Something Strange Is Happening to Developers Who Work With AI — Transcript
Full transcript
- 0:00This AI just saved
- 0:01you three weeks of
- 0:02hard work.
- 0:03It wrote two thousand
- 0:04lines of code in
- 0:05thirty seconds.
- 0:06But if you click
- 0:07merge right now, you
- 0:08are making a huge
- 0:09mistake.
- 0:10You are actively destroying
- 0:11the exact neural pathways
- 0:13that make you a
- 0:14senior engineer.
- 0:15Imagine it is a
- 0:16beautiful Monday morning.
- 0:18You open your laptop.
- 0:19You ask your AI
- 0:20assistant to build a
- 0:21new feature.
- 0:22You burn some tokens.
- 0:23The AI gives you
- 0:25a clean pull request
- 0:26The tests pass.
- 0:27The code looks reasonable.
- 0:29You have five minutes
- 0:29before your next meeting.
- 0:31So you click merge.
- 0:33That feels amazing.
- 0:34You just did a
- 0:35week of work in
- 0:36five minutes.
- 0:36But what exactly are
- 0:38you checking?
- 0:39And how do you
- 0:40know you still understand
- 0:41the change?
- 0:42Software development has always
- 0:43been a heavy mental
- 0:45job.
- 0:46You have to keep
- 0:46strict rules in mind.
- 0:48We build mental models
- 0:49of new systems.
- 0:51We recognize patterns.
- 0:52We make predictions.
- 0:54We decide what evidence
- 0:56we can trust.
- 0:57AI has changed how
- 0:58fast we write code.
- 1:00But our brains still
- 1:01need to understand the
- 1:02system.
- 1:04And AI makes that
- 1:05part very easy to
- 1:07forget.
- 1:08CTO of Criteo, Diarmuid
- 1:10recently wrote an editorial
- 1:12about this.
- 1:13He said engineering is
- 1:14changing fast.
- 1:15We are moving away
- 1:16from writing code line
- 1:17by line.
- 1:18We are moving toward
- 1:19defining intent.
- 1:20We provide the context.
- 1:23We validate the outcomes.
- 1:25The easier it gets
- 1:26to produce code, the
- 1:27more these human skills
- 1:29matter.
- 1:29You might wonder why
- 1:30a software engineer is
- 1:31talking about the brain.
- 1:33but
- 1:33Do not worry.
- 1:34I will break down
- 1:35these complex concepts using
- 1:36simple analogies.
- 1:38if the AI is
- 1:39doing all the heavy
- 1:40lifting, what happens to
- 1:41your brain?
- 1:42Let us talk about
- 1:43how your brain actually
- 1:44learns.
- 1:45Cognitive scientists call it
- 1:47the generation effect.
- 1:49This means you remember
- 1:50things better when you
- 1:52build them yourself.
- 1:53Think of our brain
- 1:54like a dense forest.
- 1:56When you solve a
- 1:57hard problem, you cut
- 1:58a path through the
- 1:59trees.
- 2:00It takes a lot
- 2:00of effort.
- 2:01You get lost.
- 2:02You hit dead ends.
- 2:04But eventually, you find
- 2:06the way out.
- 2:07The next time, the
- 2:08path is already there.
- 2:10You can walk it
- 2:10easily.
- 2:11This physical change in
- 2:13your brain is real.
- 2:15It is called myelination.
- 2:16It makes your brain
- 2:17signals move much faster.
- 2:19Think about playing a
- 2:20hard video game.
- 2:21You have to beat
- 2:22a tough boss.
- 2:23You die twenty times.
- 2:25You learn the boss's
- 2:26attack patterns.
- 2:28You learn exactly when
- 2:29to dodge.
- 2:30Finally, you beat him.
- 2:31You feel a huge
- 2:32sense of pride.
- 2:34Now imagine a friend
- 2:35takes your controller.
- 2:37He enters a cheat
- 2:39code.
- 2:39The boss dies instantly.
- 2:40You won the fight.
- 2:42But you did not
- 2:42learn how to play
- 2:43the game.
- 2:44If you face that
- 2:45boss again without cheats,
- 2:47you lose.
- 2:48AI is that cheat
- 2:49code.
