AI Raises The Bar For Tomorrow’s Engineers

Key Takeaways:

• Universities are integrating AI into engineering education while reinforcing physics, math, and other fundamentals that students need to evaluate AI-generated work.
• Critical thinking and engineering judgment are more important than ever because AI can produce plausible but incorrect answers, leaving students responsible for verifying results and assumptions.
• Schools are redesigning assessments around oral exams, in-class quizzes, labs, projects, and other tasks that test genuine understanding while allowing responsible AI use.

Today’s science, technology, engineering, and math (STEM) students will be entering the workforce at a pivotal moment. They grew up on iPads, attended high school classes on Zoom during the COVID pandemic, and are navigating college amid an AI revolution. Now, with generative and agentic tools able to accelerate many tasks, a burning question remains: How can universities adapt to this new era and help students harness AI’s advantages without weakening the foundational knowledge and judgment they will need as engineers?

Academic institutions are working to help students get the most out of AI tools while ensuring the next generation of engineers doesn’t cut corners on foundational knowledge. One trend is universities placing greater emphasis on critical-thinking tasks so students don’t swallow GenAI slop in totality. Just as large language models need guardrails, universities are developing tests and quizzes to ensure students learn properly and to minimize cheating. [See Table of University Approaches below.]

“We are seeing people who skate through or miss something fundamental that was important for a particular job, and they have to go back and relearn it,” said Alexandra Werth, assistant professor in biomedical engineering at Cornell University. “Higher ed is in trouble if we don’t work hard to make sure that doesn’t happen. I just went to an American Society of Engineering Education conference, where there were thousands of engineering educators, and this was what everyone wanted to talk about and figure out because this is what makes education relevant. People are trying a lot of different things, and we’re still in a weird place where AI technology is constantly changing. Saying, ‘This is the right approach,’ or ‘This is the wrong approach,’ is still in flux.”

Students are adopting AI tools faster than industry and academia can determine the right balance. “Everybody’s kind of struggling with it, to be honest,” said Matthew Ozalas, master application development engineer, scientist at Keysight. “I’ve talked to professors and heard a lot of different perspectives across the gamut. Some professors are saying, ‘No technology in my class — just blue books back to the old days.’ Others say, ‘Use whatever you want, but you have a time limit.’ I’ve been really impressed with how AI-savvy young people are. At the same time, there’s this idea about how much they need to know in the engineering field. People have asked, ‘Do they even need to know everything anymore? Can’t they just ask their large language model?'”

That means validation of AI responses is key. “The biggest shift we’re seeing is that next-generation engineers need to be good at defining problems, providing the right context to AI, validating its output, and understanding tradeoffs,” said William Wang, founder and CEO of ChipAgents. “Engineering judgment has become even more important, not less, because engineers are now responsible for validating AI-generated work, recognizing incorrect assumptions, and knowing when deeper investigation is required.”

That uncertainty only raises the stakes for mastering the fundamentals, especially when AI can produce answers that sound convincing but are wrong. “From an industry perspective, it is strongly recommended that universities should teach students how to combine fundamentals with AI-assisted workflows,” said Andy Nightingale, vice president of product management and marketing at Arteris. “Specify the problem clearly, generate alternatives, validate assumptions, test results, and document decisions. The graduates who will stand out are not those who use AI, but those who can use it safely, critically, and productively.”

Students also must be able to defend their findings orally, without AI help. “Graduates need soft skills tailored for a generation who have grown up on screens,” said Ashish Darbari, CEO of Axiomise. “In industry, engineers have to explain their reasoning on the spot — in design reviews, with customers, and when something goes wrong. Clear, confident verbal communication matters more in the AI era because it shows thinking that AI can’t produce for you. Universities should make it a deliberate part of the curriculum.”

In short, students still need the knowledge and judgment to evaluate what an LLM tells them.

Experience makes the risks clear. “If you work with AI and you’re experienced, you can see how confidently it gives you completely wrong answers or will hallucinate its way through,” said Ozalas. “It’s the balancing act that we have to strike. I can imagine it’s really hard right now to be a student or a young person trying to get into this.”

One professor, who spoke on condition he not be named, pointed to a potentially more disturbing issue. “AI is a resource that has been helping facilitate student research and learning over the past several years,” he said. “However, over this period, I have observed an overall decline in student retention and performance in the same type of curriculum, suggesting that the benefits of AI may be outweighed by increased reliance on this tool. The overall trend has appeared to be that students are having decreasing attention spans as AI becomes more integrated into education and knowledge transfer.”

Another professor, speaking on condition of anonymity, declined to discuss students’ use of AI. His reason: “Because then I’ll be bitter.”

