I remember being at a workshop a while back and watching someone interact with an AI agent using voice-to-text.
It was impressive. They didn't stutter or pause; they just spoke a perfectly articulated, well-laid-out paragraph. They gave the AI the context, the constraints, and the goal in one smooth take. It seemed effortless for them, and the result they got back was exactly what they needed.
I went home that night fired up to try it myself. I opened the app, hit the microphone button, and… it just felt clunky. I found myself speaking slower than usual, trying to "format" my thoughts before I said them. It wasn't quite the smooth experience I had witnessed earlier.
But the expert at the workshop had sworn by the results, so I stuck with it.
It took a few sessions, but the awkwardness faded pretty quickly. I realised that the person I saw wasn't using a magic script. They were just explaining what they wanted, the same way they would to a colleague.
From "Prompting" to Coaching
Once I got past that initial bit of friction, I realised why this matters. When you type, you naturally try to be brief. You cut corners. But when you speak, you provide context. You tell the whole story.
At Plain Logic, this has become a core part of how I approach technical problems. For example, I was recently vetting some infrastructure options (Docker and containerization). I was using AI as a sounding board, and it gave me a standard, textbook answer about "reliability" impacts.
Because I was in a conversation mode rather than just reading text, I naturally pushed back. I asked it to explain why. I kept asking questions, digging into the logic, effectively coaching it until it gave me a breakdown that actually made sense for the specific project economics.
The Takeaway
It feels a little weird to talk to a machine at first. Most people stop because of that initial friction.
But the trade-off is huge. By getting over that small hump, you stop getting generic "outputs" and start having high-context problem-solving sessions. It allows me to build solutions that aren't just technically effective, but are actually logical for the specific problem we are solving.