- 2:50It wins the fight
- 2:51for you.
- 2:51But it leaves you
- 2:53completely helpless for the
- 2:54next fight.
- 2:55When you use AI,
- 2:56you skip the forest.
- 2:58The AI drops you
- 2:59in a helicopter at
- 3:00the finish line.
- 3:02You get the perfect
- 3:02result.
- 3:03But you never cut
- 3:04the path.
- 3:05Your brain does not
- 3:08change.
- 3:08Your memory does not
- 3:09grow.
- 3:10You bypass a part
- 3:11of your brain called
- 3:12the hippocampus.
- 3:13The hippocampus is your
- 3:15brain's learning center.
- 3:16It looks a bit
- 3:17like a seahorse.
- 3:19When you struggle, the
- 3:20hippocampus lights up.
- 3:21It saves the new
- 3:22information.
- 3:23When you just read
- 3:24AI code, it stays
- 3:26quiet.
- 3:27You forget the code
- 3:28by tomorrow.
- 3:29We are already seeing
- 3:30the results of this.
- 3:31By 2026, the tech
- 3:33industry faces a big
- 3:35problem.
- 3:36We call it the
- 3:37seniority cliff.
- 3:38We have automated all
- 3:39the entry-level work.
- 3:41But we also automated
- 3:42friction.
- 3:43Friction is what builds
- 3:44senior-level judgment.
- 3:46What exactly makes someone
- 3:47a senior engineer?
- 3:49It is not typing
- 3:50speed.
- 3:51It is pure judgment.
- 3:52It is knowing what
- 3:54not to build.
- 3:55You only get this
- 3:56judgment through pain.
- 3:58You spend three days
- 3:59fixing a terrible bug.
- 4:01You pull your hair
- 4:02out.
- 4:02You vow to never
- 4:04make that mistake again.
- 4:06That pain is your
- 4:07tuition.
- 4:08You pay it to
- 4:09become a senior.
- 4:10AI takes away the
- 4:11pain entirely.
- 4:12But it also takes
- 4:14away the tuition.
- 4:15If you never struggle,
- 4:16you never learn the
- 4:17hard lessons.
- 4:18Junior developers can read
- 4:20code perfectly fine.
- 4:22But they cannot map
- 4:22the system in their
- 4:23heads.
- 4:24They lack the spatial
- 4:25memory for it.
- 4:27Because they never built
- 4:27the map themselves.
- 4:29They suffer from AI-induced
- 4:31neuro-atrophy.
- 4:33Atrophy just means a
- 4:34muscle shrinks from lack
- 4:35of use.
- 4:35Your brain is that
- 4:36muscle.
- 4:37But skipping the struggle
- 4:39creates a hidden danger.
- 4:40Wait, let's take a
- 4:41quick pause.
- 4:43If you think this
- 4:43video is worth sharing
- 4:44with others, hit the
- 4:46like button to tell
- 4:46the algorithm about it.
- 4:48If you use social
- 4:49media and have friends
- 4:49in tech, please share
- 4:51this video.
- 4:52If you are not
- 4:52subscribed yet, please do.
- 4:54So you might think
- 4:55reviewing the code is
- 4:56the answer, right?
- 4:58You read the AI
- 4:59output before you merge
- 5:00it.
- 5:00But that brings up
- 5:02a whole new danger.
- 5:03Reviewing human code is
- 5:05actually fun for your
- 5:05brain.
- 5:06Humans make predictable mistakes.
- 5:08We forget to check
- 5:09for blank values.
- 5:11We use terrible names
- 5:12for things.
- 5:13When you spot a
- 5:14human error, you get
- 5:15a reward.
- 5:16Your brain gives you
- 5:17a quick hit of
- 5:17dopamine.
- 5:18Dopamine is the chemical
- 5:20that makes you feel
- 5:21good.
- 5:21But AI does not
- 5:23make simple typos.
- 5:24AI code is syntactically
- 5:26flawless.
- 5:27That means the grammar
- 5:28of code is perfect.
- 5:30The spacing is beautiful.
- 5:32It looks incredibly confident.
- 5:34But the core logic
- 5:35might be completely wrong.
- 5:37Think of it like
- 5:37tasting soup.
