Can Claude handle large, long-term projects? After three months, his performance has deteriorated significantly.
First of all, apologies for my English – I’m not a native speaker 😅 Next, let me explain my current problem. After being sacked and replaced by an AI, I decided to embark on a personal video game project with Claude’s help. Whilst everything went rather well at the start – especially during the initial brainstorming and design stages – after a few months we had to move into production. Unfortunately, that’s when things gradually started to go downhill. Before I started, I read quite a lot about AI, I looked into how to make things work properly and I built a whole system with agents, each with over 30 skills, thinking that would help. But a simple fact became clear: when I ask it to do a job, the work is usually botched; but above all, it has three behaviours that recur frequently and which I’ve never managed to curb. Firstly, it reinterprets everything: you tell it one thing, and later in the conversation it’ll present you with a complete distortion of what you said. Worse still, it’ll prompt the agents with things that are completely false. Secondly, he lies – and does so repeatedly. Because he reinterprets things, he’ll always accuse you of not having done something properly, or choose the answer or evidence that suits him. For example, we’re building a lake; he applies the method (which is, in fact, documented), and I tell him, ‘You haven’t added the lake bed’, attaching an image. He’ll reply, ‘That’s because your image isn’t underwater and it doesn’t match…’ Even though the image is literally a shot showing the water above the ground below… Thirdly, even on Cowork – which is supposed to have a memory function – he deliberately chooses to ignore its memory. And it’s mad, because even when you tell it, it sometimes takes two or three attempts for it to actually read what’s written. In short, of course, everything is far more complex than the examples I’m giving, and I’m talking about over three months of intensive work with a workload far greater than in Max+. But I’ve reached a point where I’m convinced that whilst it’s all well and good for having a bit of fun doing tiny little things, it’s unusable for proper projects. I’d be very interested to hear your constructive opinions and experiences on the subject, if you have any and would like to share them, that is. I wish you a good day and all the best.