Why AI hasn't replaced software engineers, and won't
This article challenges the popular narrative that AI will lead to mass layoffs for software engineers, arguing that AI primarily automates coding (the 'execute' layer) while human judgment in 'decide' and 'deliver' remains critical. It critiques 'AI washing' in corporate layoffs and explains why increased productivity from AI may lead to more demand for software engineers, not less. The piece sparked a spirited debate on HN, with some commenters agreeing with the nuanced view while others insisted that AI will indeed replace many roles, especially those focused solely on development.
The Lowdown
The piece delves into the widespread anxiety surrounding AI's potential to replace jobs, specifically focusing on software engineering where AI capabilities are most advanced. Authors Arvind Narayanan and Sayash Kapoor contend that the data does not support the narrative of mass layoffs due to AI, even within the tech sector.
- AI Washing: The article reveals that many corporate layoffs attributed to AI (e.g., Block, Snap, Intuit) are often financially driven, with AI being used as a convenient, positive-sounding excuse. Surveys show most companies aren't ready for AI to replace these jobs.
- The Decide-Execute-Deliver Sandwich: Software engineering is broken down into three conceptual layers: deciding what to build, executing (coding), and delivering (testing, accountability). AI excels at compressing the 'execute' layer, but the 'decide' and 'deliver' layers, requiring human understanding, judgment, and accountability, remain resistant to automation.
- Coding is Not the Bottleneck: Historically, writing code constitutes a small fraction of a developer's time. The real bottlenecks involve problem specification, verification, and deep understanding of the codebase and business needs.
- Agentic Engineering vs. Vibe Coding: The article distinguishes between haphazard 'vibe coding' (unsupervised AI use) and 'agentic engineering' (human-supervised AI as a tool). The latter is becoming the norm, with engineers finding supervising AI agents surprisingly exhausting.
- Increased Demand, Not Displacement: Applying economic principles like price elasticity and Jevons' paradox, the authors suggest that as AI makes software creation cheaper, the overall demand for software (and thus software engineers) will increase, similar to historical patterns in other industries where technology boosted output.
In conclusion, the article posits that AI will transform the role of software engineers, making them more like 'crane operators' supervising AI agents. While individual roles may shift, the aggregate demand for human software engineering skills is expected to remain strong, focusing on the critical human elements of decision-making and accountability.
The Gossip
Replacement Realities
Commenters were split on whether AI would actually replace software engineers. Some vehemently argued that AI would indeed replace developers, especially as tools for deployment and maintenance become productized, enabling non-technical users to build full applications. Others nuanced this, suggesting AI would replace 'software developers' (focused on execution) rather than 'software engineers' (who handle the full 'sandwich').
Productivity Paradoxes
Many commenters discussed the practical impact of AI on productivity. Some believed that while AI might not fully replace roles, it would significantly increase the output expected from remaining engineers, leading to fewer hires. There was also a strong sentiment that the initial 'holy f*** I'm so productive' phase with AI coding would lead to a 'slop curve' of half-finished projects and a new set of problems, as managers misinterpret AI's capabilities.
Historical Interpretations
One comment challenged the article's historical comparison regarding job displacement. While the article cited elevator and telegraph operators as examples of jobs automated away, the commenter pointed out the much larger scale of job displacement in agriculture (from 15% to 2% of the workforce), suggesting that significant shifts due to technology are not unprecedented.