Universities must address such concerns. “It is essential for university curricula to target GenAI weaknesses,” said Tamara Popovic, educational developer for IT and electrical engineering at ETH Zurich. “One of the biggest challenges we face is what’s called the illusion of competence. When a student prompts an LLM to explain a concept like heterojunction formation, the tool generates a beautifully structured, highly confident response. The student reads it, feels an immediate sense of clarity, and mistakes that passive clarity for actual mastery. They skip the difficult mental strain required to truly internalize the physics behind the phenomena. To break that illusion of competence, we started to design learning activities that target the specific cognitive blind spots of GenAI, namely, complex spatial relationships and non-standard visual abstractions.”

For example, the university introduced exercises based on multi-variable graphic shifting. “We present a standard plot, like a MOSFET C-V curve, but introduce a unique physical anomaly,” said Popovic. “Text-based LLMs, for now, cannot accurately map these multi-layered physical transformations onto new visual representations without making fatal logical leaps. Forcing students to navigate these visual-spatial shifts ensures they are relying on their own internal mental models rather than an external algorithm.”

If AI can mimic understanding, universities must decide how to teach the real thing.

Developing curricula in the AI era
Universities are figuring out how to allow AI tools while preserving academic rigor and integrity.

“Faculty is uncomfortable,” said Stanley Chan, professor of electrical and computer engineering and director of engineering AI education at Purdue University. “It’s the first time that we are navigating this path, and I tell administration and faculty members that we don’t have a crystal ball.”

The Purdue board recently approved a plan to integrate more AI into each successive year of a four-year degree. “We have to make the transformation because otherwise we will be outpaced by AI,” said Chan. “It’s a matter of time that higher education will be fundamentally reshaped because a new technology is here.”

To avoid loss of learning, professors have to ask students tough questions, and students need to learn how to ask AI tough questions, noted Chan. “Why are you doing this? Can you give me the evidence? Can you show me that this is working? Have you considered other design options? Why is this design done in this way, and have you thought about approaching the problem in another way? Maybe you should think about the other direction.”

Syllabi also need to strike the right balance. “There is that question of how to even write a good syllabus nowadays,” said David Janes, professor of electrical and computer engineering and the director of the semiconductor degrees program at Purdue. “In your syllabus you need to define what is and is not appropriate use of AI, and the answer is usually that you can’t use AI. If you use AI, you are still responsible for verifying and checking it. You are responsible for independently learning and understanding these concepts.”

Fig. 1: A student and professor working in Purdue University’s Birck Nanotechnology Center. Source: Purdue

Given this complex semiconductor work, AI fluency must be grounded in engineering fundamentals, whether students are tackling network-on-chip congestion or semiconductor test.

“AI can propose a fabric configuration in seconds, but knowing whether it will hold up under real traffic patterns, achieve physical implementation, and survive routing and other tradeoffs requires that underlying domain depth,” noted Nandan Nayampally, chief commercial officer at Baya Systems. “That combination — AI plus real expertise — is where you get the actual productivity boost, not from AI alone. Without the fundamentals to validate it, AI-driven design accelerates disasters as fast as it accelerates progress.”

For semiconductor testing, students need deep, domain-specific knowledge that takes time to develop, observed Ryan Moran, head of talent acquisition at the Center of Excellence at Teradyne. “Universities that treat AI literacy as complementary to rigorous foundational training are producing the graduates we most want to hire. The more forward-thinking programs are redesigning assignments so that AI tool use is assumed. New graduates who arrive having used tools like GitHub Copilot, Claude Code, or similar assistants can produce working code drafts faster than earlier cohorts. The challenge is to maintain technical knowledge to both understand and verify what the code actually does.”

Foundational skills and deep disciplines such as physics and mathematics bring many job opportunities. “You can go very deep in linear algebra or calculus,” said Pradheepa Raman, chief people officer at GlobalFoundries. “If you go into the linear algebra side, you become a data scientist or AI engineer. If you go into the calculus side, you can become a quant in financial institutions. These old types of deep skills are in many curricula to help students become deep experts. I would prioritize that first because it’s hard for companies to teach students unless you have the basics.”

Ideally, graduates will know both science and AI. “The strongest candidates are those who know how to use AI as a productivity tool while also demonstrating solid engineering judgment, a problem-solving mindset, and a strong grasp of first principles,” said Gabriela Cruz Thompson, director of university research collaboration at Intel. “In other words, we are not looking for graduates who simply know how to prompt an AI model. We’re looking for graduates who understand the underlying science and engineering well enough to know when an AI-generated answer is correct, incomplete, or wrong.”

Producing those graduates requires assessments that reveal genuine understanding, not merely polished AI output.

