AI skills · Career · Software engineering
Researchers gave 16 experienced developers 246 real tasks, randomly allowed AI on some, and timed everything. They predicted AI would make them 24% faster. They were measured 19% slower. The interesting part: afterwards, they still believed AI had made them 20% faster. This is about you even if you have never written a line of code, because that gap between what we feel and what is true may be the most dangerous thing in your career right now.
Prefer to read? The full article is below. Watch on YouTube.
The short version
I will not sell you an outdated number. That study used tools from early 2025. The same researchers re-ran it with newer 2026 tools and the result came out roughly flat; they now think AI probably does help. So the 19% is history. Do not quote it to your boss.
But the gap between what people feel and what is actually true keeps appearing in study after study. Your sense of your own improvement is not evidence.
I run a company and I no longer write code by hand. When these tools got good, I built things in a weekend that would have taken far longer before. It genuinely felt like a superpower, and I will not pretend otherwise.
Then I went back and properly read what I had built. Some was fine. Some was messy. Some was straight-up unsafe, the kind of thing that would have been a serious problem if the wrong person had looked at it. The AI did not lie to me. It gave me exactly what I asked for. The problem was that I did not know enough to ask the right questions, or to spot what was missing.
And AI does not take responsibility. If what I shipped breaks, nobody calls Claude to complain. They call me. My name is on it. That is not a coding lesson. It is a lesson about accountability.
“AI can generate anything. It cannot be accountable for anything.”
The popular line is that AI makes lazy people look good for a while until they get exposed. The data says something different. Junior developers using AI see speed gains of roughly 27% to 39%; seniors around 8% to 16%. AI helps less experienced people more, and structurally, not temporarily. It is an always-available mentor that never gets annoyed by basic questions. That is genuinely good. I am not going to complain that the ladder got easier to climb.
But look at what it means. On raw output, how fast something appears on screen, the gap between a beginner and a professional got much smaller. If your career plan is "I am faster than a junior", that plan is dead. It was never a moat, and it certainly is not one now.
So where did the senior-junior gap go? Into checking. Security research suggests that somewhere around 30% to 40% of AI-generated code snippets contain at least one known class of vulnerability. One study of 304,000 AI-written commits across 6,200 projects found AI-introduced security problems had a 41% survival rate, meaning many were never caught. Meanwhile, one analysis of 211 million lines of code found the average developer pushed 75% more code in 2025 than in 2020.
Put those together: 75% more code, but nobody hired 75% more reviewers. The bottleneck was never typing. It was always knowing whether what was typed is correct, and that part is untouched. That is your gap. Not speed. Judgement.
This is not just coding. McKinsey found around 46% time savings on routine tasks and under 10% on complex work. AI eats the routine. The complex part barely moved, and the complex part is where the money is, in every job.
If you are under 25, sit with this one. A study on skill formation found juniors using AI scored 17% lower on comprehension tests. They produced more and understood less.
Here is the trap: you will not notice. You ship without really reviewing. Your manager is happy. Your output looks identical to someone who understood it. For a year, maybe two. Then something breaks in a way AI cannot explain. A client asks why. Or you sit in an interview with someone hiring for real skill, they ask one follow-up question about your own codebase, and there is nothing behind the answer.
I interview people. I do not care whether you used AI; I want you to. But I always ask one extra question: why did you do it this way? I have watched many confident people fall apart on that single question, because they never made the decision. They accepted what AI gave them. Developers know this: in one of the biggest industry surveys, favourable views of AI tools fell from 70% to 60% in a year, 46% said they do not trust the output, and the top complaint (66%) was code that is almost right. Almost right is the most expensive output there is, because finding the flaw costs more than writing it properly. We explore this in how to explain code you built with AI.
Between 1 January and 12 June this year, 42,807 Malaysians lost their jobs according to PERKESO, and that only counts people who registered. To be honest about cause, the data suggests AI is not currently a major driver of these retrenchments; businesses closing and downsizing are. Meanwhile, 605,000 jobs were advertised nationwide, and TalentCorp's critical occupations list for 2024/2025 names 66 occupations across 18 sectors where demand exceeds supply, with software developers and ICT roles among them.
So the jobs are not gone. Nobody in Malaysia is replacing developers with a chatbot in 2026. The part of the job AI does is now worth close to nothing because everyone has it. Carrying the consequences and being accountable for the work is worth more than ever.
Four practical moves, whether you code or not:
01
Stop measuring yourself by output
How much you ship says nothing about you any more; it went up for everyone. If output is your whole case for being employable, you have no case.
02
Build the review muscle on purpose
Every time you accept AI output, answer two questions first: what would break this, and why this approach instead of another? Ninety seconds that separate two years of experience from one year repeated twice.
03
Juniors: go slow on purpose
Once a week, do one thing without AI. Not for productivity, but for comprehension. You are buying back the understanding that the 17% gap says you would otherwise lose.
04
Get outside signal
Feelings are not evidence. Does anyone pay more for your work? Do seniors trust you with harder things? Can you explain your decisions under pressure?
AI did not make average people permanently look good. It made them faster, and that is not going away. What it did was take what most people were selling, output, and make it nearly free. If that was your whole value, the problem is not a few years away. It is now.
If your value is judgement, taste and being willing to put your name on the outcome, you have just been handed the tool of your life, and half your competition deleted their own advantage without noticing. Everyone who takes on more accountability gets more valuable as the tools improve. That is the one strategy I would bet on.
Building that judgement in software is exactly what our AI-Native Software Development Programme is designed around: you build with AI, then mentors make you explain and defend every decision. You can try the first lessons in the free trial. And be honest with yourself: when did you last ship something with AI that you could not explain? For me, it was two months ago.
It depends. An early-2025 study found experienced developers were 19% slower with AI while believing they were faster; a later run with newer tools came out roughly flat. Other data shows juniors gain more speed (around 27% to 39%) than seniors (around 8% to 16%). The consistent finding is that people's sense of their speed is unreliable.
Not by default. Security research suggests roughly 30% to 40% of AI-generated code snippets contain at least one known vulnerability class, and many AI-introduced issues survive into production. AI code needs the same careful review as human code, and more of it is being produced, so review skill is increasingly valuable.
It can. One study found juniors using AI scored 17% lower on comprehension tests despite producing more. Using AI is fine, but juniors should deliberately review and question its output and sometimes work without it, so they build the understanding needed to debug and explain their work.
Judgement and accountability: knowing whether output is correct, what could break it, why one approach beats another, and being willing to own the result. Output itself has become cheap because everyone has AI. The ability to check, decide and explain is where value has moved.
Your next step
Our 12-week AI-Native Software Development Programme is a live, mentor-reviewed cohort where you build and ship real products with AI, and learn to explain every line. Start with the free trial to see how we teach, then decide.
Whatever you end up doing, the people who understand how software and AI actually work get to direct them. Try the first lessons free, no card needed, or see the full 12-week AI-Native Software Development Programme.
