Qualixar film · 31:24
Should you still learn DSA in 2026 when AI can solve the puzzles? I tested it instead of giving you an opinion: DSA vs AI, three rounds, with the rules written down first. AI won 3-0, and the student's two years still weren't wasted. Here's why, and what to practise tonight.
0:00 Three Cents
3:01 One Elevator
9:27 Who Checks?
13:07 What The Nights Built
16:58 The Real Job
25:14 The Fire
27:33 Would You Get In?
What was tested
Round 1: one of the strongest AI models answered two hard coding puzzles, 50 times each. All 100 answers passed thousands of test cases. On the newer puzzle, each answer cost about three cents.
Round 2: I changed a plus to a minus in a correct answer. A fresh AI found the bad line in 25 of 25 reads, with no false alarms in 25 reads of the correct code.
Round 3: four teams of AI agents built elevator software, one change at a time, with the same AI, the same job and the same two and a half hours. Two worked alone and passed 5 of 5 versions. Two also got notes written by a separate AI told to think like an architect, and passed 3 of 5.
What this doesn't show
This is one project and a four-run case study from November 2026. Four runs can't say why the teams with notes did worse, or prove the notes caused it. Earlier versions of the test had a money limit, a limit on rounds and a hole in the brief, and some teams stopped short, so I fixed those. A practice team then finished all five before the final test was sealed. None of those runs count here. People wrote the brief, the limits, the rules and the referee, and the AI teams wrote the code. The referee caught 15 of 17 planted bugs, and 6 of 9 in another team's code, so a pass only means the days it tried. The fire is an un-scored simulation. Vamika, the student in the film, is made up.
What to do next
Keep the gym and close the notebook of answers. Solve one puzzle a day by hand. Then try to break the AI's answer with your own slow checker. Build one small real thing with agents, after writing down on paper what right means. In an interview, show the bug your check caught. This work is called AI Reliability Engineering.
Run the test yourself: https://github.com/qualixar/dsa-vs-ai
Sources
Andrej Karpathy, Sequoia Ascent 2026: https://karpathy.bearblog.dev/sequoia-ascent-2026/
Chase and Simon (1973), Perception in Chess: https://iiif.library.cmu.edu/file/Simon_box00005_fld00354_bdl0001_doc0001/Simon_box00005_fld00354_bdl0001_doc0001.pdf
McKeithen et al. (1981), Knowledge Organization and Skill Differences in Computer Programmers: https://deepblue.lib.umich.edu/items/1d548813-cc8c-4d00-bc3b-ef1cbd11ed0d
Anthropic's study of 52 developers, one small study: https://www.anthropic.com/research/AI-assistance-coding-skills
Stanford Digital Economy Lab, 12 Aug 2026, a gap that doesn't prove the cause: https://digitaleconomy.stanford.edu/news/canariesaug26/
Website: https://qualixar.com
#DSA #DSAvsAI #AIReliabilityEngineering
Connecting official YouTube player…
Press play in the official YouTube player. Playback is never started automatically.
Watch on YouTube ↗Evidence status