This article digs into a heated debate: are large language models (LLMs) automating routine coding and changing what software engineers actually do? It spotlights some bold claims from AI industry leaders, shares real productivity data, and looks at how educators and companies are rethinking computer science education to focus more on reasoning and system design instead of rote coding.
Industry voices on LLMs and software engineering
In a viral thread, Perplexity AI CEO Aravind Srinivas argued that LLMs are starting to handle boilerplate coding, nudging computer science back toward its roots in math and physics. Meanwhile, Anthropic CEO Dario Amodei has floated the idea that we could be just 6 to 12 months away from AI taking on most software engineering tasks end-to-end. He even suggested some engineers might soon stop writing code altogether.
Replit’s CEO chimed in too, bluntly saying the traditional software engineering job could “sort of disappear.” These statements paint a picture of a field in flux. Automation is picking up speed, but it’s also forcing people to ask what engineers should do beyond just writing code.
From automation to architecture: what changes for developers
As LLMs take on routine tasks, engineers are shifting focus to bigger questions: system behavior, failure modes, trade-offs, and scalable architecture. The day-to-day now leans more on verification, judgment, and building new systems, even as long-standing coding chores get automated.
This shift doesn’t erase software engineering as a craft. Instead, it’s changing the skill set and the kinds of problems engineers work on.
Evidence of productivity gains and task coverage
There’s some real data backing up these changes. In a 2023 Microsoft experiment, developers using GitHub Copilot finished tasks about 55.8% faster than those who didn’t use the tool.
The Anthropic AI Exposure Index estimates that programmers have about 75% task coverage by LLMs—the highest among tracked professions. That suggests coding tasks are especially easy to automate.
- Productivity gains: Copilot users completed tasks much faster, showing clear efficiency improvements on routine work.
- High task coverage: LLMs can handle a big chunk of typical programming tasks, according to the AI Exposure Index.
- Quality and safety considerations: With more automation, engineers spend more time on verification, architecture, and designing systems that won’t break in unexpected ways.
- Divergent viewpoints: Skeptics point out LLMs still struggle with complex, novel problems, and that experienced engineers are critical for inventing new systems.
Limitations and the continued need for expertise
Not all coding tasks are simple, and LLMs can stumble on new or highly specialized problems. Senior engineers are still crucial for verification, judgment, and inventing systems that fit unique requirements.
That’s probably why many teams use AI tools for efficiency but keep humans in the loop for important decisions and creative work. It’s not all black and white—there’s a lot of gray area as the field shifts.
Education and curriculum responses
Educators and industry leaders are rethinking computer science curricula. The focus now leans more toward logical reasoning, problem decomposition, and architectural thinking.
Students get trained to articulate system behavior, failure modes, and trade-offs. Programs want to prepare the next generation of engineers for a world where design thinking and verification matter most in software creation.
As AI tools keep maturing, the journey for software professionals isn’t really about coding disappearing. It’s more about what expertise actually means now.
The best practitioners blend mathematical thinking with hands-on architectural judgment. They use automation to speed up innovation, but still care about building robust, scalable, and trustworthy systems.
Here is the source article for this story: Perplexity AI CEO Aravind Srinivas agrees that Computer Science is gradually returning to the domain of…