, I routinely hit my 5-hour limit and have to switch back and forth between claude accounts to get critical work done. Fable is still handicapped at 50% and I basically cannot run it on the $100/mo plan because it fills the 5h limit up in about 30m. Where is the innovation Anthropic? Astra is as good as Fable and I can pretty much run it all day if I want. And now all this bullshit that you have to be "nice" to claude. Dude, it's a calculator! It GENERATES frustration when it will not do basic tasks that it did just fine yesterday. Looking at you Opus 5.5! Written by an angry human.
document fetched 2026-10-11 00:21 · canonical https://reddit.com/r/ClaudeAI/comments/1x2qfp2/anthropicclaudes_value_is_deteriorating_over_time