State of AI: Dictation and Conversation — Product Discovery Group

New ideas are the currency of Silicon Valley.

Most founders even deep into their latest venture can’t stop coming up with them. (think @boredelonmusk pre-2020)

Most of these ideas aren’t scalable, feasible, or monetizable.

In looking for an idea to use for this exploration of AI tools, I picked one of the not-better-than-my-day-job ideas that I regularly scratch on my whiteboard.

The Idea → Avoiding street cleaning parking tickets

Have you ever gotten a street cleaning parking ticket?

For urbanites, it happens enough that we think of it as a budget line item.

Last year, my household of 4 drivers got 5 of these tickets. At $97 each, that’s $485 real dollars and much more in dirty looks from my wife (who never seems to slip up).

I’ll use AI tools to develop an app to eliminate street cleaning tickets altogether (always dream big).

In addition, I’ll be using both Claude and chatGPT simultaneously to find their respective strengths and weaknesses.

Dictation and Conversation

Most of us are patterned to creating precise inputs in order to get valuable outputs.

For over a decade, Google has taken care of our spelling mistakes but we are still crafting just the right keywords.

With AI chatbots, we can skip the step of crafting keywords and just type what comes to mind in a stream of consciousness. These chatbots deal with spelling mistakes, grammar mistakes and don’t require as much forethought when we interact with them.

Given this flexibility, it’s a natural next step to use voice where most of us are not trained in concise dictation (like say a doctor would be).

So as I started thinking about my app I went straight to voice mode where I could ramble a bit and trust that the AI would organize my thoughts.

Brainstorming out loud is how I interact with my good friends when we discuss new ideas and using the AI in this way was a no-brainer.

I love starting a thread with AI using voice. I’m usually in the car and tired of my podcasts. I put chatGPT into voice mode and start having a conversation (footnote 1).

The conversational mode of chatGPT is still in its early days and can be awkward at times. The robot voice doesn’t always know when to chime in and it misses some short phrases (car noise?). But I’ve consistently used it for long conversations for 10-20 minutes at a time.

I’ve learned how to reverse sear a steak, plan gluten free meals for camping, which community banks offer business bank accounts, the subtleties of TAM-SAM-SOM and so on.

The voice user experience has a lot of potential.

Claude does not have a back and forth conversational voice mode yet (March 2025) so my starting point is typically chatGPT. If I use voice during this exploration, I’ll just copy and paste the transcribed prompt from chatGPT into Claude (using Claude’s latest model, 3.7, at this point) to keep them at the same level of context.

When using voice mode to describe the street cleaning ticket avoidance app, I tried to play dumb so the AI could fill in more but I couldn’t help myself and added lots of detail.

However, for my app idea, I quickly moved back into typing mode. Doing everything in the voice mode seemed clunky since I want to review the specifications that it was making before I gave it more prompts.

The voice mode shines most when the result is learning something, not producing a document.

Changing the AI Personality

To explore the corners, edges and depths of ideas, a common hack for AI conversations is to change the AI into a specific personality.

If you’re interviewing for a job, ask the AI to be the hiring manager.

“Ask me questions as if you were a Head of Product interviewing for a Sr. Product Manager”

Be sure to give it context in advance about the company, job, and other relevant info including uploading the job description, your resume and other related docs and website URLs.

If you’re looking to prepare for a meeting where your assumptions will be challenged, ask the AI to take the opposite approach to solving a problem and debate with you

“Take the fiscally conservative approach and respond to my ideas from that perspective.”

To prepare for board meetings or executive meetings, some advanced AI users ask the AI to create a group of personalities all with different perspectives to simulate the conversation and potential questions beforehand.

Side note: When using voice chat, it’s all transcribed into a chat in your chatGPT app for reviewing at a later time (footnote 2).

Here’s the transcript of the beginning of the voice conversation about my app idea (footnote 3):

1. At the time I was using the “o3-mini-high” model because it seemed like that was the right one (I’ve since just moved back to 4o, o1 and o1 pro…good luck keeping them straight). In the chatGPT app, I used a normal chat (not a project).

2. Note: I am not paid by any of these AI companies for this analysis.

3. Chat transcript:

Me: Okay, I'd like to create an iPhone app, and I'm going to tell you about that app. Does that sound good?

chatGPT: That sounds great! I’m all ears. What kind of app are you thinking about creating?

Me: This is an app that's going to help you keep track of where your cars are in San Francisco, and the idea is to avoid getting parking tickets due to street cleaning. Street cleaning happens on different time frames, on different streets, there's a data file or an API available from the San Francisco government that we'll need to get and process, and ideally we'd be able to, when the app opens, show where I am on the map with the car, and then be able to set that as a location, and then it would give me some sort of like countdown or timer, and some sort of ability to alert me with a push notification and or text, such that I have time to move the car, like an hour before, maybe the night before, because it might be early, like 6 a.m., and then I really want to be able to share this location with my family. There are two other people in my family that have iPhones, and ideally it would alert all of us if a car was still in a location that was having street cleaning, and the best feature of all would be able to read off of an air tag that is in the car, that I already have associated with my Apple Find My account, that's already shared to the people that I'm talking about. So the idea is to have street cleaning with data from San Francisco, with data from the air tag that's in the car, and an alert system with the goal of avoiding getting street cleaning parking tickets.

chatGPT: That sounds like a great app idea! Incorporating the San Francisco street cleaning schedule, map integration, notifications, and AirTag support could be super helpful. How far along are you with the development process?

Me: I've done nothing. So I'd like a plan for the app itself, the design, the coding, and how I would do the coding, and where the data is going to be stored, where I should launch. There's going to be some sort of back-end service. I don't really quite know what I need, so maybe you can help me with, like, all the things.

Next Article: Ideation and Specification

Jim coaches Product Management organizations in startups, growth stage companies and Fortune 100s.

He's a Silicon Valley founder with over two decades of experience including an IPO ($450 million) and a buyout ($168 million). These days, he coaches Product leaders and teams to find product-market fit and accelerate growth across a variety of industries and business models.

Jim graduated from Stanford University with a BS in Computer Science and currently lectures at University of California, Berkeley in Product Management.