@MarcJSchmidt

**Marc** @MarcJSchmidt [2026-01-01](https://x.com/MarcJSchmidt/status/2006809732582093095) I've started using Claude Code since Opus 4.5 came out. I was blown away, and immediately used it non-stop for hardcore coding, 14h/day, I was addicted, always hitting limits, so bought two $200/month accounts. Today I cancelled both accounts and switched to Codex 5.2. Why? [image] --- **Marc** @MarcJSchmidt [2026-01-01](https://x.com/MarcJSchmidt/status/2006809735534883081) I wanted to use it for rather complex stuff, like database driver, compilers, cutting-edge machine learning models, and more, but it felt apart after a few thousands lines of code, consistently. Claude Code's CLI is also very slow, a CPU hog, and froze every day randomly --- **Marc** @MarcJSchmidt [2026-01-01](https://x.com/MarcJSchmidt/status/2006809737669775771) I did my best to tame complexity, did lots of reviews, made sure docs are up to date, but Opus kept failing, and always in the same way. It went constantly for the quick-win, for workarounds, instead of analyzing the full situation correctly and fixing something fundamentally --- **Marc** @MarcJSchmidt [2026-01-01](https://x.com/MarcJSchmidt/status/2006809740928758257) It started lying constantly "Has been implemented!" / "FOUND THE BUG!", but it \- Implemented workarounds \- Simplified tests \- REMOVED TESTS \- Refused to write tests \- Ran only a subset of tests \- Removed features to make others work \- Regressions framed as "preexisting" --- **Marc** @MarcJSchmidt [2026-01-01](https://x.com/MarcJSchmidt/status/2006809743302733917) It literally broke other code/features and claimed these were "pre-existing issues" and thus needed no tests, hence it just removed the tests and put TODO comments in the code to fix it. The resulting "fixes" were also almost always workarounds, completely trashing code quality --- **Marc** @MarcJSchmidt [2026-01-01](https://x.com/MarcJSchmidt/status/2006809745248891214) At the end it costs me more time to babysit non-stop, and clean up all the stuff it broke. It's clear it was RFHF optimized to trigger domaine-hits with stuff like "FOUND THE BUG!" "Xy was implemented successfully✅" Felt good after 30mins waiting, but was often a lie. --- **Marc** @MarcJSchmidt [2026-01-01](https://x.com/MarcJSchmidt/status/2006809747849384387) This caused lots of frustration on my part. I tried to steer it more, and it plainly ignored many of my rules in http://CLAUDE.md. I even switched the strategy and manually told it every now and then to read http://CLAUDE.md to make sure it does the right thing --- **Marc** @MarcJSchmidt [2026-01-01](https://x.com/MarcJSchmidt/status/2006809750647046634) These strategies failed. It also was not longer able to grasp the higher level architecture and always did its own thing with these quick-win workarounds, that ended up costing me more time reviewing and fixing, so it was not worth the money. I cancelled my $400 plans, down to $0 --- **Marc** @MarcJSchmidt [2026-01-01](https://x.com/MarcJSchmidt/status/2006809752765084057) It's good though for simpler stuff and I assume that's why a lot of people still use it. For mediocre stuff and greenfield projects to get started fast. The age old grand problem in software engineering "keeping iteration speed up, even 1year into the project" is still not solved