Three months ago, engineers all over Silicon Valley unwrapped what felt like a present designed just for them: a new version of Claude Code, Anthropic’s AI coding tool. Developers spent their holiday breaks experimenting with it, marveling at its power. They jokingly called it “Claude Christmas.”
But the giddiness didn’t last. Many emerged from the holidays deeply unsettled, having watched the tool, which was released Nov. 24, autonomously build projects they would have spent weeks coding by hand. For some, the breakthrough confirmed a darker fear: that they might soon be relegated to what tech workers have taken to calling the “permanent underclass.” In San Francisco and San Mateo counties, where around 190,000 jobs are tied to tech, those concerns have hit especially hard.
The anxiety spilled into public view last week when an essay (opens in new tab) by an AI CEO went viral, arguing that tech workers have spent the past year watching AI surpass them at their jobs, and that other white-collar workers are about to experience the same thing.
It used to be that coders spent around 20% of their time designing and 80% writing code, said Daivik Goel, an engineer working on his own startup. “But now it’s rare that you write any code at all.” Some engineers are even plotting what their post-software lives will look like.
Coding is especially vulnerable to automation because the work is digital. Unlike jobs that require physical presence or navigating human relationships, software can be written, tested, and improved entirely by machine. AI has an enormous advantage: It can learn from billions of lines of code that programmers have made publicly available over decades. Engineers have, in effect, built the perfect training ground for their own replacement.
What makes this moment different from earlier waves of automation is that AI does more than simply accelerate the work. Previous upgrades, from better programming languages to cloud infrastructure, required engineers to design the systems they were building. But today’s coding agents show signs of having “ideas” — they can propose architectures, follow their own road maps, and execute sophisticated projects with little — or no — human guidance.
“If suddenly we have a machine that’s able to do all the things that society thought you were valuable for, that’s very existentially upsetting,” said James O’Brien, a computer science professor at UC Berkeley.
Some believe engineers are just the first sector to feel the pain, with customer service reps, legal aides, and consultants next in line. Exactly how, and on what timeline, automation plays out across other professions is up for debate. Anthropic CEO Dario Amodei warned (opens in new tab) last year that AI could wipe out half of all entry-level white-collar jobs in one to five years. Verizon CEO Dan Schulman recently floated (opens in new tab) the possibility of overall unemployment hitting 20%, or even 30%, in two to five years.
Some experts are skeptical about these timelines. They argue that when the barrier to building software drops, more software gets built — expanding the overall market and creating more jobs. And those who can masterfully deploy agents will find themselves in even higher demand.
“For the engineers who can get the most out of these tools, it’s like giving them a nuclear-powered six-axis mill,” said Lee Edwards, an investor at Root Ventures and a software engineer by training. “It’s a single-person software factory.”
But in San Francisco, many rank-and-file engineers feel like they’ve stepped into an unsettling future. “If AI is a rising water level, it’s recently reached a point where it has submerged the skilled engineer,” O’Brien said. “In a year, I expect coding agents will be better than any human.”
‘Coding is practically solved’
Until 2025, AI coding tools were powerful but functioned more or less like an engineer’s assistant. But in November, Anthropic released a new model for its AI coding tool, Claude Code, a dramatic step up from existing AI tools. For many developers, the leap felt bigger than the moment ChatGPT arrived in November 2022. “It just broke my brain,” said Edwards.
After a brief prompt, Claude Code — a tool that lets Claude work autonomously — can build features, run tests, fix bugs, and check its own work, all without direct human supervision. Since the “Claude Christmas” launch, Anthropic has released an even more capable model: Opus 4.6. OpenAI put out its advanced coding model, GPT-5.3 Codex, this month.
Edwards likens his newfound ability to tell Claude to build things for him as a kind of superpower. He has written hundreds of thousands of lines of code in two weeks across six passion projects, including a “Star Wars” card game — and read almost none of it. He describes the arrangement as akin to managing a team of four software engineers, but for the relative steal of paying Anthropic a few hundred dollars a month.
The excitement, though, is undercut by unease. “The AI has reached a point of sophistication where it’s telling itself how to work based on the best engineering practices,” Edwards said. When it does stop to ask him a question, he added, it’s designed to never need to ask the same thing twice, effectively creating a closed loop that doesn’t need a human. “So what does it need me for?” he asked.
A crucial step toward artificial general intelligence — machines that can perform any intellectual task that a human can — is thought to be “recursive self-improvement”: AI models that can keep making themselves better. Boris Cherny, the creator of Claude Code, has said signs of self-improvement are already emerging.
“Today coding is practically solved,” Cherny said this week (opens in new tab) on Y Combinator’s podcast. “We’re going to start to see the title of software engineer go away. It’s just going to be ‘builder’ or ‘product manager.’”
