AI coworkers need the same onboarding as human hires. Map your team's behavioral patterns before buying tools that force a single workflow on your best performers.
I believe we still have a ways to go before AI cowork will be a real thing. It's cool to have an assistant, but for now that's all it is. I wouldn't delegate important work without close supervision.
Depends on what you mean with important work & close supervision
In AI coding, high-level architectural guidance and loose supervision are already possible
I'm looking to push more of my creative and knowledge work in the same direction
Of course you're never absolved from doing the thinking and setting the course, but I also wouldn't want to have it any other way
Having super powerful and competent AI research assistants that are specialized in specific scientific domains alone could open up a whole new frontier for human knowledge, 100x ing the output of researchers that know how to use them
I code with AI everyday and I can assure you that it's nowhere close to truly understanding the scope of work even when specifically told what to do and how different parts of the code interact with one another.
I mean sure, it can create plans, it can write super optimized code and tests, it can debug and it can even recommend best approaches, but once it accidentally makes an error it doesn't stop to fix it, rather it moves on. That's why you need to double check its work all the time.
I code with GPT 5.2, Opus 4.5 and Gemini 3.0 Pro, regardless of what model i use they all have flaws and need close supervision.
In any case, it's January 20th 2026, who know what models we'll have in 6 months or next year. The difference between Gemini 1.5 Pro (launched 2024) and Gemini 3 Flash (launched 2025) is unfathomable. If the trend keeps up we'll have AGI in no time and everything I just said till' now will no longer be applicable.
I believe we still have a ways to go before AI cowork will be a real thing. It's cool to have an assistant, but for now that's all it is. I wouldn't delegate important work without close supervision.
Depends on what you mean with important work & close supervision
In AI coding, high-level architectural guidance and loose supervision are already possible
I'm looking to push more of my creative and knowledge work in the same direction
Of course you're never absolved from doing the thinking and setting the course, but I also wouldn't want to have it any other way
Having super powerful and competent AI research assistants that are specialized in specific scientific domains alone could open up a whole new frontier for human knowledge, 100x ing the output of researchers that know how to use them
I completely agree with your point on research assistants, but not about AI coding.
While you can leave an agent code on its own for days on end, the final result will be sub par (see this experiment ran by Cursor's CEO https://www.financialexpress.com/life/technology-this-ai-agent-wrote-3-million-lines-of-code-in-1-week-built-web-browser-from-scratch-internet-reacts-4112313/).
I code with AI everyday and I can assure you that it's nowhere close to truly understanding the scope of work even when specifically told what to do and how different parts of the code interact with one another.
I mean sure, it can create plans, it can write super optimized code and tests, it can debug and it can even recommend best approaches, but once it accidentally makes an error it doesn't stop to fix it, rather it moves on. That's why you need to double check its work all the time.
I code with GPT 5.2, Opus 4.5 and Gemini 3.0 Pro, regardless of what model i use they all have flaws and need close supervision.
In any case, it's January 20th 2026, who know what models we'll have in 6 months or next year. The difference between Gemini 1.5 Pro (launched 2024) and Gemini 3 Flash (launched 2025) is unfathomable. If the trend keeps up we'll have AGI in no time and everything I just said till' now will no longer be applicable.
Apparently, the folks at Anthropic are way ahead of us AI plebs https://open.substack.com/pub/importai/p/import-ai-441-my-agents-are-working
Didn't know about Devin - off to Research them now 🤔