Semiconductor Engineering 圆桌：AI 时代电气工程毕业生需求未降但要求在变

Key takeaways:

• There is no drop in demand for electrical engineering graduates, but the demands on their education and work experience are changing.
• Agents are now doing many tasks that were considered the work of junior engineers, but there are new higher-level tasks for them to do.
• There is an opportunity for junior engineers to help retrain senior engineers in the use of AI.

Experts at the Table: Semiconductor Engineering sat down to discuss what is happening with the latest college graduates in the face of AI adoption, with Chuck Alpert, fellow at Cadence; Thomas Andersen, vice president of AI and ML at Synopsys; Simon Davidmann, AI and EDA researcher at Southampton University; Nagesh Gupta, CEO of llmda.ai; Dave Kelf, CEO of Breker Verification Systems; and Kexun Zhang, Head of Research from ChipAgents. What follows are excerpts of that discussion. This is the first article in a three-part series.

Clockwise from top left: Southampton University’s Davidmann; Breker’s Kelf; Synopsys’ Andersen; ChipAgents’ Kexun; llmda’s Gupta; and Cadence’s Alpert.

SE: AI is disruptive to EDA, the semiconductor industry, and to their customers. One early indication is that computer science (CS) grads are struggling to find jobs. In the past, they were highly sought after. Today, they have moved to a higher unemployability rate than the general averages. What’s going on?

Gupta: From my observations, data science is much more significantly affected compared to computer science. I’ve been looking for computer science grads, but from good universities. There are not that many good CS grads available. It’s more of the data science that seems to be affected.

Alpert: We know that the vibe-coding tools — Cursor, Claude Code, GitHub Copilot — are improving the productivity of computer scientists, and so it’s natural there will be a hiring slowdown. Companies are spending a lot of money on Claude and OpenAI for these things, and they don’t want to hire quite so many engineers at quite the same rate.

Andersen: When we talk about computer scientists, or folks in our industry, there are two aspects to consider. There are folks that write EDA software, and then there are chip design companies. I would agree that when it comes to software development, everybody’s a little cautious because they want to see how much AI can automate things. At the same time, when we work on new technology and new releases, everyone wants to know how you can do it faster. If I can do things faster and more efficiently, if I can make releases every three months and develop software faster, that is definitely beneficial. On the chip designer side, it’s not quite at the same level because they are not like traditional software programmers. They are primarily experts in chip design, and they are users of tools. There’s been a shortage of those people for quite some time. To some degree, AI is helping address that shortage, but it’s not that humans are going away. Humans are just elevating the tasks that they’re doing, and can be more efficient, and then they can develop more chips. How long does it take to get a chip out the door? If somebody says I can do it in half the time, everybody would love it.

Alpert: I’m guessing none of this applies to electrical engineers (EE) because there is still a huge gap in demand for design expertise. I don’t know if there’s a slowdown there.

Davidmann: I did some digging around for the UK market, and they reckon that the graduate market is difficult, but AI is not the only cause. They said that from 2022 to 2026, graduate recruitment among leading employers fell 24%, but that is about economic uncertainty, reduced budgets, corrections, and post-pandemic hiring changes. AI has come into an area that was already complicated. It’s not easy to work out what effect AI is having. From an engineering point of view, they said that graduate engineers are finding employment within 15 months of graduating. So 75% of them got jobs, and that’s pretty typical. It is definitely having some effect, but it’s very hard to pinpoint what the actual AI impact is on changing employment because there are so many other things going on.

Kelf: I ran AI to ask this question and got similar answers. I was talking to someone who’s leaving college and trying to find a job. They can’t find one. The problem is a lack of practical experience, and we’ve all seen this over the years. We’re trying to hire people, and if you hire a new grad, you’ve got to spend a year to get them up to speed on real designs. That seems to be getting worse. Something we have seen is that AI is somewhat replacing the need for interns, and so finding internships seems to be getting harder, and therefore they’re getting even less practical experience. The colleges aren’t really helping with that. When we hire a new grad, we’ve got to spend even more time training them. There’s also a clear financial question. If we can put AI on some of the more junior tasks — tasks that you would have used to train new hires — then we can’t afford to hire them anymore. If we hire them and our competitors are using AI, then we have a problem with profitability versus our competitors. We’ve got to keep up with those.

Davidmann: That is a short-term problem for graduates because AI only does certain things. There are many skills that you need to learn and put in practice for real engineering, which AI doesn’t really help you with. If you don’t have graduates coming in and apprenticeships, you will have a problem because they never become the experts. And as your experts get old, then you’ve got this big gap, which AI isn’t going to fix.

Andersen: I completely agree. While you can say I don’t need the junior guys anymore because AI can do those simple tasks, if I don’t replace the existing experts, who’s ever going to do it?

Gupta: From my own experience, my daughter is a recent CS grad, and fortunately she did find a job. I was talking to her about the other CS grads, and she told me pretty much everyone in the graduating class got jobs. But she made a very interesting observation. Previously, if you were a new grad, you would get assigned to a mentor who trained you and got you up to speed. What is happening today is that all the senior folks in the company are using AI to do a lot of their work. The young grads are not able to get time from the experienced folks to train them. This is a big problem, and the large company she is working for realized this. They have started doing a focus group to figure out how to solve this problem, where nobody is interested in taking on the younger grads and getting them up to speed. That seems to be a key problem.

