For the past few weeks, I’ve been down a rabbit hole. The mission? A personal knowledge project that had me relearning Linux, wrangling Python environments, and generally feeling like a tech newbie again. But this time I wasn't alone. My project partner in this venture was Google's latest Gemini 2.5 Pro (I realise there are many other powerful ones, but amazingly this one via AI Studio is currently available at no extra cost if you have a gmail account), and our collaboration was one of the most fun learning experiences I've ever had.
We've all gotten used to bossing AI around. "Summarise this." "Write that email." "Tell me a joke." (Q. Why do programmers prefer dark mode? A. Because light attracts bugs!) Groan.
But what happens if you stop treating AI like an intern and start treating it more like a helpful colleague? Things can get interesting.
I found that when we hit dead ends together we could re-architect new solutions by riffing off each other's ideas. At times the AI would get confused, and I learned to restart our sessions when our chat history got ridiculously long (over 120,000 tokens, for the curious - it doesn't take as long as you think to hit those limits!).
At one point, while I was waiting for a data process to finish, the AI said to me: "Yes, absolutely.
This is the perfect time to work in parallel. The ingestion process is I/O and CPU/GPU bound, but our development work is human-brain-bound." "Human-brain-bound."
Cheeky AI blighter! I scoffed and laughed out loud. But...maybe it was right. Partly in retribution for its comment, I then spun up three different project development sessions at once to see if my AI collaborator could keep up. Ha - take that! Despite its immense power, it humorously struggled to keep up with all three, proving that even advanced AI has its own "Oh, wait" moments. The little mistakes it made along the way were annoying and amusing at the same time - one of my favourite instances was when it apologised for a mistake by saying, "Ah sorry— that’s one of my classic copy-paste errors—I forgot a parenthesis." Wait. What? Really?
This experience taught me a few things about moving beyond simple prompts:
- Frame AI up as an idea machine, not an answer machine: When you treat AI as some type of intellectual colleague, you will find it really can start to volunteer ideas. I was consistently surprised by the novel approaches it suggested, pushing my project in directions I hadn't considered. In turn, it riffed off my ideas - good fun. And amazingly rapid.
- Have a dialogue, not a monologue: Instead of just taking orders, the AI started asking me questions, helping to refine the project's scope and ensuring we were on the right track. It felt less like a tool and more like a true co-designer.
- It gets better if you work on a long-running theme together: Just like with a human colleague, the more we "worked" together, the better the AI understood my goals and tailored its suggestions. It was reliable at updating my goals with our joint project phases so we never lost track and I could easily resume after a new chat session (after it got "tired" - i.e. started to perform poorly). And those three separate projects? At my suggestion it integrated them back into one - but for sure it did the heavy lifting of that code refactoring!
Shifting from simple answer prompting to interactive collegiate partnering doesn't just make work more productive; it makes it more fun. It turns problem-solving into a dynamic, creative dance between human ingenuity and artificial intelligence.
But rest assured I won't be forgetting the movies and for sure I will NOT be treating Gemini Pro like a human partner. I'm not tempted to go down a HER rabbit hole. And partly for this reason I don't use the voice interface at all for my project work. I need the distancing that the keyboard gives me, and maybe it's me, but I find the keyboard interface more natural for "computer" types of projects.
So...I suggest having a go at changing how you prompt to make it more of an interactive dialogue and try it out on a project idea - the results might just surprise—and delight—you.
PS - it is not perfect - every session generated errors which had to be debugged. But it was good at that, and I found the personal learning and satisfaction with progressing my project rapidly made up for that. What's been your experience?
Leave a Comment