- 5:38If you taste a
- 5:39terrible soup, you know
- 5:41immediately.
- 5:42It is too salty.
- 5:43You spit it out.
- 5:44That is like reviewing
- 5:46human code.
- 5:47The mistakes are obvious.
- 5:49But AI code is
- 5:50like tasting a perfectly
- 5:51bland soup.
- 5:52It is not too
- 5:53salty.
- 5:54It is not too
- 5:55spicy.
- 5:56It looks like soup.
- 5:57It smells like soup.
- 5:59You taste it fifty
- 6:00times.
- 6:01Eventually, your taste buds
- 6:02just give up.
- 6:03You say it is
- 6:05Your brain just gets
- 6:06tired of looking for
- 6:07the flavor.
- 6:08Staring at the perfect
- 6:09code is dangerous.
- 6:11It drains a specific
- 6:12part of your brain.
- 6:14It is called the
- 6:14anterior cingulate cortex.
- 6:16This is your brain's
- 6:17error detection center.
- 6:19Think of it like
- 6:20a smoke alarm.
- 6:21When it sees a
- 6:22typo, the alarm rings.
- 6:24But staring at flawless
- 6:25AI code drains the
- 6:26battery.
- 6:27The alarm stops working.
- 6:29We call this Turing
- 6:31fatigue.
- 6:32You get tired of
- 6:33looking for hidden flaws.
- 6:34You see the green
- 6:35checkmarks.
- 6:37You assume the AI
- 6:38is right.
- 6:39So you rubber-stamp terrible
- 6:40logic.
- 6:41You approve code that
- 6:42ruins your project.
- 6:45And you do not
- 6:45even realize it.
- 6:47But bad logic is
- 6:48only half the problem.
- 6:50Let us talk about
- 6:50the biggest pain point
- 6:51of 2026.
- 6:53It is not the
- 6:53AI hallucinating functions.
- 6:56real problem is the
- 6:57AI hallucinating architectural
- 6:59That means it forgets
- 7:01the main goal of
- 7:01the project.
- 7:02It is like having
- 7:03a really fast intern
- 7:05who has read every
- 7:05book in the world
- 7:06but has zero common
- 7:07sense.
- 7:08Imagine you hire a
- 7:09builder to add a
- 7:10room.
- 7:11You want a simple
- 7:11wooden room.
- 7:13The builder works incredibly
- 7:15fast.
- 7:16But he uses steel
- 7:17instead of wood.
- 7:18He adds extra doors
- 7:20you did not want.
- 7:21The room stands up.
- 7:22It passes the basic
- 7:23safety test.
- 7:25But it completely ruins
- 7:26the style of your
- 7:27house.
- 7:28AI agents do this
- 7:29to your codebase.
- 7:30They silently introduce micro-dependencies.
- 7:34They drift away from
- 7:35your core design.
- 7:37Let me tell you
- 7:37a story about a
- 7:38developer.
- 7:39We will call her
- 7:39Sarah.
- 7:41Sarah uses an AI
- 7:42agent to build a
- 7:43payment system.
- 7:44She asks the agent
- 7:45to add a discount
- 7:47feature.
- 7:48The agent writes five
- 7:49hundred lines of code.
- 7:50It takes ten seconds.
- 7:52Sarah runs the local
- 7:53tests.
- 7:54The tests pass perfectly.
- 7:56She clicks merge and
- 7:57goes to lunch.
- 7:58But the AI did
- 7:59something sneaky.
- 8:00It did not just
- 8:01add a discount.
- 8:02It changed how the
- 8:03database handles currency.
- 8:05It imported a massive
- 8:07math library
- 8:08Sarah did not need
- 8:09this library.
- 8:10The old code worked
- 8:11fine.
- 8:12But the AI thought
- 8:13the new library was
- 8:14better.
- 8:15Three months later, the
- 8:16math library gets a
- 8:18security update.
- 8:19It breaks the whole
- 8:20payment system.
- 8:21Customers cannot buy anything.
- 8:23Sarah tries to fix
- 8:24it.
- 8:25But she does not
- 8:26understand the new math
- 8:27library.
- 8:28She never built the
- 8:29mental model for it.
- 8:30She is staring at
- 8:31a phantom architecture.
