From the Desk of
Summoning Software
At a house party a few years back in an unfamiliar part of the Bay Area, I was talking to a seventy-something-year-old dentist about his favorite Italian restaurant on the other side of town. Midway through the recommendation he launched into turn-by-turn directions— take this road, turn right there, left at the light— and I nodded along, pretending I’d remember, though it had been like a decade since I’d last tried to commit directions to memory and I didn’t know the area well enough to integrate the instructions into my mental map, but he was older, and kind, and in that moment he was a human time machine taking me back to the twentieth century.
Nowadays nobody knows where anything is except the Google Maps icon on your Home Screen. We tap it, type where to, and blindly follow the turn-by-turn without ever looking at the whole route because who even knows the nuances of routing anymore?! But for most of the twentieth century, everyone who drove a car1 knew how to get around town to some degree. And at the extreme end, professional drivers weren’t just people with yellow cars and an uncanny openness to interacting with strangers in need of a change of scenery, they were people with an otherworldly working knowledge of the local environs that let them move you from anywhere to everywhere.
That knowledge was the barrier to entry— most famously, London’s Knowledge test required aspiring cabbies to memorize twenty-five thousand streets and thousands of landmarks before they could get a license— and it meant not everyone who wanted to earn money driving could. It’s strange to think about now because everyone can go anywhere with Google Maps, but for about a century professionals with that knowledge were in such tight supply that most people only rarely paid someone else to drive them anywhere.
It started to change slowly, then all at once. The Department of Defense started launching satellites into orbit for its global positioning system in 1978, by 1993 it was globally operational, and finally in 2000, President Clinton unlocked GPS for civilians. But it wasn’t until iPhone connected those satellites to the internet— and all from the palm of our hands— that Uber could simultaneously transform the transportation industry and reshape the world for everyone who wants to be somewhere they aren’t.
At this point in history like millions of words have been written about Uber as revolution for consumers: summoning rides to anywhere, high resolution ETAs, cashless payment, the transformation of cities and nightlife and the way we think about getting places. Less has been said about how completely Uber reshaped the transportation industry itself— the hiring practices, the staffing plans, the careful balance of supply and demand that taxi companies had spent decades refining. Imagine telling them in 2008 that the market for paid rides was about to grow 10X larger …and they were going to end up with a smaller slice of it.
The transportation incumbents lost share because they didn’t perceive GPS as an improvement— they saw it as a crutch for amateurs because in the beginning, it was: for the first decade or so, a veteran cabbie’s mental map was faster, integrated traffic knowledge, and unlocked shortcuts. In yet another classic example of the innovator’s dilemma, the new technology looked like a toy to the people whose skills it would eventually replace.
Smartphones with GPS didn’t just let people who would’ve failed the taxi company hiring tests get around town, they obviated almost all of the value that professional drivers provided in the first place. Suddenly anyone with a car and an iPhone could substitute for a professional driver, and that meant a massive explosion in the supply of people who could do the work.
Uber could have commercialized that insight by selling on-demand surge capacity to the taxi companies: like, more drivers on Friday and Saturday nights, holidays, overflow capacity whenever the professionals weren’t keeping up with demand. And that might’ve grown the industry by a modest percentage, and it would’ve left the lion’s share of the value on the doorsteps of incumbents.
OK here let me save you two years of tuition at Harvard Business School: people don’t want a drill, they want a hole in the wall. And likewise, people don’t want a taxicab, they want to be somewhere they aren’t. Uber as a consumer product goes far beyond the driver labor market revolution, it transformed the multi-act drama of wondering/wanting/hailing a taxi into— to the extent current technology allows— a teleportation device: tap some buttons and you’re on the way there.
But this isn’t an essay about the history of rideshare; it’s an essay about the future of software. Because just like GPS and smartphones radically reduced entry requirements for professional drivers, LLMs are blowing away the specialized training and specific knowledge required to produce software.
So far2, the transformation is more evolution than revolution. Today’s vibe coders are, in my mind, analogous to the folks in the 20th century who applied to be taxi drivers but got rejected for not being able to memorize the map, now equipped with smartphone GPS. Everywhere we look, we’re flooded with stories about people creating software without the previously requisite engineering skills —and most of these stories have predictably horrific endings. And, all the vibe coding platforms make the same grand promises about being able to submit your app to the App Store in a copy-and-paste analog to that B2B version of Uber we daydreamed about together above.
Uber was a revolutionary for its transformation of consumer expectations around the availability of drivers. Before Uber, you knew cabs would only be in certain parts of certain cities— so you didn’t even try to hail one on a random street corner in Brooklyn or Des Moines or wherever —and your expectations about supply guided both your demand choices and the perceptions of consumer demand growth potential inside the incumbent transportation businesses.
Now we’ve seen a few years of that same transformation on consumer expectations happening as ChatGPT rapidly replaces Google as the knowledge engine folks start searches with. Pre- generative AI, you knew Google could only serve you things in its index— like a physical store can only sell you the goods on its shelves— and because you’ve shopped Google so many thousands of times before, you’ve got an internal sense of what kinds of queries will return useful results akin to knowing what kinds of produce your grocer stocks (and finding that ube flour’s going to mean a special trip to the ethnic grocery via Google Maps). You know better than to search Google for a cocktail pairing for the recipe you’re cooking tonight, but ChatGPT’s generative-ness its independence from an indexed inventory, inspires you to ask it things you don’t expect anyone to have already written about.
The future of software isn’t millions of apps made by people with a vague inkling that they should be successful entrepreneurs while lacking the chutzpah to dig in and learn the skills to engineer useful, reliable, delightful systems. The revolution comes when consumers can summon software like they summon an Uber; when they stop intuitively restricting their demand for software based on their expectations and experiences with what the software market has provided to date.
Maybe the truth is simply that the technology just isn’t there yet. Today’s vibe coding tools take minutes and cost real money per iteration— maybe it’s just the GPS without the smartphone bundle— and they’re so much slower than searching the App Store for something extant and close enough that for an end user it’s an outrageous imposition. Who really wants to make holes in the wall…
So who’s the customer for this technology right now? Not App Store customers, clearly, but the poor rubes who …but for the unfairness of the difficulty of learning to code3… want to skip the Knowledge test and go straight to the App Store riches. Uber could have been just a provider of marginal4 labor for the incumbent transportation businesses, allowing them to scoop up an extra few percent of revenue on Friday and Saturday nights, and likewise the vibe coding platforms popping up every day5 appear to have a ceiling of producing merely the Newest Worst App in the App Store.
The revolutionary promise of generative AI is dropping the requirement that software be created by someone, and in advance of someone wanting to use it. We’re going from software as an asset that is inventoried to one that is produced just in time, dynamically, in response to the user’s query. Just like Uber lets you summon a ride to your location that takes you to exactly where you’d rather be, the future of the App Store is software on demand: summoned to solve exactly the problem you have, then disappearing, without all the bullshit like a marketing newsletter and an onboarding flow and billing crap that exists because today every app is trying to be an entire business.
We don’t want an App Store stuffed with more junk, we just want a hole in the wall.