What Tom Scott Knew Before the Rest of Us

On January 2nd, 2024, I was sitting in the backseat of a car heading east through Slovenia with the worst hangover of my young year — and it was only the second day.

We’d celebrated New Year’s Eve in Ljubljana, which had been great. But the evening of January 1st turned out to be the real damage. It started slow — our group wandering through the quiet post-holiday streets — until we found this cozy bar with a tablet on the table that let you queue songs on the bar’s Spotify. Two-for-one Long Island Iced Teas. Two Australian guys, younger than us, who we ended up adopting for the night and teaching our German drinking rituals to. The kind of evening that starts with “just one more” and ends with a blackout.

By mid-morning on the 2nd, squinting at the Slovenian countryside from the passenger seat while three friends debated which road to take for our little post-holiday retreat, I knew with absolute certainty: the biggest hangover of 2024 had already happened. Nowhere to go but up.

I did what I always do when I’m hungover and have no obligations — I opened YouTube. The kind of passive, guilt-free scrolling that only feels earned after you’ve truly destroyed yourself the night before.

The first video in my feed was Tom Scott’s farewell. The one where he hangs from a helicopter and flies into the horizon to swelling strings, the kind of outro that tells you this isn’t a break, it’s a goodbye. I don’t know if the hangover had left me more emotionally porous than usual, or if I’d genuinely built up that much of a connection to someone I’d never met through years of ten-minute videos — but I teared up. Sitting in that backseat, sunglasses on, mountains drifting past, a quiet tear running down my cheek for a YouTuber. Not my finest moment. Maybe one of my most honest ones.

Still riding the serotonin afterglow of the trip, I decided to post something about Tom Scott on Instagram — something I don’t do all that often. Not the farewell video, though. That felt too obvious, too much like bandwagoning on the collective goodbye. No, I thought of my actual favorite video of his. Even if it happens to be his most-watched one — not exactly an underground pick, I know.

This Video Has X Views.

Most people know it for the self-updating title. The comments are almost entirely about whether the code still works. But sitting in that car, still slightly drunk and unreasonably emotional, I knew exactly which part of that video had always hit me hardest. Not the clever hack. Not the API archaeology. The ending.

Tom Scott, standing on the White Cliffs of Dover, talking about chalk and entropy and the end of all things:

“The White Cliffs of Dover are a symbol of Britain, they are this imposing barrier, but they’re just chalk. Time and tide will wash them away, a long time in the future. This, too, shall pass. But that doesn’t mean you shouldn’t build things anyway. Just because something is going to break in the end, doesn’t mean that it can’t have an effect that lasts into the future. Joy. Wonder. Laughter. Hope. The world can be better because of what you built in the past. They don’t have to be big projects, they might just have an audience of one.”

An audience of one. I kept turning that phrase over in my head as Slovenia scrolled past the window. Because two years later, that line describes something Tom Scott couldn’t have predicted when he filmed it in 2020 — but something that has quietly become the defining reality of software in 2025 and 2026.

The Prophecy Nobody Noticed

Tom Scott filmed that video in April 2020. The world was in lockdown, and the open web he’d grown up tinkering with was dying. Not from neglect — from enclosure.

The middle section of the video is a quiet eulogy for Web 2.0 — the era of open APIs, mashups, and tools like Yahoo Pipes that let anyone stitch services together without asking permission. Scott walks through how he originally built the self-updating title hack using the open YouTube API, and how that kind of playful tinkering became harder and harder as platforms locked down. Facebook killed its open graph. Google shut down APIs with little warning. Twitter went from the most developer-friendly platform on the internet to one of the most hostile. The web didn’t break — it got walled in.

What’s ironic is that the thing Tom Scott was mourning — the ability for regular people to build things with code without gatekeepers — is exactly what AI coding tools are now bringing back. Not through open APIs, but by making the code itself almost free. The walled gardens are still there. But now you can build your own garden in an afternoon.

Scott couldn’t have known that in 2020. He was describing a mindset that AI would later make accessible to millions of people. The idea that code is temporary, that the act of building matters more than the artifact, that your project doesn’t need to serve a million users to be worth making — these weren’t AI talking points. They were a YouTuber’s philosophy, delivered from a chalk cliff in Kent.

Five years later, they read like a manifesto for an entire movement.

”Ninety Percent”

In March 2025, Dario Amodei — CEO of Anthropic, the company behind Claude — sat down at the Council on Foreign Relations and made a claim that lit the internet on fire. Within three to six months, he said, AI would be writing ninety percent of code. Within twelve months, essentially all of it.

The reaction was swift and derisive. An Anthropic employee publicly disagreed with his own CEO, estimating the real internal number at closer to fifty percent and calling the prediction “overly aggressive.” IBM’s CEO offered a more conservative twenty to thirty. Tech journalists marked the date on their calendars to revisit in six months, sharpening their knives.

Six months later, the fact-checkers came back — and they were partially right. Most developers weren’t having ninety percent of their code written by AI. One study showed experienced programmers sometimes got slower with AI assistance, not faster. The skeptics had their victory lap.

But then something interesting happened. Instead of the prediction quietly dying, it started becoming true — just not in the way anyone expected. John Gruber, the notoriously skeptical Apple commentator, revisited Amodei’s claim a year later and arrived at a surprising conclusion: Amodei wasn’t wrong about the percentage. He was wrong about the mechanism. The ninety percent didn’t come from AI replacing human programmers. It came from AI enabling an explosion of code that would never have been written at all.

The total amount of human-written code hadn’t meaningfully decreased. But the total amount of all code — human plus AI — had multiplied by an order of magnitude. The pie isn’t the same size with different slices. The pie is ten times bigger. And most of the new slices are things no human would have bothered to bake — a custom expense tracker for a five-person team, a one-off data migration script, a weekend app that scratches exactly one itch for exactly one person.