Zhang: There is a big mismatch between what people expect new grads to do when they join the company in the AI era, compared to how they’re trained either in schools, universities, or the internship programs. AI is evolving so fast, and we are now expecting the new hires, the new grads to be AI-native, to be able to use AI tools to solve problems at a higher level. We’re not expecting them to do the traditional grunt work that junior engineers or interns did. We’re expecting them to solve higher-level problems with taste, with their own initiative for defining problems. Solving these problems and implementing the solution has become much easier, but university education hasn’t really adapted, and it’s impossible for it to adapt this fast given the pace of AI and its evolving ability.

Gupta: This is especially true for the recent college grads, because they’ve been in this transition.

Zhang: Universities are trying to adapt, and the recent grads are the generation who are in this gap. Most of their education has not been AI-native, but the companies that are seeing the economic value of using AI have already changed their expectations for new grads.

Davidmann: Being back in the university scene, there’s schizophrenia in the universities about what they do with AI, about how much they let people use AI, and what type of assessments they apply. They need to build assessments that can’t be done by AI, where the engineer, the student, has to have learnt the skills. I have two views on it. Education has to change and separate between two types of teaching, a foundation mode and a professional mode. The professional mode is what you expect them to do in a company. This means using AI, using the agents as part of the job. Then there’s the foundation mode where you need to know the mathematics of things, how the algorithms work, and how you debug things. There are areas where you shouldn’t be using AI, because you need the understanding skills. But that is of no use in the industry because you’re going to be using agents and AI. You really need to have these two things. While working in one of the large EDA companies, I looked after the university program in Europe. One thing we did was to lend them technology for verification and give them a project to do with coverage. First they got to understand the problem, manually got the simulation going, the test generators going, the coverage going. They did all of this manually and did manual tests. They learnt the whole process, and then they tried to use AI to improve it. It was interesting to listen to them present their project. Education has to change with the use of AI. In the same way, companies are learning how to make use of AI and how the staff can use it. The educators, the lecturers, have to understand it, and these policies aren’t sorted. Similarly, they’re not in companies. It’s all changing very quickly.

Alpert: I’m finding that a lot of the junior engineers, people who have been hired within the last two years, are some of the best experts to rely on for AI expertise. It’s really exciting for EDA because we have a lot of mature people. Three to five years ago, if I had asked what a young engineer could teach me, the answer would have been nothing. But today I can. I learn a lot from talking to these guys.

Andersen: I fully agree with that. With a lot of the established folks, there’s some resistance to using new technology. That happens with every generation. It’s also true that maybe there is a gap, in the sense that universities have not prepared the current graduates for these technologies. But in general, younger people are very eager to pick up new stuff. They can be self-learning, and that’s a big asset.

Kelf: There have been waves within EDA. Think about logic synthesis. In the past, when synthesis started coming in, senior engineers were working at the gate level. They were so used to that, and they were good at it. They were resistant to picking up synthesis, and it was the new grads and new engineers who figured out HDLs, and they were the ones who started using synthesis. We saw it again when randomized verification was introduced. The senior verification guys weren’t used to using test benches, which required expertise in object-oriented programming, but the new grads were learning that at school. The question with AI is what they could learn about AI to do these jobs more practically compared to senior engineers who are having to figure this out by themselves. I wonder if there’s an expectation that the kids coming out of school will be experts in AI because they’re learning it, but they’re not there yet. Maybe in another year or two they will start having much more intuitive AI courses, and they’ll be able to figure things out like hallucinations and other things that the senior engineers will have trouble figuring out. Maybe this is the next wave.

SE: It has been said that this won’t affect the EEs, that it’s the CS grads who are impacted. But people within the EDA industry are saying that the current agents are as good as junior engineers. They also say we are going to have senior engineers managing 10 of these agents. They call them virtual engineers rather than people. This says to me that EEs are going to get impacted in the same way.

Alpert: There’s such a huge productivity gap, and the designs are more complex. I’ve been hearing that everything is 10× more complex because we have to do the packaging and 3D-ICs, and we have to verify all these different fields, and we have to do drop tests. There is so much more to do to get things out the door. This is over and above doing 10× more designs with no more resources. The timeframes are the same. EDA has always been about productivity. That’s what we do as an industry. Now we need to accelerate that with agents, and we can. But that’s just another productivity gain that’s going to help close the gap. I still don’t think it’s going to be closed. The number of designs people want to do means there’s never going to be a shortage of EE’s, even if designers are 10× more productive.

Andersen: There are many studies that show there is a talent shortage. When it comes to chip designers, there’s definitely more demand than supply, and that means it would be improved by AI, where people can be more effective. When we have terms like junior versus senior engineer, it simply describes the tasks that they’re doing. If the current task of a junior engineer is being automated, then they do a higher-level task. That doesn’t necessarily mean they need to spend 20 years getting experience so that they can do the higher-level tasks. They just need to level up from the beginning.

Alpert: All engineers are now senior engineers. That’s what we mean. When you hire someone, you get some help.