“SWE will be replaced by AI in 2-3 years” - Dario Amodei. That’s not what Dario means, read on…
I recently watched a video where Dario Amodei, CEO of Anthropic, offered a refreshingly measured perspective on AI’s impact on software engineering—one that contrasts sharply with typical hype and headlines.
When asked about the timeline for full SWE automation, Amodei believes we’ll reach it in 1-2 years.
By full SWE automation, he means:
- Environment setup: Compiling code and setting up clusters or environments
- Testing & Documentation: Testing features and writing memos or design documents
- High-Level Strategy: Understanding the context of a problem and setting technical direction
However, he explains that the problem should be viewed as a spectrum of benchmarks:
- 90% of code written by AI: This has already occurred at Anthropic and other organizations. However, Amodei calls this a “weak criterion” because it doesn’t mean you need 90% fewer engineers; it simply means the model assists with syntax.
- 100% of code written by AI: This represents a significant leap in productivity but still only covers code generation itself
- 90% of end-to-end SWE tasks: The model handles the vast majority of the workflow
- 100% of today’s SWE tasks: The model can perform every task currently required of a software engineer, including setting technical direction and understanding the context of the problem
He notes that even when 100% of today’s SWE tasks are automated, engineers likely won’t disappear immediately, instead, they may transition into oversight and management roles responsible to oversee of these AI systems.
Timeline
Amodei has a “hunch” (50/50 probability) that we’ll reach full end-to-end SWE automation within one to three years. He considers it a “super safe bet” (95-99% certainty) that this will happen within ten years.