Snap announced this week it's eliminating 1,000 jobs - 16% of its workforce - while simultaneously claiming that AI now generates over 65% of the company's new code and handles more than a million customer support questions every month.
The CEO framed the cuts as efficiency gains. The AI does the work, the humans go home. Simple maths. $500 million in projected cost savings.
This is the conversation we've been dancing around for two years. Not "will AI replace jobs?" but "what does it look like when it starts happening at scale?"
The numbers tell two stories
Sixty-five per cent of new code. That's not a pilot programme. That's not experimentation. That's production systems being built by machines, shipped to millions of users, maintained by... well, that's the question, isn't it?
A million support questions handled monthly. Those are conversations that used to require people. Now they require prompts and guardrails and the occasional human escalation when the model gets confused.
The honest version of this story isn't "AI creates efficiencies." It's "AI does work that people used to do, and those people are expensive."
For business owners watching this unfold, the calculation is straightforward. If AI can generate two-thirds of your codebase and handle your support queue, why wouldn't you lean into that? The cost savings are real. The shareholders care about margins, not headcount.
What nobody's saying out loud
Here's what makes this different from previous automation waves: the speed. Factory automation took decades to replace assembly line workers. AI code generation went from interesting experiment to production dependency in under three years.
Developers I speak to are split. Half are using AI tools daily and shipping faster than ever. The other half are watching these announcements and wondering if they're next.
The uncomfortable truth is that both things can be true simultaneously. AI makes remaining developers more productive and you need fewer developers overall. Snap's 65% figure suggests they're not replacing developers one-for-one with AI. They're replacing teams.
For junior developers entering the workforce, this changes the game entirely. The "write boilerplate for two years while you learn" pathway is gone. If your value proposition is writing code that AI can write faster, you don't have a value proposition.
The maintenance question
What happens when AI-generated code needs fixing? When edge cases emerge that the model didn't anticipate? When business requirements shift and the generated code needs fundamental restructuring?
The optimistic view: remaining developers spend less time writing and more time thinking. They become architects, editors, validators. Higher-level work, better paid, more interesting.
The realistic view: companies discover that maintaining AI-generated code at scale is harder than they thought. That the cost savings on the front end create technical debt on the back end. That you still need humans who understand the systems deeply, not just people who can prompt models effectively.
We'll find out which view is correct over the next 12 months. Snap is the test case. A major tech company betting its engineering efficiency on AI code generation at production scale.
What this means for everyone else
If you're running a business, the pressure to adopt AI tools just intensified. Your competitors are looking at Snap's $500 million cost savings and asking their engineering leads why they're not doing the same thing.
If you're a developer, the skillset is shifting. Writing code is becoming table stakes. Understanding architecture, debugging AI output, knowing when to override the model - those are the differentiators now.
If you're entering the workforce, understand that the job market has fundamentally changed in the last 24 months. The roles that exist today might not exist in five years. Not because AI will be conscious or AGI will arrive, but because the economics of software development are shifting faster than anyone expected.
Snap's announcement isn't a warning. It's a datapoint. One company, one set of metrics, one set of decisions. But when a major tech company publicly states that AI generates 65% of new code, everyone else is doing the same calculation.
The question isn't whether this will spread. It's how fast, and who's ready for it.