An audience of one.

The Vibe Shift

On February 2nd, 2025 — almost exactly five years after Tom Scott’s video — Andrej Karpathy posted a tweet that would become its own kind of cultural marker. The former Tesla AI director described a new way of programming he called “vibe coding”: you talk to an AI in plain language, accept whatever code it generates without reading it, and paste error messages back in without comment. The code grows beyond your comprehension, and you don’t care, because it works. By year’s end, Collins Dictionary had named it their Word of the Year.

And then something genuinely new started happening: people who had never written a line of code began building real software. A growth marketer built a safety app that reached ten thousand users. A corporate marketing team saved six figures in software costs by building their own internal tools. Y Combinator reported that a quarter of their latest startup batch had codebases that were almost entirely AI-generated.

Simon Willison captured the optimistic case perfectly: everyone deserves the ability to automate tedious tasks with computers. You shouldn’t need a CS degree for that. But he also drew a sharp line — if you review the code, test it, and can explain how it works, that’s not vibe coding, that’s software development. What’s fascinating is that both sides of that line are expanding simultaneously. Professional developers ship faster with AI as a collaborator. Non-developers build things that were impossible for them a year ago. The same tools are transforming an existing craft and creating an entirely new one.

I see this in my own life. I’m building a court booking and team organization app for a small tennis club — the kind of thing that would normally mean an overpriced SaaS subscription or a shared spreadsheet that nobody updates. With AI-assisted coding, I had a working prototype in days. The club has maybe two-hundred members. The audience is tiny. But the courts get booked, the team lineup gets sorted, and I learned something building it. Tom Scott would approve.

Disposable Code and Permanent Value

There’s a philosophical shift happening underneath all of this that goes deeper than productivity numbers. It’s about what code is.

For decades, code was treated as an asset. You maintained it, refactored it, documented it, handed it off to the next team. Entire methodologies — clean architecture, SOLID principles, domain-driven design — existed to make code more durable. But when the cost of writing code approaches zero, durability stops being the primary virtue.

One developer described it as software that fits you like a craftsman’s tool fits their hands — custom, personal, maybe not built to outlast you, but built to serve you right now. A Google Cloud engineer suggested the right mindset isn’t “build to last” but “build to evolve.” Others coined the idea of “disposable software” — code that exists to test an idea, prove a concept, solve a Tuesday afternoon problem, and then get thrown away without guilt. This sounds radical if you’ve spent twenty years learning to write maintainable code. But it’s also exactly what Tom Scott said on that cliff in 2020. The code was never the important part.

There’s a beautiful tension in this moment. Some experienced developers report that the joy has drained from their craft — one described feeling like he’s on a factory production line, checking AI output, next, next, next. A Finnish developer half-jokingly proposed “luomukoodi” — organic code, written by hand without AI, as a meditative practice. But at the same time, the people gaining access to coding for the first time are experiencing the exact joy the veterans are mourning. As one essay put it: the same tools that are draining joy from the craftsmen are creating it for people who never had access to the craft in the first place.

Entropy and creation, happening simultaneously. The cliff eroding while someone builds a sandcastle at its base.

Builder

There’s a word gaining traction that I think captures this moment better than “developer” or “engineer” or “vibe coder.” The word is builder.

Boris Cherny, who created Claude Code at Anthropic, said it plainly: we’re going to see the title of “software engineer” start to go away. Sal Khan of Khan Academy predicted that the people who thrive won’t be the ones waiting for specs — they’ll be the ones who meet the customer and build the solution themselves. Across startups and big tech companies, employees are already swapping their titles.

I find “builder” compelling because it’s honest about what’s changed. A builder doesn’t need to understand every line of code. A builder needs to understand the problem, envision a solution, and use whatever tools are available to make it real. Sometimes that means writing code by hand. Sometimes it means talking to an AI. The output matters. The method doesn’t.

And if that sounds like it diminishes the skill of software engineering — I don’t think it does. It elevates it. When the mechanical act of writing code becomes cheap, the expensive skills are taste, judgment, architecture, empathy for users, and the ability to know what’s worth building in the first place. The craft moves up the stack. From syntax to systems. From systems to outcomes.

The Code Was Never the Important Part

I still think about that car ride through Slovenia. The hangover, the mountains, the quiet tear for a YouTuber I’d never met. In retrospect, I wasn’t crying because Tom Scott was leaving YouTube. I was crying because he’d articulated something I’d always felt but never put into words — the bittersweet reality of building things that won’t last, and why you should do it anyway.

The tears dried somewhere in eastern Slovenia. Tom Scott’s retirement didn’t hold either — he came back last week, and uploaded his first real video again just three hours ago. Apparently even farewells are subject to entropy, just in reverse. Nothing lasts. Not even goodbyes.

Two years later, I use AI to write code every single day. My workflow has changed more in the past year than in the previous decade. I build more, ship faster, experiment with ideas I would have dismissed as too small or too niche before. Some of it is professional. Some of it has an audience of exactly one.

And none of it will last forever. The APIs will change. The models will evolve. The tools I rely on today will be deprecated, acquired, or abandoned. Entropy will get us all in the end.

Tom Scott knew this when he stood on those chalk cliffs and pointed his camera at the horizon. He knew the code updating his video title would break someday. And he said the thing that, five years and an AI revolution later, still feels like the truest statement anyone has made about software:

“And yes, at some point, the code that’s updating the title of this video will break. Maybe I’ll fix it. Maybe I won’t. But that code was never the important part.”

Build things anyway.