Using Generative AI Tools for Faster Prototyping

Product Managers: It will save you time.

I’m not convinced generative AI is as revolutionary as it’s made out to be. I am also not Ed Zitron. I am someone who believes that you have to keep learning to stay competitive. That’s why I value curiosity as the top trait in product people - (designers and engineers included). If you have a gremlin inside you that always asks why, you’re going to find yourself on a growth path.

I’m always looking for ways to execute faster. I code for data analysis and visualization, not to make software. As a non-coding technical product manager, I have wanted to quickly spin up working prototypes and proof of concepts. Before generative AI, I created wireframes. If I had more time, I’d make a basic interactive prototype. Wireframes don’t provide proof of concept - they only kickstart the design process. Prototypes take too long to make and won’t cover cases where designing in code can. Working on niche components and experiences amplifies my need to design in code due to the lack of design patterns.

This is where LLMs and gen AI tools have saved me time. However, this will only work well on small solutions.

Using Claude or ChatGPT, I can pound my keyboard with a stream of consciousness flow, telling it what I want to do in the order of capabilities that comes to mind. Then, I’ll ask it to analyze, organize, and transform my thoughts into specs and requirements for an engineer. Through a back and forth, I can get the requirements I need in 10 minutes.

Spelling Bee App created with bolt.new. Integrates with a free text to speech API to play the words to the user. Integrates with the free dictionary API to show definitions and the word used in a sentence.

I take those requirements and give them to a gen AI coding tool (really just a Claude or ChatGPT wrapper with additional functionality) to create my prototype. If your requirements are precise, you’ll get 90% of what you want within 15 minutes. This wasn’t always the case. Since mid 2023, I’ve used this same workflow. The coding part would still take me hours, which I blame on the model’s capability and not my prompting skills. But this isn’t the case anymore. I use this workflow at work and on my side projects. In the last 2 weeks, I created a spelling practice app for my kids to help them win their school spelling bee. For work, I wanted a proof of concept of a standalone American football pass location collection app. I’ve tried this before with some success but nowhere near what I was able to make today.

Unless you can design in code, use this workflow to get from fuzzy idea to proof of concept. If you don’t, others will.

Coding tools used can be found at the bottom in the References section.

American Football Pass Location Data Collection App. Can collect data for as many quarterbacks as you wish. Visualized pass chart can be exported to jpeg.

Data table listing the passes collected location data with a dropdown to collect the status of the pass. User can also delete the pass location if they made an error.

References

• Made the spelling bee app with bolt.new

• Made the pass location collection app with Lovable