AI and the Copy-Paste Problem

As the number of my AI service subscriptions has increased, a workflow naturally emerged where I pit AIs against each other. This is typically when planning projects, but I also use the same process to evaluate and improve AI-written code.

It often goes something like this:

• Gemini drafts a plan
• Claude evaluates it
• ChatGPT refines the final draft

Which AI I use for each task is irrelevant here, especially at the rate they release new versions, which frequently shifts their dominance in different areas. The important part is the unavoidable context switching, and particularly the text-jockeying that needs to take place to complete the flow. It was a minor inconvenience when I was only using ChatGPT and then discovered Claude. I'd compare outputs and see which I preferred for the given task.

But the number of AIs and tools I use on a daily basis continually grows. At the moment I use Claude, Gemini, ChatGPT, Manus and Cursor, sometimes for the same aspect of the same project. As a result, I can end up spending a lot of time moving outputs between tools. Also, too often I have to remind them of my preferences; be concise, offer useful critique, use 2 spaces for indentation, etc. This gets annoying.

This is, of course, because memory capabilities vary considerably between models. Claude and Gemini have none. Manus has a limited memory feature, akin to the memory snippets ChatGPT had prior to its recent upgrade, making it capable of accessing all past conversations. That upgrade makes a remarkable difference, creating a greater sense of continuity and an assistant that is actually getting to know you. But even here, the memory access is limited to specific models.

I'm looking forward to when it's a standard feature of AIs, as I believe it's the path to truly useful AI assistants. I was having a conversation with family a few months back. They're not in the tech industry and only really learned through me about a lot of the amazing things AI can do now. I enthusiastically shared the latest developments at the time, including how I'd been using AIs to help me build far more rapidly, and with far less actual coding. The conversation naturally shifted towards the future and I shared with them some ideas about how it—and our use—would evolve. For a long time I held the belief that it would become deeply integrated into our lives, yet seeing it manifest is still incredibly exciting and even surprising at times. I think inevitably it becomes not just deeply integrated, but deeply personal.

However, I don't think it will be a "one AI to rule them all" scenario. It's clear that I'm not alone in my use of multiple AIs, and I think this trend only continues as more models are released for specific verticals, with increasingly narrowing use-cases. I'm regularly using 5 AI services today. In two years it could be double or triple that. This exacerbates the Copy-Paste Problem even further.

Maybe I'm overthinking this, or overly sensitive to this perceived inefficiency. Perhaps others have found better approaches than what I've described above—I'd love to hear them. Either way, it's interesting watching the evolution of our workflows, and the exponential growth of the tools that are evolving them.