AI-proof assessments
While students can enter prompts for just about anything, from emails and job applications all the way to semiconductor designs, the most valuable new skill is not prompt writing by itself. “It is the ability to use AI as an engineering assistant while maintaining technical ownership of the result,” Arteris’ Nightingale said. “The challenge for universities is to move beyond policing AI use and toward teaching responsible, verifiable engineering with AI in the loop.”

However, no universal silver bullet can prevent students from using AI to cheat. Assessments must be AI-proof, and professors need teaching assistants to help them adapt the curriculum course by course, Purdue’s Chan noted. “We need to ask the question, ‘Can AI solve all the homework problems by itself?’ If the answer is yes, then the class itself has to be rethought. Otherwise, no matter how many countermeasures you put in, it’s going to be ineffective.”

That need for course-by-course redesign is widely shared. “AI can now answer most of the questions we used to set in exams and coursework, and universities are under pressure to make assessments AI-proof,” said Axiomise’s Darbari. “That isn’t only an assessment problem. It changes how students learn and what they are equipped with when they graduate. The risk is that we produce graduates who can get answers but can’t stand behind them.”

At the same time, students must still memorize some math and physics. “Some of the higher-level concepts they have to learn by heart,” said Chan. “We still want to keep some of the paper and pencil exams so that when they go to the design stage, they will know what questions to ask. That cannot come in a day.”

Creative assessments are also key, and some faculty hold in-class competitions. “There is an engineering design goal such as building a robot, and then there are targets, and you see who performs better,” he said. “You cannot cheat by just using AI. The robot has to physically run on the road.”

Students lose out if they take shortcuts. “At the Bachelor level, mastering a course such as semiconductor devices requires a massive amount of mental effort,” said ETH Zurich’s Popovic. “When students use generative AI as a cognitive shortcut, they bypass the very struggle that builds deep engineering intuition.”

Students using AI to cheat is not viewed differently from students copying others’ code or written work. Purdue has used plagiarism checkers on graduate theses for quite some time to ensure students write their own work and cite appropriate sources, Purdue’s Janes says. “It’s a very parallel thing with AI, though we would go about it differently if it’s an engineering analysis than writing. We all struggle with the same problem.”

Physics cannot be traded for AI. “When engineers don’t know physics, they don’t know reality,” Keysight’s Ozalas said. “It’s the same with coding. If you don’t know coding and you just prompt everything, you lose the connection with the real world, and that’s a concern.”

Ultimately, traditional tests may still be best for fundamental knowledge. “Then provide more workshops, more projects, more labs, but definitely allow them to use the AI tools,” said GF’s Raman. “If you don’t, it’s like saying, ‘Hey, I have all these tools, but I want you to go the traditional way even though it will take enormous time.’ Time is also of the essence. They can learn more if they are allowed to use the tools.”

All the company’s summer interns had access to AI tools, Raman noted. “They were amazing using these tokens. They were happy to get access and build things. It was awesome to see.”

Given the direction the industry is heading, denying students access to AI tools is pointless, even though the industry still hasn’t decided on the best approach to manage access. “From the AI perspective, you do need to build AI,” said Dana Neustadter, senior director of product management at Synopsys. “Everything will have AI built in as a technology eventually, so you need AI security specialists to make sure that they build secure solutions for AI. AI tools might change the job landscape, but I don’t think positions will go away. It’s just that they’re going to change in their nature. To keep up with the technology, you’ll have to use AI tools and rely on AI assistance. I’ve heard from academia that there is a need for having specialized skill sets.” However, she added that the exact nature of these skills is still evolving.

This also means students need both hard knowledge and soft communication skills. “The graduates who do well in the AI era will be the ones who can use the tools, defend their reasoning, and communicate it clearly in the moment,” said Darbari. “Universities that build all of that into their curriculum will produce the engineers the semiconductor industry most needs.”

Darbari’s tips for universities:

• Guide students through the new way of learning.
• Make students defend what they know.
• Focus on applying knowledge, not recalling it.
• Polish communication, especially spoken communication.
• Build breadth and depth that don’t depend on AI.
• Invest in expert academics.

Table of University Approaches

The pace and depth of change in university curricula vary considerably across institutions. Both academia and industry benefit from open discussions about navigating the AI era.

Swipe left to view all five columns →

Related Articles
Can A Computer Science Student Be Taught To Design Hardware?
To fill the talent gap, CS majors could be taught to design hardware, and the EE curriculum could be adapted or even shortened.

AI’s Impact On Engineering Jobs May Be Different Than Expected
Workflows and the addition of new capabilities are happening much faster than with previous technologies, and new grads may be vital in that transition.

Even With AI Inroads, Human Chip Designers Still Essential
Engineers are still needed at key points throughout the design pipeline.