Why AI hasn’t replaced software engineers, and won’t
AI Snake Oil · Arvind Narayanan · 2026-06-11
Arvind Narayanan argues, using economic data and a 'decide-execute-deliver sandwich' model, that AI will not cause mass layoffs in software engineering because automation only compresses the middle execution layer while human judgment and accountability at both ends remain essential.
Appears in
Extraction
Topics: ai-labor-economicssoftware-engineering-jobsai-productivityagentic-engineeringfuture-of-work
Claims
- Reported AI-driven layoffs in software engineering are largely 'AI washing' of financially motivated cuts, with 59% of U.S. hiring managers admitting they emphasize AI when explaining layoffs to stakeholders.
- Software development is a 'decide-execute-deliver sandwich' in which AI compresses the middle execution layer but leaves decision-making and delivery accountability largely unchanged.
- Writing code was never the bottleneck in software development; studies show developers spend as little as 9% of their time on coding.
- AI agents produce an eight-fold increase in lines of code written but only a 30% increase in software releases, demonstrating that human bottlenecks in the Decide and Deliver layers persist.
- Vibe coding and agentic engineering are fundamentally different practices; production software still requires skilled engineers who supervise and remain accountable for AI output.
- Demand for software engineering may increase overall due to price elasticity — cheaper software production will drive demand for more software, not fewer engineers.
Key quotes
AI compresses the 'execute' layer — the middle of the sandwich — but the other two layers resist automation in a way that will not be overcome by capability improvements alone.
59% of U.S. hiring managers admitted they emphasize AI when explaining hiring freezes or layoffs because it plays better with stakeholders than citing financial constraints.
Across 100,000 developers on GitHub, the researchers found that AI agents led to an eight-fold increase in the number of lines of code written... But this led to only 30% more releases, strongly suggesting that human bottlenecks (the Decide and Deliver layers) remain in place.