While working on this article through voice conversations with AI, I noticed that our exchanges were starting to feel like requesting work and receiving reports. I explain an idea and ask for something to be done. I read what comes back and check it myself. Then, based on what I find, I ask for the next piece of work.
The process feels similar to asking another person to do a job. One side makes a request, the other does the work, and the results come back for review before the next step is decided. Voice conversations and on-screen review have made that back-and-forth possible in my own work with AI.
This article took shape through repeated requests and reports
I started by talking about how voice makes it easier to share intuitions and concerns I haven't yet sorted out. When I type, my attention is divided between entering text and editing it. When I speak, I find it easier to concentrate on the conversation. I asked AI to turn that experience into an article.
But when I read the draft, my reaction was essentially, "Okay, I see." It offered another way of looking at things, but that was about as far as it went. The explanation made sense, yet it hadn't quite reached what I wanted to say.
So I talked through what felt missing. I find it easier to react candidly when the other party is AI. Receiving information and perspectives I didn't already have, and responding to them, can help make my thinking more concrete. I added these points as we talked, asked for a revision, and read it again.
During those exchanges, we arrived at a possible pattern: make a request by voice, read the report that comes back, decide what to do, and make the next request by voice. I realized how closely that resembled asking someone to do a job and reviewing the results. The process of writing this article had become an example of the idea itself.
Speaking helps me work out what to ask for next
Reading the result doesn't always give me a clear revision request straight away. Sometimes I start with a reaction like, "It isn't wrong, but something is missing." Being able to say that aloud, before I've organized it, helps.
When I prepare a text message, I organize my thoughts, type them, adjust the wording, and check what I've written. Along the way, I can end up editing out my initial intuition or unease. Speaking lets me pause halfway through a thought, rephrase it, and explore what is bothering me. I find it easier to put the attention I would have spent typing and editing into thinking about the substance.
Having AI on the other side also makes it easier to keep this exchange going. Being too blunt with another person can damage the relationship. With AI, I have less of that interpersonal concern, so it is easier to say, "I'm still not convinced." I can work through that reaction in conversation and turn it into the next request.
Those requests can go beyond rewriting. I can ask for relevant research or a different perspective on the issue. On my own, I tend to frame questions within the limits of what I already know. When AI brings back new material, I want to read it and include both what makes sense and what raises questions in my next request.
This back-and-forth also seems useful for exploratory work, when I don't yet know exactly what needs investigating. AI can report interim findings and unresolved questions, and we can discuss what to investigate next or how to proceed. I want to leave room for uncertainty and changes of mind along the way. It feels like working out the request itself through discussion.
The person making the request reviews the report and decides
After I ask for work to be done, AI returns the results. What I need at that point is a clear view of what has been produced, what has been learned, and what needs checking. Reading the report becomes the time to decide how to move forward.
For an article revision, it helps to have the draft itself along with an explanation of what changed, why it changed, and what still needs review. For research, I want to check the sources and dates behind the conclusions, as well as the conditions under which those conclusions apply.
I discuss how to organize these reports in this article. When the conclusions, supporting evidence, unknowns, and decisions ahead are easy to find, it is easier to choose whether to request more research, ask for a revision, or consider the work complete.
Having the material on screen lets me read at my own pace and return to anything that needs a closer look. Even when a spoken explanation sounds convincing, I still need to check the information and whether the result serves the purpose of my request. AI may carry out the work, but I decide whether to use the result and what to ask for next.
Once I've read it, I can communicate that decision by voice. The report becomes the starting point for the next request.
Making requests by voice and reviewing results on screen
My own work with AI may increasingly revolve around speaking to make requests or discuss ideas, then receiving detailed results on screen. I will still use text when I need to communicate precise conditions or numbers, or when my thoughts are already organized. I want to choose the medium that suits what I need to communicate and what I need to check.
I make a request. AI does the work. I read the report, check it, and decide what comes next. Then I make another request. Rereading this article and considering the next revision is part of that same process. Looking at my work with AI this way makes it clearer what I need to think through, what I can ask AI to do, and where I need to make a decision.
Top comments (0)