studying cs in 2026 life before 2022 was peak. it was back in the stone ages of coding. i am talking when stack overflow had hundreds of thousands of users daily & being a cs student was the creme de la creme. with just a couple of side projects and some leetcode problems around your belt, you could easily get internships at big tech companies, earn big bucks over the summer and land full time jobs with great compensation. securing a full-time offer post grad was not super easy but it wasn't as hard as today. i cant even say i miss those times cos i never really experienced it lol. then sam tweeted "we launched chatgpt.com, try talking to it". less than 10 words but it has changed the trajectory of everything so far. wild to think it's been just 4 years but boy, has it been a long and hard 4 years. the entire notion of what it means to be a software developer has been almost entirely redefined since then. from reading stack overflow and documentation to now prompting ai agents or as boris said "creating loops for your ai agents", it's been a dramatic change. the change has been for the better in terms of the software we can now build. some of the things that would once take months can now be done in days with ai. startups are being spun faster than we can say "jack robinson". whether an avalanche of startups building the same thing is for the good or for the worse isn't certain yet to be honest. one thing that's statistically certain at least is that the job market has dealt really badly with the change. "if ai can do it better, faster and cheaper, why not fire a bunch & give the rest access to ai" summarizes the memos companies send out when announcing layoffs. a lot of people would argue about the "better" but companies care more about the faster & cheaper anyways so they have all been cutting down on their headcount like a new vegetarian cutting down on meat. According to Trueup, 245,953 tech jobs were affected in 2025 globally. crazy number that puts it to at least 674 people per day. 2026 is looking worse with around 153,208 already. we aren't even done with the first half of the year??? all these really terrible numbers makes studying cs feel like a very very bad choice. i lowkey dont have any thoughts on whether you should study it or not tbh. it's always going to be your choice to study cs. although, i would argue that it shouldnt take people telling you before u switch. neither should layoffs as well if you ask me. the ball will always be in your court. if you still chose to study computer science, you can't keep on approaching it the same way as before. "only a fool keeps doing things the same way yet expects different results". ai has entirely changed how things work so it's only natural that the method of studying to thrive in the sector changes as well. there needs to be a change in how you maximize your 4 or 5 years in college as the case maybe. the old cs approach was to grind leetcode and learn the application layers. to ensure your resume gets selected, especially for swe roles, you learnt how to build things like apps & websites. you spend time learning frameworks like react, nextjs and a whole bunch of apis that you can use to build fancy things then slap them on your resume. to prepare for interviews, you also subscribed to the leetcode grind & you can now invert a binary tree from memory. you apply to like 20 to 30, get a bunch of interviews and then your offer arrives. sweet jesus! saying that this strategy should be abandoned would be equivalent to us completely jettisoning something that has worked countless times in the past. the goal, or at least the essence of my writing, isn't to propose forgoing it. i am not gonna join the bandwagon of those who tell you "dont do leetcode, just build stuff & opportunities will come to you". that statement is as foolish as it is wise. opportunities at big tech will never come to an undergrad from posting. and you want to get that big name on your resume. an ex-microsoft intern will get more opportunities than a "builder". unless you go to waterloo or some t10 us university, your chances of getting hired for internships at any company increases exponentially if you have a big name company on your resume. so, grind that leetcode— i am not interested in motivating anyone to do leetcode. most cs students, except for a good number of bay area students, are already pretty motivated to do so already. you cant blame the bay area students by the way. they are in close proximity to startups so they can get it easily! what we want to define is what exactly should we learn as cs students? ai is diabolically good at the application layer. anyone can spin up a fully functional website in less than 10-15 minutes with ai. just learn how to prompt and bam, you will be done. if we all know & acknowledge this, then why are you still fixated on learning react. why are you trying to be the 10000 and 1th react developer? react is just an example that came to my head by the way, it's not the problem. what is really wrong is fixating on learning smth that is already commoditized with ai. sorry to tell you this but unless you are a godlevel designer, you can't build a website better than ai that's prompted by someone with taste. just a $20 codex subscription & the person blows you out of the water. all that talk of "if you can build without ai, you build better with ai" isn't technically wrong but if a novice has access to opus 4.8 or gpt 5.5, just tell them to screenshot the type of websites they want & give it to claude. i assure you they will give you good stuff. so, the application layer, as we can all see, is commoditized already with ai. the idea of side projects doesn't really do that much to help anymore. everyone knows you can claude projects