I’ve been using Google’s AI Studio to build engineering apps using mostly iterative vibe coding. The model used is Gemini 3 Pro. Every so often I would want to prompt with a query and not a feature request but I had to fight like the dickens to stop the ‘agentic LLM’ from making code changes and not simply answering my question. I most often lost the battle.
I just tried this for the first time this morning. In about three minutes, I had a fully functional time blocking app. It was amazing. Although I'm quite sure that the app is nowhere near ready for deployment, it could work well as a personal app.
I'm noticing more and more with ChatGPT when I give it commands that it has a tendency to want to do its own thing. Sometimes, I have to tell it multiple times to do X, Y, and Z. It does make you wonder where this is going to go in the future, and with that kind of friction, how many people are going to keep arguing with it? I do, but that's because I'm very particular.
Yeah, I've noticed the same thing. Mostly I ignore the extra bits (or just scan it for anything useful), but sometimes it needs to be reigned in. Interesting times. (I find myself saying that more and more!)
Craig - I had a similar reaction with a few caveats. What I am sensing - and some of the other business AiIwriters I follow on substack make this exceptionally clear - is that agentic AI is incredibly useful for engineers and others in fields where automating significant parts of their work not only makes sense but is not threatening to their skill set - these are the kinds of drudgery and repetitive tasks that AI is exceptionally good at. The problem of course for those of us in education is that the completion of that work is not so much the point as the process employed. Not only can students now generate an entire presentation based on a few sentences, snippets of a reading, or link to a web source, but they won't be incentivized to double check or do the proofreading necessary to make the work their own because the quality is improving significantly. We're moving from a phase where AI can do almost anything we ordinarily require of students in terms of product. There is going to have to be another way to evaluate student understanding. But try having the new version of Gemini build you an app or something more creative that might be outside the scope of your ordinary capabilities and you can do some interesting things. But I agree with you, I've tried a few of these agentic browsers and I'm just not ready for them yet - I liken it to the analogy that I never really made the leap from single person video games to multi-player. There was just too much going on in the multi-player versions! I liked slowing down and being more methodical which is kind of what you're describing. My 25 year old son, of course, is completely the opposite. Ominous directions but I think it's going to continue to get even more complicated. If this feels like it's moving too fast for folks that have stayed on top of this, can you imagine what it will seem like to people who have been more or less in the dark for the past few years?
Thanks for your thoughtful comment. I think it boils down to discerning when friction is bad and when friction is good. For mundane drudgery, friction is bad, for learning ... well that's a different story.
I’ve been using Google’s AI Studio to build engineering apps using mostly iterative vibe coding. The model used is Gemini 3 Pro. Every so often I would want to prompt with a query and not a feature request but I had to fight like the dickens to stop the ‘agentic LLM’ from making code changes and not simply answering my question. I most often lost the battle.
I just tried this for the first time this morning. In about three minutes, I had a fully functional time blocking app. It was amazing. Although I'm quite sure that the app is nowhere near ready for deployment, it could work well as a personal app.
I'm noticing more and more with ChatGPT when I give it commands that it has a tendency to want to do its own thing. Sometimes, I have to tell it multiple times to do X, Y, and Z. It does make you wonder where this is going to go in the future, and with that kind of friction, how many people are going to keep arguing with it? I do, but that's because I'm very particular.
Yeah, I've noticed the same thing. Mostly I ignore the extra bits (or just scan it for anything useful), but sometimes it needs to be reigned in. Interesting times. (I find myself saying that more and more!)
I hear you, Craig. I'm right there with you. I find myself shaking my head more than ever.
Craig - I had a similar reaction with a few caveats. What I am sensing - and some of the other business AiIwriters I follow on substack make this exceptionally clear - is that agentic AI is incredibly useful for engineers and others in fields where automating significant parts of their work not only makes sense but is not threatening to their skill set - these are the kinds of drudgery and repetitive tasks that AI is exceptionally good at. The problem of course for those of us in education is that the completion of that work is not so much the point as the process employed. Not only can students now generate an entire presentation based on a few sentences, snippets of a reading, or link to a web source, but they won't be incentivized to double check or do the proofreading necessary to make the work their own because the quality is improving significantly. We're moving from a phase where AI can do almost anything we ordinarily require of students in terms of product. There is going to have to be another way to evaluate student understanding. But try having the new version of Gemini build you an app or something more creative that might be outside the scope of your ordinary capabilities and you can do some interesting things. But I agree with you, I've tried a few of these agentic browsers and I'm just not ready for them yet - I liken it to the analogy that I never really made the leap from single person video games to multi-player. There was just too much going on in the multi-player versions! I liked slowing down and being more methodical which is kind of what you're describing. My 25 year old son, of course, is completely the opposite. Ominous directions but I think it's going to continue to get even more complicated. If this feels like it's moving too fast for folks that have stayed on top of this, can you imagine what it will seem like to people who have been more or less in the dark for the past few years?
Thanks for your thoughtful comment. I think it boils down to discerning when friction is bad and when friction is good. For mundane drudgery, friction is bad, for learning ... well that's a different story.