Why I built Midnight
About fifteen years ago, I made a small scripted voice demo inspired by 2001: A Space Odyssey. The text-to-speech sounded pretty bad. HAL wasn't exactly a reassuring example of AI, but the idea of being able to talk that naturally stuck with me.
For a long time, the technology wasn't good enough. I stayed interested in voice interfaces, but only in the past year or so have they started to feel natural enough for the conversations I had in mind.
I wanted Midnight to understand the full picture, ask when something was missing, and push back when it disagreed. If I was preparing to speak with a customer, investor, or candidate, it could take the other side so I could practice important conversations first.
I started using Midnight to figure out Midnight. I kept coming back to how broad the product should be and who it was really for. Those conversations usually led me to remove something. A lot of ideas that sounded useful on their own didn't belong in this product.
Research grew out of those sessions too. A quick web search wasn't always enough. Before speaking with a founder, I wanted Midnight to pull together information from different sources about them and their company, with a short bio and a few talking points I could use in the conversation.
Practice took more work than I expected. Midnight could hold a role for a few turns, but it didn't reliably know when to stay in character or return as an advisor. A prompt wasn't enough; I had to build the structure around it.
I also got memory wrong at first. My instinct was to remember as much as possible, but old information could send a new conversation in an irrelevant direction. I cut it back. Midnight saves conversation briefs, and I choose when to return to earlier work.
As a designer, I've enjoyed giving Midnight its own atmosphere. The identity draws on transitions in the night sky, with animations that respond as you talk and listen. The name comes partly from that interest in space and partly from the private feeling of talking something through later in the day.
I've spent an unreasonable amount of time on some of those details. In client work, there's a budget, a deadline, and a point where you have to move on. With my own product, I can keep going when a tiny detail still bothers me. I'm not claiming every extra hour was sensible, but I enjoyed being able to care that much.
Some of my best sessions happen while I'm walking, with my phone in my pocket. I can talk through an idea without first turning it into a document.
Fifteen years after that scripted demo, I can finally use the thing I wanted then.