Since getting back to Colorado I have been meeting with a lot of family and friends of friends. I love finding shared ground with people, and learning about what interests people, but it is extremely hard for me to stay up to date all the goings on of all my friends hobbies/interests.
Example: I know my uncle-in-law might be really into Colorado Nuggets Basketball, reading sci-fi books, and going to new Restaurants. However I don’t know how the nuggets are doing, what sci fi books are out recently, and if Denver has had any new restaurants open. Yes I could ask AI questions for all the hobbies/interests my friends have, but I want to learn some AI and this feels like a good topic to build something about. SO I am going to build a little utility so when I am sitting on the couch next to that uncle I can steer the conversation AWAY from politics to a safer topic, like AI taking over the world. God speed to me.
That is where Serving Up Some Conversation Starters comes in.
First step for me was to get the rough idea from my mind onto the browser window. I have been using Google AI tooling lately, so I fired up Google CLI and built out the basic prompt. I pondered building a RAG pipeline to help me get recent events, but after experimenting with the results I found myself satisfied with Google’s “Grounding” feature. From what I understand “Grounding” just allows the API to do searches. As a designer I feel like this differentiation is above my paygrade, and if I did understand, it would only make me frustrated when people misuse the words. A few rounds of iterations and experiments later I found something that worked okay for my use case.
Some sample outputs
Have you been following the news about Denver’s recent moratorium on new data centers, announced by Mayor Johnston on February 23, 2026?
With the Avalanche gearing up for the Stanley Cup Playoffs, which begin on April 18, have you heard about Cale Makar’s upper-body injury from the game on Monday, March 30, 2026? Hoping he’s back on the ice soon for their playoff run!
Next came the design step. Since we are trying the Google’s AI stack when ever we can I will try a new tool Google Stitch.
It wasn’t too bad to get the tooling setup. The MCP in Gemini CLI only had a few hiccups. A feature I heard Stitch could do is conversations from code to design and back, so that was what I was excited to try! Boy was I was surprised, but in hindsight, I don’t know why. Google Stitch felt very comfortable taking liberties on initial import to add that SaaS marketing page sheen to everything. As a snowflake designer of course my feelings were crushed when it messed with my design, but it altering my content felt presumptuous. It completely transformed the “hackathon grunge” design I had. Shame on me for having expectations. The slapped together look of simple boxes and containers will always give me those nostalgic vibes from my early career. The AI generated SaaS has a disingenuousness aroma that can be smelt from a while away. Doing design work in stitch isn’t as comfy as Figma, but I am guessing if I spend the time to learn it, I would be okay. The process of bringing it back into code was equally as non-deterministic as the other direction. I know it is a normal designer trope to complain that things weren’t implemented like the Figma file, this is that at the next level. That’s AI baby!
My favorite anecdote of the switch back was Google Stitch insisting we use tailwind. In my limited time as a software developer I never understood why tailwind was popular, it seemed like a worse abstraction for CSS, but everyone around me liked it, so I used it and felt competent. I deeply enjoy that AI is trying to convince me to integrate tailwind into my hobby projects.
My takeaways:
- This workflow is mid - I don’t think this is “the way”. I would like to try a very similar experiment with Claude Code + Figma, and see if it does better. But for now I think its better to carefully create a design/pattern library, then use an agentic IDE/CLI. When talking to the agent act as a hyper specific art director. It always tested my patience when someone stood over my shoulder saying “could we try rounding the corners 1 pixel less”. When the AI takes over, please know I am sorry and I empathize with your situation. It was cruel what I did, but we all need start somewhere, and I had to start there too.
- Always double check your .env file before committing. No project would be complete if I didn’t accidently dump my API keys to GitHub, luckily I caught it with in 30 seconds. Welp, damage done, I have deleted the key, now to replace everywhere its used.