Arvind Narayanan (Princeton) and Sayash Kapoor provide a detailed analysis of why AI cannot yet replace software engineers, despite appearing to be the profession most easily disrupted by AI. Data from New York's WARN Act reveals that in 2025, over 160 companies filed layoff notices, but none checked the "reason: AI" box.
AI Only Accelerates "Typing Code Into Computer"
In reality, AI only speeds up the code-writing portion, which isn't the true bottleneck of software work. What makes software work complex isn't writing code, but: Task breakdown - Decomposing large problems into manageable smaller parts Understanding requirements - Communicating with clients and other teams Debugging - Finding and fixing complex defects Communication - Team collaboration and explaining technical concepts
AI Affects Through "Slower Hiring," Not "Increased Firing"
Data shows AI's impact on the labor market isn't through firing workers, but through companies hiring more slowly. Mass layoffs citing AI are mostly "theatre" — low labor costs are the real reason.
"Sandwich Model": Knowledge Work = Decide → Do → Deliver
This model shows knowledge work consists of 3 layers, where AI helps only in the "Do" layer. Decision-making and delivery still require humans with tacit knowledge from experience.
The Importance of Tacit Knowledge
Engineers with tacit knowledge become more valuable in the AI era. Firing senior teams to cut costs and replacing them with AI destroys accumulated organizational knowledge — a valuable asset.
In summary, AI is a useful tool for software engineers but cannot replace them because software work isn't just writing code — it's complex problem-solving and understanding that requires deep knowledge and experience. Reference: https://news.ycombinator.com/item?id=48487540