This Company Says Humans Should Never Write Code Again (And They Might Be Right)

Anthropic CEO Dario Amodei made a prediction that's sending shockwaves through the tech industry: "We might be 6-12 months away from models doing all of what software engineers do end-to-end."

He's not alone. Jensen Huang says coding isn't a viable career anymore. Marc Benioff says Salesforce is "seriously debating" hiring engineers. Zuckerberg claims mid-level AI engineers will replace professionals.

But here's what they're all missing—and what founders actually need to understand.

The Claim vs. The Reality

Amodei's statement is striking because he's not a hype merchant. Anthropic is arguably the most thoughtful AI lab, and Dario has historically been measured in his predictions. When he says engineers at Anthropic "don't write code anymore," that's worth taking seriously.

But let's examine what that actually means.

Anthropic engineers using Claude to write code are still engineers. They're reading the output, verifying correctness, architecting systems, debugging when things break. The model writes; they think.

That's different from "AI replacing software engineers." It's AI changing what engineering work looks like. And that distinction matters enormously for how you build your company.

The Vibe Coding Reality

A new term emerged in 2025: "vibe coding." Coined by Andrej Karpathy, it describes a mode where you "fully give in to the vibes"—letting AI write code while you focus on direction and outcomes rather than implementation.

Here's what research actually shows about vibe coding in practice: professional developers who tried it report mixed results. Simple tasks accelerate dramatically. Complex systems break. The developers who succeed aren't abandoning their engineering skills; they're applying them to a different layer of the stack.

The engineers who "don't write code anymore" are actually doing more engineering than ever. They're just doing it at the requirements and architecture level rather than the syntax level.

What This Means for Hiring

If you're a founder wondering whether to hire engineers, here's the framework: AI is making execution cheaper but judgment more valuable.

The engineer who can write clean Python is less differentiated than before. The engineer who can decompose an ambiguous product requirement into a coherent technical approach, then verify that an AI's implementation actually meets that specification—that engineer is more valuable than ever.

Salesforce isn't "debating" hiring engineers because they don't need engineering work done. They're debating whether junior engineers who primarily write boilerplate provide enough value versus what AI can do. The answer may be no. But senior engineers who think in systems? Those are still scarce.

The Startup Implications

For early-stage companies, AI coding tools create genuine leverage. A founding team of 2-3 strong engineers can now ship what previously required 5-6. That's transformative for capital efficiency and speed.

But there's a trap: thinking that AI can substitute for engineering judgment entirely. The startups that will fail are those that "vibe code" their way into production systems without anyone who actually understands what they've built.

Technical debt compounds. Security vulnerabilities hide. Performance problems emerge at scale. AI-generated code has all these problems in spades because it optimizes for "works in the demo" rather than "works in production."

The Real Bottleneck

Amodei acknowledged something crucial: the feedback loop isn't closed. AI can write code, but chip manufacturing, training time, and physical infrastructure still create bottlenecks. You can't AI your way to infinite AI.

This maps to company-building. The bottleneck in most startups isn't writing code. It's understanding users, finding product-market fit, making strategic decisions about what to build and what not to build. AI doesn't help with those—yet.

A startup with infinite AI coding capacity but no understanding of their market will build the wrong thing faster. That's not progress.

What Smart Founders Should Do

Lean into AI tooling aggressively. Cursor, Copilot, Claude—use them. Your engineers should be shipping more with these tools, not resisting them.

Hire for judgment, not just output. The engineer who questions whether a feature should be built at all is more valuable than one who builds it instantly. AI makes the second type abundant; the first type remains rare.

Don't mistake AI-assisted velocity for readiness. You can ship faster, but shipping the wrong thing faster doesn't help. Maintain discipline about what deserves to exist.

Plan for a transition, not a revolution. The 6-12 month timeline is aggressive. Even if models reach that capability, organizational adoption takes longer. But in 3-5 years, engineering teams will look very different. Hire with that in mind.

The Meta-Point

When every CEO in tech says the same thing—"coding is dead"—ask why. They're talking their book. They want to reduce labor costs. They want to signal that their companies are AI-forward. They want to justify massive AI investments.

That doesn't mean they're wrong. But it means their incentives aren't perfectly aligned with truth. The reality is messier than "engineers are obsolete." It's closer to "engineering is being redefined, and the definition will keep changing."

The founders who win will be the ones who ride that change rather than either denying it or overreacting to it.