Студенты-программисты переходят от написания кода к управлению ИИ: как меняются требования к стажировкам

Students pursuing software engineering (SWE) are entering an internship market reshaped by generative artificial intelligence (AI). The fundamentals of the job’s responsibilities have changed due to an increased dependability on AI programs. As AI tools take on more of the code-writing that once defined the job, two UCI computer science (CS) students say that qualifying for an internship is less about writing code from scratch and more about knowing how to direct, review and build on what AI produces. Fourth-year CS and intelligence systems student Nathan Ong is currently a SWE intern at Leidos. However, Ong doesn’t see internships as necessary until the summer between an undergraduate’s third and fourth years, where the pressure for a full-time position after graduation peaks. He recommends acquiring industry experience through school organizations and personal projects before applying, having landed positions through both cold applications and internal connections in separate cycles. Ong prepared for CS recruitment, especially its technical aspects, mostly outside of university resources. Aspiring SWEs including Ong commonly practice timed coding problems on platforms like LeetCode, which is heavily dependent on individual initiative. At Leidos, Ong said he’s learned how a tech corporate setting truly runs: Managing work in sprints, communicating team updates and making sense of design decisions made by former employees in legacy codebases, including picking up a new programming language on the job with minimal guidance. His current internship focuses more on teamwork than solo engineering unlike his past ones because the current corporate world stresses knowing how to use AI tools like Claude Code. Ong said understanding AI is expected and recruiters now check whether candidates can prompt in generative AI to improve workflow efficiency for the same tasks that took longer pre-AI. “[AI is] just more efficient, and it helps get stuff done faster, and that’s what a lot of companies care about, efficiency,” Ong told New University. Fourth-year CS major and SWE intern at NASA’s Jet Propulsion Laboratory (JPL) Yousef Khan echoed the same sentiment. According to Khan, Claude generates code “all day” at JPL, while interns review Claude’s work and make sure it functions within a “human context.” AI-generated code has maximized his team’s efficiency, letting them focus more on implementing code to their project instead of writing it from scratch. As a result, Khan’s team can now see the full development of their project and advance it further in the same length of their internship program. Khan also expects the SWE interview process to shift toward evaluating how well candidates can debug AI-generated codes on top of LeetCode problems. What Khan didn’t expect in his internship, though, was learning industry-standard platforms like Go and Docker because they weren’t covered extensively in his classes. Alternatively, he credited UCI’s Python sequence — I&C SCI 31-33 — for teaching him how to think about program design, saying that learning programming languages is built through practice and application over sole memorization of concepts and terminology. Both students said UCI’s CS courses have adapted to AI mainly on the academic-integrity side rather than directly teaching AI tools and its application to workforce tasks. “They’re like, oh, disclose AI whenever you use it. Some classes say to not or do use AI, so they’re not really encouraging it or helping you use it,” Ong said. Khan added that teaching AI to students is tricky because AI grows too fast to reflect in weekly lectures. “It’s an impossible feat to ask education to migrate curriculum at the pace that AI kind of boomed in the last three years,” Khan told New University. He noted that some of his professors have modified their projects to accommodate for AI usage, one conducting a poll to ask whether AI usage should be fully allowed for class projects, given they increase in scale, while another promoted open AI usage if students could defend their design decisions in oral interviews. Khan said he noticed job descriptions over the course of the AI boom increasingly list AI fluency as a requirement, though simply listing knowledge of certain softwares on a résumé doesn’t exactly tell recruiters how strongly the applicant knows how to use AI for SWE responsibilities. Additionally, the distinction of applicant capability and integrity with using AI for SWE responsibilities isn’t always clear when put in condensed bullets, leading to a potential lack of transparency between applicants and employers. Khan gave an example with self-made projects: Rather than accumulating a large portfolio of AI-built projects, an engineer could display a few projects that integrate both their own knowledge and AI-assisted work. Students may face pressure to show recruiters strong passion by building a large number of projects, which might mean an applicant builds numerous softwares primarily by AI instead of utilizing their own skill for a select number of quality work. Based on Ong and Khan’s experiences, UCI’s CS curriculum seems to be slower to formally integrate generative AI than the workplace expects it, creating a large gap both students said falls largely on individual professors to address rather than the program as a whole. For students preparing to enter the field, their accounts suggest that internships remain a critical entry point, but the skills that make a candidate competitive, such as fluency in directing and reviewing AI-generated code, are ones they say they’re picking up on the job, not in the classroom. Piyush Vaishnava is a Features Intern for the summer 2026 quarter. He can be reached at [email protected]. Edited by Ennes Kahf and Elizabeth Gregg.