I am seeing a dangerous situation happening in the tech industry, especially in VLSI. Many companies have stopped hiring freshers, and many AI silicon startups are hiring only for senior-level positions (5-10 yoe). This means in the near future, we might have fewer qualified engineers. Is this situation fine? I think companies are trying to push AI too hard without preparing the next generation. Placing a huge bet on AI. Hope it won't destroy the industry.
I recently tried to use AI to make Verilator support whatever the remaining unsupported features are. I hate to say this, but it gave me completely BS. If you don't know what is verilator, it is a open source simulator for verilog/systemverilog.
So, I got 1000$ ai credits from hyperagent, so I set up a workflow using github actions and local wsl env. So that agents can push to github, github actions triggers the workflow and a local script will pull and run the tests, so I can check if the new changes are not breaking the tool. (im not a dev so, I dont know what I wrote there but you get the idea, I suppose). I made 2 threads, one is opus 4.8 and another one is gpt 5.5. I made opus to code and gpt to review it. At the end, I got a half-baked tool and complete tokens burnt.
I had to babysit both of them; they constantly gave up and said whatever we did until now is good enough, we can stop here and submit for public release (I kept the repo as private).
At the end, it checked the main source code and said that whatever we built was not complete and impossible to build on it anymore (I really wish I had taken a screenshot; Claude removed my older messages, and it was gone). The main source code had already implemented these things in a different way (like wtf!!!).
I mean what I felt is these agents are good, but they are mostly good at webdev, Python at max, the Verilator is written in c++, and it was giving up whenever I was telling it to build this feature, but when I gave some suggestions, it was like "ahh!! we can do that". I was trying to make it run a verification IP. Whenever something failed, it was blamed on the VIP for it.
But yeah, I may not have given it exactly what to do, but if I gave many clear instructions, I should have just done the coding myself. I wonder how these companies will use loop engineering. I wish them the best of luck.
Just wanted to share how I wasted 1000$ credits, good thing it was given for free.