Perplexity AI CEO Aravind Srinivas backs viral post saying AI is pulling computer science back to maths and physics
Perplexity AI chief executive Aravind Srinivas has lent his voice to a growing debate about what artificial intelligence is doing to software engineering, endorsing a viral post arguing that large language models are quietly automating the grunt work of coding and pulling computer science back toward its mathematical and physics-heavy origins.
On March 13, Srinivas quote-tweeted a post by physics and AI/ML student @TheVixhal with two words: "Well said." The original post, which garnered more than 15,000 likes and nearly a million views, made the case that LLMs are shifting what engineers spend their time thinking about, away from syntax and boilerplate, and toward the kind of foundational reasoning that underpins physics and mathematics.
The endorsement adds to a chorus of prominent voices signalling a fundamental shift in the profession. Anthropic chief executive Dario Amodei has said the industry is six to twelve months away from AI handling most of what software engineers currently do end-to-end, noting that some engineers at Anthropic no longer write code at all. Replit's chief executive was more direct, suggesting the software engineering role as currently defined "sort of disappears."
The data supports the shift. A 2023 Microsoft-run experiment found developers using GitHub Copilot completed tasks 55.8% faster. Anthropic's own AI Exposure Index places programmers at roughly 75% task coverage by LLMs, the highest of any profession tracked.
What is changing, observers note, is not just speed but the nature of the work itself. Boilerplate is increasingly being handled by machines. The questions that remain, including how systems fail, what trade-offs to make and whether an architecture holds at scale, are closer in nature to physics and mathematics than to typing syntax correctly.
Not everyone is convinced the transition is straightforward. Critics point out that LLMs still struggle on novel and complex problems. Junior developers are seen as gaining the most from these tools, while senior engineers remain essential for verification and judgement calls. The six to twelve month timeline Amodei has outlined applies to existing tasks, not the harder work of inventing entirely new systems.
The debate has reached education as well. Code.org's founder is already rethinking computer science curricula, shifting emphasis away from syntax and toward logical reasoning. "Coding is dead," he said. "Long live coding."
On March 13, Srinivas quote-tweeted a post by physics and AI/ML student @TheVixhal with two words: "Well said." The original post, which garnered more than 15,000 likes and nearly a million views, made the case that LLMs are shifting what engineers spend their time thinking about, away from syntax and boilerplate, and toward the kind of foundational reasoning that underpins physics and mathematics.
The endorsement adds to a chorus of prominent voices signalling a fundamental shift in the profession. Anthropic chief executive Dario Amodei has said the industry is six to twelve months away from AI handling most of what software engineers currently do end-to-end, noting that some engineers at Anthropic no longer write code at all. Replit's chief executive was more direct, suggesting the software engineering role as currently defined "sort of disappears."
The data supports the shift. A 2023 Microsoft-run experiment found developers using GitHub Copilot completed tasks 55.8% faster. Anthropic's own AI Exposure Index places programmers at roughly 75% task coverage by LLMs, the highest of any profession tracked.
What is changing, observers note, is not just speed but the nature of the work itself. Boilerplate is increasingly being handled by machines. The questions that remain, including how systems fail, what trade-offs to make and whether an architecture holds at scale, are closer in nature to physics and mathematics than to typing syntax correctly.
Not everyone is convinced the transition is straightforward. Critics point out that LLMs still struggle on novel and complex problems. Junior developers are seen as gaining the most from these tools, while senior engineers remain essential for verification and judgement calls. The six to twelve month timeline Amodei has outlined applies to existing tasks, not the harder work of inventing entirely new systems.
The debate has reached education as well. Code.org's founder is already rethinking computer science curricula, shifting emphasis away from syntax and toward logical reasoning. "Coding is dead," he said. "Long live coding."
Next Story