Google Street View started as a project to scan the real world. Cars with cameras photographed streets to collect data. It worked, but everything that made a place feel like a place was gone. Until @andyszy prototyped what it could feel like as a product designed for people. The prototype was completely fake. The map didn't work and the imagery wasn't real. But it let people feel what it was like to stand on the street. An engineer saw it and said "let's ship it." Andy had to explain there was nothing to ship. This "smoke and mirrors" prototype convinced the team to throw out years of work and rebuild Street View through the eye of a human instead of a machine. I talked with Andy about the whole story, including the prototypes, the key design decisions, and how Street View became one of the most beloved products on the internet. Full story (with more early prototypes) in the replies 👇

I was an early designer on Google Maps. I still use it everyday. The Maps AI update is awesome b/c natural language has the potential to solve Maps' oldest limitations. And not so awesome b/c "Ask Maps" feels tacked on instead of deeply integrating AI into Maps. Here's why: — Google Maps was built around discrete searches. This works well for simple tasks like Where is… or How do I get to… "548 Market St." But many real-world navigation challenges aren’t simple queries. They involve intention and tradeoffs. My actual thinking is something more like: "I need to get to 548 Market St and find parking nearby. I’m happy to walk a block or two if it saves $20, but these shoes will likely give me blisters if I have to walk too far." Historically, I had to translate that into Maps’ UI with a series of separate searches like: "548 Market St" then "Parking" But by the time I see parking places, 548 Market St has disappeared. So I have to scan the results and pop back and forth between searches to find a solution. Natural language has the power to fix this. AI gives users the power to describe what’s actually in their head instead of translating it into sequential queries for the Maps UI. How amazing would it be to tell Maps: "I’m on my way to Tahoe with my family. I want to stop at an In-N-Out near a freeway exit and a fast charger. But I want to stop later in the trip since I have plenty of charge and nobody’s hungry yet. What are my best options?" It’s exactly these kinds of tradeoffs (distance, timing, price, convenience, etc.) that AI should excel at handling. — But Google isn’t revolutionizing Maps with an all-powerful, natural language input box. Instead, it cleared some pixels to add an “Ask Maps” button that serves as a portal into Gemini. So now Google Maps has two search boxes: 1. Classic Search 2. Ask Maps AI This makes the UI more complex. And it misses the opportunity to deeply integrate AI into Google Maps so that it feels native. Why not start with one, AI-fueled omni-box that can handle both: “548 Market St” and "I am heading to 548 Market St and running late. I need a place to park and get into the building ASAP!" — The promise of AI is to make things simpler. Giving people the ability to express the nuance in our heads is a first step. Hopefully the next step is using AI to augment and improve Maps’ core tasks natively. And for the steps after that, I really hope the team is dreaming big! I would love to see some of the bolder integrations that I’m sure the team explored internally.

A few months ago, I got a text: "State Farm: Friendly reminder that your State Farm payment is past due. Please pay online at st8farm..." And I was like yeah, whatever — I know a scam when I see one. I confirmed the check was cashed and I forgot about it. Until last week... I was going through the mail, and there was an urgent notice that our policy was about to be canceled for non-payment. I called State Farm. They showed no record of payment. So I sent them images of the cashed check. They opened a case and escalated to central payments. A few days passed and I called again. No update. Today (the official policy cancellation day), I call again. State Farm declares they never received nor cashed the check, so someone else must have cashed it. To keep the policy active, I need to pay again while they sort it out. So they send me a payment link and I pay it — not thinking twice this time. Then I call our bank. The agent tells me I’ll need to open a forged-endorsement case. To do that, they will send me a paper form in 7-10 business days. I ask if they can send an electronic version, but the agent says: “No. After all, you wrote a physical check.”  Touché. At this point my head is spinning, thinking about the absurdity of paper checks. But something isn’t sitting right. And then it hits me. How do I actually know the second payment I just made was to the real State Farm? Did I just fall for a different scam? Is this some elaborate scheme where the scammers have gotten relentless and weirdly sophisticated? I look back at the link “State Farm” sent and I start freaking out.The payment URL isn’t statefarm. It’s statefarmservice. Fuck. I start panic-searching, verifying URLs and emails. I call “State Farm” back, assuming they’ll ghost me. But no, they pick up. They try to reassure me. The agent I’m speaking with sounds as rattled as me. They aren’t sure how to convince me they are legit. I search back through my old emails. I’ve been exchanging emails with people from this State Farm office for years and all of their email addresses have the same weird 4 character code… At this same time, the supervisor calls me. She assures me where they are located and tells me who the different people are in the office that I’ve been speaking with. She sincerely apologizes for all the trouble. And finally, I exhale. Everything seems legit. Which means the problem isn’t just the scammers. It’s that in 2025, we have to be so vigilant with cyber-security that we can’t trust anyone. Because anyone can pose as someone else, it is nearly impossible to verify anyone’s identity — even when they’re actually real. Maybe the best strategy is to move from full blown paranoia to reasonable caution? But then again, a man posing as a delivery driver in SF just bound a homeowner with duct tape and stole $11M in crypto… For the moment, I’m setting aside my tin-foil hat and just trusting my gut that the State Farm people are who they say they are. But I may burn my last remaining checkbook while I wait out this 90-day forged-endorsement process... Stay tuned.