“We don’t need the junior developers anymore ,” said Amr Awadallah, chief executive of Vectara, summarizing the tremors running through the best computer science programs in the United States. A degree in software engineering from Stanford ensured a Stanford engineer a spot in the Silicon Valley elite. But now even the best-of-breed students are having trouble scheduling an interview. What they are up against was not an overnight talent gap but the ensuing development of generative AI coding assistants, which are able to write hours of high-functioning code in a matter of minutes with fewer bugs than most new graduates. While the initial version of chatbot technology, ChatGPT, in 2022 was only able to write code for 30 seconds at a time, the newer version, such as the assistive tool called Claude, can power through an eight-hour block of coding.
As one Valley hiring manager put it, The new math is this: instead of needing to hire ten engineers in the past, they can now hire two senior engineers and a large language model agent to get the job done. The CEO of Anthropic, Dario Amodei, has said that his company’s 70% to 90% of the code for some products is being generated by Claude. His prediction? Half of white-collar jobs will likely be eliminated by AI in the next five years. The statistics are not reassuring, either, as a 20% loss in employment for software developers aged 22 to 25 since late 2022 is reported by the Stanford Digital Economy Lab in other AI-impacted jobs, such as customer service and accounting.
The technology’s capabilities are certainly impressive, but nowhere near perfect, Legato continues. They’re great at unstructured, repetitive programming-such as to create boilerplate or patch bugs, and writing tests, but they’re ‘jagged’-full of logical holes, and they tend to produce ‘garbage’ output, Nenad Medvidović, a USC researcher, told scientists in *The Persistence*. Besides, research led by METR came to the conclusion that programmers assisted by software are 19% slower at practical work since they use lots of time to assess, correct, and compile their ‘garbage’ into complex systems. And here’s the paradox: industry players re-thought their processes of programming, which boosted interest in managers’, not programmers’, jobs.
What IT employees want, they can no longer easily get, Porges wrote in his prophecy of doom for the IT sector in 2019. For IT people, the only people who can now compete for the best jobs in the field are ‘cracked engineers’ with tons of product experience and a research background. Everyone else is lowering their sights, taking jobs in less exciting companies, and startups are proliferating. Many are going into academia; the number of students in the master’s program at Stanford in the fifth year has skyrocketed, giving them more time to learn about AI and beef up their resume.
Universities are scrambling to modify their curricula, with projects such as the GenAI in CS Education Consortium at the University of California, San Diego, developing “ready to go” classes to teach students to weave AI inputs into their software development processes and deal with the outputs to understand the benefits. What’s more, this disruption effect of AI is uneven in itself: while programmers in coding positions are in decline, a rise in demand occurs for those with experience in AI design, verification, and cross-discipline tasks.
According to Brex’s engineering director, T-shaped or multi-spired engineers who have deep expertise in their technical niche and also product management, UX design, or business will thrive. AI, for instance, is still unable to link innovations across multiple fields or simulate empathy during customer understanding. This has not prevented the fast-tracking of AI into development tools, however. GitHub’s Copilot, OpenAI’s GPT-4, and Anthropic’s Claude are already baked into top coding editing software, with professional developers adopting at a rate of more than 60 percent.
This profoundly changes the cost structure in software development and is driving organizations toward lean staffing. According to Coinbase’s Rob Witoff, the result of shifting lower-level tasks for developers to AI systems is bottlenecks for those higher up, such as software engineers now load-heavy with code reviews. This opens up more opportunities for additional automation. But the message to the current crop of graduates is crystal clear: AI fluency in development is no longer nice-to-have; the world of entry-level employment is being rewritten, and the traditional notion of the “bottom rung” in the corporate world is becoming extinct. Success now depends on an individual’s competency at working with AI both as a colleague and as a system that needs to be audited.