Engineers at Anthropic and OpenAI have said in recent weeks that AI writes 100% of their code (opens in new tab). Meta CEO Mark Zuckerberg has predicted (opens in new tab) that by midyear, AI will write most of his company’s code.
AI apocalypse or explosive employment?
The labor market has already started to shift, with cracks first appearing at the bottom of the ladder. Since 2019, hiring of new graduates at the 15 largest U.S. tech companies has fallen 55%, according to venture capital firm SignalFire, as companies lean more heavily on automation and a smaller pool of seasoned engineers. So far this year, there have been AI-driven layoffs at Pinterest, Autodesk, Amazon, and Salesforce. Software job growth has remained sluggish despite the AI boom in San Francisco.
College students are taking notice. For years, computer science was one of the fastest-growing majors, with enrollment surging as tech work promised high salaries and job security. But for the first time since the dot-com bust in the early 2000s, undergraduate computer science enrollment across the UC system declined (opens in new tab) — falling 6% in 2025 and 3% in 2024.
In San Francisco, fully employed software engineers are pondering how long their jobs will last — and what comes after.
“It’s such a weird time to be a junior software engineer,” said one employee of a large SF-based tech company, who said all of his code is now written by AI. “I’m basically a proxy to Claude Code. My manager tells me what to do, and I tell Claude to do it.”
The employee, who spoke anonymously because he is not authorized to speak to the press, said his predominant feeling about AI being a better coder than him is grief. “The skill you spent years developing is now just commoditized to the general public. It makes you feel kind of empty.”
He described a constant fear that the next model release could make his job obsolete — which he thinks could happen in a year or two. “Then I guess I’ll move to Yosemite and become a park ranger,” he said.
Another engineer at a medium-size tech company said that since he started using AI to write code, he understands only about half the work he produces. “Maybe we’ll all just get fired and put into the permanent underclass,” he said. “I will fix cars if we end up in the AI apocalypse.”
The “permanent underclass” refers to the idea that many people who were once all but assured of well-paying jobs and growth opportunities might be locked out of economic mobility because of AI. The term has become a common reference in San Francisco tech conversations. The idea emerged after ChatGPT’s launch in 2022, when human-level AI seemed a decade or more away. Now it’s more like everyday gallows humor.
Ridden with uncertainty, many engineers are grasping for historical precedents. Optimists point to the compiler, a program invented in the 1950s that translated human-readable code into instructions that a computer can execute. When compilers arrived, some feared they would eliminate programming jobs. Instead, the opposite happened: Coding became easier and cheaper, creating far more jobs than before. The pattern repeated with new programming languages, debugging tools, and shared code libraries — each breakthrough made individual engineers more productive, and the industry kept expanding.
The doomers, however, argue that the tractor offers a better analogy. “Tractors didn’t just change farming,” wrote one commenter (opens in new tab) on Y Combinator’s discussion forum Hacker News, on a thread discussing whether AI will replace software engineers. Instead, they “emptied entire regions” of farmers. In this scenario, engineers are the farmers.
Enrico Moretti, a labor economist at UC Berkeley, offers a grain of hope for the local workforce. He thinks San Francisco may be an exception. The city, he argues, is uniquely positioned to benefit from AI rather than be hollowed out by it. While other regions may see net job losses as AI displaces workers, San Francisco’s growing AI sector could offset and eventually surpass weakening demand for traditional software engineers.
“San Francisco is probably the only city in the U.S. where, on net, the AI revolution will add to employment and not subtract,” Moretti said. His reasoning: The AI era is in its early phase, with most jobs concentrated in research and technical development. OpenAI and Anthropic employ fewer than 10,000 people combined. Moretti expects broader job growth as the technology moves from experimentation to selling AI-powered products at scale — a shift that will favor the Bay Area. “I think the trend will be explosive in terms of employment,” he said.
But that growth may take years to arrive — too long for engineers who feel their job security eroding in real time. Many feel they’re getting the raw end of a bargain they thought was secure. They followed the rules, earned computer science degrees, and built careers on the premise that technical skill was a path to stability. Now they scroll through “permanent underclass” memes on X during lunch breaks, while Claude Code does the work they trained for years to do themselves.
In the worst-case scenario, O’Brien has advocated for universal basic income (opens in new tab) funded by taxing the AI systems that will eventually be doing the bulk of society’s work. That future, the professor believes, is no longer distant — and the shape it takes depends on choices being made now.
We could end up with an idealized future, O’Brien said, like “Star Trek.” Or we could end up with a darker existence, like “The Hunger Games.”
“Both stories,” he said, “start with the machines doing all the work.”