- 8:33A phantom architecture is
- 8:34a system nobody understands.
- 8:36The original design is
- 8:37completely gone.
- 8:39This is agentic context
- 8:40drift in action.
- 8:42It creates massive
- 8:43Technical debt
- 8:44Eventually, the project becomes
- 8:46a mess.
- 8:47Nobody can fix it.
- 8:48So how do elite
- 8:49teams actually fix this?
- 8:51They use a new
- 8:52method.
- 8:53It is called the
- 8:54adversarial audit.
- 8:55You have to change
- 8:56your job title.
- 8:57You are no longer
- 8:58a code creator.
- 9:00You are an adversarial
- 9:01scientist.
- 9:02Adversarial means you play
- 9:03against the AI.
- 9:04You do not trust
- 9:05it.
- 9:06You treat every AI
- 9:08output as a guess.
- 9:09Your job is to
- 9:10break that guess.
- 9:11Think of the adversarial
- 9:12audit like a car
- 9:13mechanic.
- 9:14A bad mechanic just
- 9:15turns the key.
- 9:17If the car starts,
- 9:18he says it is
- 9:18fixed.
- 9:19A good mechanic does
- 9:20not trust the
- 9:22car.
- 9:22He revs the engine
- 9:23hard.
- 9:24He listens for weird
- 9:25noises.
- 9:26He hits the brakes
- 9:27suddenly.
- 9:28He actively tries to
- 9:29find the hidden flaw.
- 9:30You have to be
- 9:31the good mechanic.
- 9:32Elite teams use a
- 9:34three-step routine.
- 9:35First, you use a
- 9:36concept called evals as
- 9:38code.
- 9:38Eval is short for
- 9:40evaluation.
- 9:41You spend most of
- 9:42your time writing tests.
- 9:43You write the tests
- 9:44before AI writes the
- 9:45code.
- 9:46You define the exact
- 9:47rules.
- 9:48You build a strict
- 9:49cage.
- 9:50Then you let the
- 9:51AI work inside the
- 9:52cage.
- 9:53If the AI steps
- 9:54out of line, the
- 9:55compiler yells.
- 9:57This stops the AI
- 9:58from drifting.
- 9:59Second, you hide the
- 10:00AI output.
- 10:01Do not look at
- 10:02it immediately.
- 10:03Force your brain to
- 10:04rebuild the logic from
- 10:05scratch.
- 10:06Write down the steps
- 10:07on paper.
- 10:08This creates synthetic struggle.
- 10:11It forces your brain
- 10:12to work.
- 10:13It keeps your neural
- 10:14pathways strong.
- 10:15It prevents that neuro-atrophy
- 10:17we talked about.
- 10:18Third, you review the
- 10:20AI code like an
- 10:21attacker.
- 10:21Do not look for
- 10:22typos.
- 10:23Look for logic gaps.
- 10:25Ask yourself how you
- 10:26can break it.
- 10:28Run it with weird
- 10:29inputs.
- 10:29You want the code
- 10:30to fail.
- 10:31When you break it,
- 10:33you win.
- 10:34Let us recap.
- 10:35AI gives you incredible
- 10:36speed.
- 10:37But it takes away
- 10:38the cognitive friction you
- 10:39need.
- 10:40If you just merge
- 10:40perfect code, your brain
- 10:42shrinks.
- 10:43You lose your spatial
- 10:44memory.
- 10:45You suffer from Turing
- 10:46fatigue.
- 10:47Your project suffers from
- 10:48context drift.
- 10:49To survive, you must
- 10:51become an adversarial auditor.
- 10:53Write tests first.
- 10:54Hide the output.
- 10:55Break the code on
- 10:56purpose.
- 10:57That is how you
- 10:58stay a senior engineer
- 10:59in 2026.
- 11:00If you enjoyed this
- 11:01breakdown, please hit the
- 11:03like button and subscribe.
- 11:05It really helps the
- 11:06channel.
- 11:07I will see you
- 11:07in the next one.
- 11:09Bye.
About this transcript
This page contains the full transcript of Something Strange Is Happening to Developers Who Work With AI by Alex - a1crew, generated from the public captions YouTube serves with the video. The transcript has 1,832 words across 562 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.