DSA still matters
Despite AI tooling, FAANG interviews continue to use data‑structures and algorithms to test reasoning, edge‑case thinking and communication rather than just code generation. Social posts emphasise that interviewers want concrete trade‑off reasoning—like choosing heaps versus hashmaps at scale—so candidates still need strong algorithmic foundations. (x.com)
If artificial intelligence could replace coding interviews, the biggest tech companies would have stopped asking candidates to reverse linked lists and walk through graph problems by now. Instead, Google job listings in April 2026 still ask for experience with data structures and algorithms, and Microsoft’s own interview guide still tells candidates to know when to use hash maps, trees, and graphs. (google.com) (microsoft.com) A data structure is just a way to organize information, like choosing between a filing cabinet, a stack of trays, or a labeled set of boxes. An algorithm is the recipe for using that structure, like deciding whether to sort first, scan once, or keep the top 10 items in a heap. (microsoft.com) That is why these interviews have survived the wave of code generators. Microsoft says interviewers want candidates to explain “when to use each one” and describe the “pros and cons,” which turns the exercise from typing code into defending choices under constraints. (microsoft.com) Amazon says most technical interviews still include coding and system design whiteboarding exercises, and its 2026 software development engineer postings still list data structure implementation and basic algorithm development as expected skills. The company is testing whether you can build a solution from scratch when the prompt is messy, not whether you can paste a perfect answer on the first try. (amazon.jobs 1) (amazon.jobs 2) Google’s hiring pages show the same pattern at multiple levels. A Software Engineer III posting and several other current roles list “data structures and algorithms” directly in the qualifications, which means the screening system itself is still filtering for that foundation before any team match happens. (google.com 1) (google.com 2) The reason is scale. A hash map can find a key fast, but it will not keep your largest 100 items in order; a heap can keep the top items ready, but it is worse for exact lookups, so the right answer depends on whether the system is ranking millions of events or checking one user ID. (microsoft.com) Artificial intelligence makes this tradeoff discussion more important, not less important. OpenAI’s SWE-Lancer benchmark, published on February 18, 2025, was built from more than 1,400 real freelance software tasks because real engineering work is bigger than code completion and includes choosing between competing implementations. (openai.com) OpenAI said on February 23, 2026 that SWE-bench Verified no longer cleanly measures frontier coding ability because the benchmark is increasingly contaminated, which is a reminder that getting a model to solve familiar coding tasks is not the same as proving a human can reason through a new one in a live interview. (openai.com) That is why candidates still get pushed to talk out loud. Microsoft tells interviewees to ask clarifying questions, make a plan before implementing, and manage a 45-minute round, because the company is grading ambiguity handling and communication at the same time as correctness. (microsoft.com) The practical shift is not “ignore artificial intelligence” but “use it after you know the basics.” If a candidate cannot explain why a breadth-first search explores level by level, or why a heap beats full sorting for a rolling top-k problem, the generated code will not save them when the interviewer changes one constraint halfway through. (microsoft.com)
