Why AI hasn’t replaced software engineers, and won’t
New York WARN Act data (2025) shows zero AI-related layoff disclosures. Software engineering bottlenecks are requirements, verification, and deep context. AI accelerates coding but not the human-centric parts of the job. The narrative of imminent mass AI displacement lacks evidentiary support.
Analysis
TL;DR
- New York WARN Act data (2025) shows zero AI-related layoff disclosures.
- Software engineering bottlenecks are requirements, verification, and deep context.
- AI accelerates coding but not the human-centric parts of the job.
- The narrative of imminent mass AI displacement lacks evidentiary support.
Key Data
| Entity | Key Info | Data/Metrics |
|---|---|---|
| New York State | First to add AI disclosure to WARN Act filings | March 2025 |
| WARN Act Filings | Total companies filing notices in first full year | >160 companies |
| AI Disclosure Box | Times checked by filing companies | 0 (Not a single one) |
Deep Analysis
The tech discourse is stuck in a simplistic loop: AI gets better at X, therefore job Y is doomed. This essay rightly punctures that balloon, and the most potent evidence comes from a mundane, bureaucratic source—the WARN Act. New York's experiment is a smoking gun. In a state that is a tech industry hub, not one of over 160 companies making significant layoffs felt compelled to blame or even cite AI. This isn't proof of zero impact, but it is devastating proof against the mass displacement narrative. If AI were a primary driver of significant workforce reductions, corporate legal teams, navigating disclosure requirements, would have used it. They didn't.
The authors then move to the more interesting, nuanced argument. They correctly identify that the "AI will replace coders" take misunderstands the job. Coding, the act of translating a known solution into syntax, is a task, not the profession. It's the translation layer. AI is brilliant at translation. But software engineering, at its core, is an act of human judgment under uncertainty. The real work happens in the murky pre-code and post-code phases.
The three bottlenecks they identify—deciding what to build, verifying its correctness and fitness, and the deep contextual understanding required for both—are fundamentally human-centric. They are about problem formulation and accountability. An LLM can generate a plausible function to sort a list, but it cannot attend the stakeholder meeting where conflicting priorities are negotiated, absorb the history of why a past solution failed, or sign off on a deployment that carries business risk. It can assist in these areas—summarizing documents, suggesting options, finding edge cases—but the core act of judgment and the bearer of responsibility remains human.
My own experience aligns with this. AI is a force multiplier for a skilled engineer. It handles the rote, accelerates exploration, and acts as a tireless pair programmer. But it also raises the value of the human skills it cannot replicate. When code generation is commoditized, the engineer's worth is distilled into their architectural taste, their grasp of system trade-offs, their communication with non-technical partners, and their ability to debug the unexpected—the failure mode no one thought to specify. AI is great for the known unknowns; humans are still needed for the unknown unknowns.
The weak point in the analysis is its potential to become a temporary truism. The argument that "software engineering is more than coding" is correct today. But the authors must grapple with the recursive problem: as AI tools improve, will the definition of "the hard parts" not shift? Today's bottleneck in "verifying what is delivered" might be tomorrow's automated audit trail. The essay correctly focuses on current evidence, but the longer-term game is about whether AI can eventually simulate the gained experience and institutional knowledge that constitute that "deep human understanding." The current answer is no, but the trajectory is what matters for careers and education.
Industry Insights
- The value of software engineers will shift decisively toward product thinking, systems design, and cross-functional communication.
- Companies will invest more in internal tools and platforms that amplify their senior engineers' contextual knowledge, creating higher-value "human-in-the-loop" AI systems.
- Technical interviewing and hiring will need to evolve to assess AI-collaboration skills and judgment, not just raw coding ability.
FAQ
Q: If AI isn't replacing engineers, why are tech layoffs happening?
A: Layoffs are driven by macroeconomic factors, over-hiring during the pandemic boom, and strategic refocusing. The article's point is that these layoffs are not being caused by AI, as evidenced by the lack of disclosure.
Q: Should junior engineers be worried about AI automating their entry-level tasks?
A: Yes, but not for wholesale replacement. The nature of entry-level work will change. Learning to effectively use and supervise AI tools will become a core junior skill, shifting focus from pure coding to problem decomposition and verification.
Q: Does this mean AI is just a fad for software development?
A: Not at all. AI is a transformative productivity tool, akin to the shift from assembly language to high-level languages. It makes developers more powerful, but the role evolves; it doesn't vanish. The hype of replacement is the fad.
Disclaimer: The above content is generated by AI and is for reference only.