away in a blast. so the solution isn't more projects. what we need to focus on is the basics or as elon likes to call it "first principles". first principles means understanding the how & why of things. so for example, you can know that retrieval of an element in a list is o(1). first principles would be to try building one yourself and then realizing that lists are contiguous in memory. they are all on a block of memory so you can just locate any of them by going to that block. i enjoyed typing that btw. i feel so smart lol. back to business! so don't spend another weekend building an app that calls 5 apis you saw on some api marketplace. build one yourself from scratch. spend sometime watching videos on servers, open a repo and try to implement one from scratch yourself. dont use codex or some ai assistant, just dig through it yourself and learn how it works. most cs students' resume contains like 5-6 different programming languages, have you tried understanding how languages work from first principles? have you spent time on the beauty and complexity of compilers? you spend hours becoming an sql or supabase demon. maybe try building a database from scratch instead of just memorizing syntax? ilovepdf was the work of some college students when ai didn't exist at all. idk any college student who hasn't used that website tbh. first principles understanding at work. you can still ask ai when you're stuck though. "i am working on this but i dont get it, explain it to me." i just didn't tell you to ask it to build it for you! "what does all of this have to do with ai taking jobs and ai being good at coding. ai could build all these as well, no?" doing those things i mentioned is difficult. and i dont mean like difficult in trying to figure out how to scrape a website with selenium. i mean the kind of difficulty where you just wanna give up after a while & accept defeat. there are resources out there but boy, you will have to search. you'll have to dig through tons of youtube videos, articles and a whole lot of noise before you see one old blog somewhere that explains it. i am currently trying to build an http server from scratch in c++ without an ai coding tool. oh boy, its tough. i swear i've read Beej's guide twice without getting a hang of what he meant. it was like going through solutions to olympiad questions. but if you stick with those tough things, your brain becomes stronger & adapts consciously to doing hard things. you become much better at solving hard problems, and you subconsciously develop a mental model for working on tough things. for my leetcode warriors, you actually get better at it too. most of the leetcode algorithms are actually fundamental things. you actually have to walk through a binary tree when building a compiler. everybody tells you nowadays that software is over because the application layer is commoditized. the reality is software has not been eaten. what has actually been eaten is only the obvious part, the part we all see: the ones about building apps and websites. but there are still hard inner problems in software that haven't been solved, and that is what diving into the basics helps you get to. the second thing, as regards hiring: will doing this make you better for jobs? for the actual job itself, it will. you understand things from first principles instead of just vibe coding or building only at the application layer. you can break things down to the fundamental layers, pick up new hard tasks faster, and write better code, because you have a natural understanding of systems instead of just vibing your way to an output. it's easy to be deceived into thinking you dont need knowledge or experience to code, which is kind of true, but the quality of what you produce is naturally higher when you understand the fundamentals. the job market is cooked. none of this guarantees an internship or a job, nothing ever does. but if you pair leetcode with fundamental projects, system design, architecture, you stand a better chance than people who only vibe code. and the good news is you should still code with ai too. shipping fast with ai is a necessity now. pick up cursor or antigravity, learn prompt engineering. boris said the new wave is "creating loops". if you have infinite tokens like him, learn that as well. as a student you have the luxury of time most working engineers dont, so dont spend all of it vibe coding and filling your brain with ai slop. companies dont care that you built an http server from scratch specifically, its a nice project to have but not the point. what they actually care about is whether you can solve problems well, and at big tech that mostly still means leetcode. big tech hiring is gameable in a way startup hiring isnt: most startups in sf have very few interns a year and most of those spots are for waterloo students, big name schools, one kid they met at a conference or the 1001th cold emailer. google & amazon collectively take more than 10k interns yearly. cloudflare took 1111 this year. oddly specific number but bottom line is big tech takes a lot. that asymmetry alone is reason enough to keep grinding leetcode alongside the fundamentals. in conclusion, there's still a lot of problems left to solve in software. when search engines came, when facebook came, it felt like there was nothing left to build. then transformer models came, then a mixture of experts, then world models, a whole load of new technologies still advancing. software is not over, there are still new architectures to develop, still new things to build. but you never get there by staying at the application layer. so this isn't a call for less computer science. it's a call for more. more fundamentals, more basics, more of the hard things that make your brain feel dumb.