YouTube2Text

Something Strange Is Happening to Developers Who Work With AI — Transcript

by Alex - a1crew · 1,832 words · 562 segments · language en · Watch on YouTube

Full transcript

  1. 0:00This AI just saved
  2. 0:01you three weeks of
  3. 0:02hard work.
  4. 0:03It wrote two thousand
  5. 0:04lines of code in
  6. 0:05thirty seconds.
  7. 0:06But if you click
  8. 0:07merge right now, you
  9. 0:08are making a huge
  10. 0:09mistake.
  11. 0:10You are actively destroying
  12. 0:11the exact neural pathways
  13. 0:13that make you a
  14. 0:14senior engineer.
  15. 0:15Imagine it is a
  16. 0:16beautiful Monday morning.
  17. 0:18You open your laptop.
  18. 0:19You ask your AI
  19. 0:20assistant to build a
  20. 0:21new feature.
  21. 0:22You burn some tokens.
  22. 0:23The AI gives you
  23. 0:25a clean pull request
  24. 0:26The tests pass.
  25. 0:27The code looks reasonable.
  26. 0:29You have five minutes
  27. 0:29before your next meeting.
  28. 0:31So you click merge.
  29. 0:33That feels amazing.
  30. 0:34You just did a
  31. 0:35week of work in
  32. 0:36five minutes.
  33. 0:36But what exactly are
  34. 0:38you checking?
  35. 0:39And how do you
  36. 0:40know you still understand
  37. 0:41the change?
  38. 0:42Software development has always
  39. 0:43been a heavy mental
  40. 0:45job.
  41. 0:46You have to keep
  42. 0:46strict rules in mind.
  43. 0:48We build mental models
  44. 0:49of new systems.
  45. 0:51We recognize patterns.
  46. 0:52We make predictions.
  47. 0:54We decide what evidence
  48. 0:56we can trust.
  49. 0:57AI has changed how
  50. 0:58fast we write code.
  51. 1:00But our brains still
  52. 1:01need to understand the
  53. 1:02system.
  54. 1:04And AI makes that
  55. 1:05part very easy to
  56. 1:07forget.
  57. 1:08CTO of Criteo, Diarmuid
  58. 1:10recently wrote an editorial
  59. 1:12about this.
  60. 1:13He said engineering is
  61. 1:14changing fast.
  62. 1:15We are moving away
  63. 1:16from writing code line
  64. 1:17by line.
  65. 1:18We are moving toward
  66. 1:19defining intent.
  67. 1:20We provide the context.
  68. 1:23We validate the outcomes.
  69. 1:25The easier it gets
  70. 1:26to produce code, the
  71. 1:27more these human skills
  72. 1:29matter.
  73. 1:29You might wonder why
  74. 1:30a software engineer is
  75. 1:31talking about the brain.
  76. 1:33but
  77. 1:33Do not worry.
  78. 1:34I will break down
  79. 1:35these complex concepts using
  80. 1:36simple analogies.
  81. 1:38if the AI is
  82. 1:39doing all the heavy
  83. 1:40lifting, what happens to
  84. 1:41your brain?
  85. 1:42Let us talk about
  86. 1:43how your brain actually
  87. 1:44learns.
  88. 1:45Cognitive scientists call it
  89. 1:47the generation effect.
  90. 1:49This means you remember
  91. 1:50things better when you
  92. 1:52build them yourself.
  93. 1:53Think of our brain
  94. 1:54like a dense forest.
  95. 1:56When you solve a
  96. 1:57hard problem, you cut
  97. 1:58a path through the
  98. 1:59trees.
  99. 2:00It takes a lot
  100. 2:00of effort.
  101. 2:01You get lost.
  102. 2:02You hit dead ends.
  103. 2:04But eventually, you find
  104. 2:06the way out.
  105. 2:07The next time, the
  106. 2:08path is already there.
  107. 2:10You can walk it
  108. 2:10easily.
  109. 2:11This physical change in
  110. 2:13your brain is real.
  111. 2:15It is called myelination.
  112. 2:16It makes your brain
  113. 2:17signals move much faster.
  114. 2:19Think about playing a
  115. 2:20hard video game.
  116. 2:21You have to beat
  117. 2:22a tough boss.
  118. 2:23You die twenty times.
  119. 2:25You learn the boss's
  120. 2:26attack patterns.
  121. 2:28You learn exactly when
  122. 2:29to dodge.
  123. 2:30Finally, you beat him.
  124. 2:31You feel a huge
  125. 2:32sense of pride.
  126. 2:34Now imagine a friend
  127. 2:35takes your controller.
  128. 2:37He enters a cheat
  129. 2:39code.
  130. 2:39The boss dies instantly.
  131. 2:40You won the fight.
  132. 2:42But you did not
  133. 2:42learn how to play
  134. 2:43the game.
  135. 2:44If you face that
  136. 2:45boss again without cheats,
  137. 2:47you lose.
  138. 2:48AI is that cheat
  139. 2:49code.
  140. 2:50It wins the fight
  141. 2:51for you.
  142. 2:51But it leaves you
  143. 2:53completely helpless for the
  144. 2:54next fight.
  145. 2:55When you use AI,
  146. 2:56you skip the forest.
  147. 2:58The AI drops you
  148. 2:59in a helicopter at
  149. 3:00the finish line.
  150. 3:02You get the perfect
  151. 3:02result.
  152. 3:03But you never cut
  153. 3:04the path.
  154. 3:05Your brain does not
  155. 3:08change.
  156. 3:08Your memory does not
  157. 3:09grow.
  158. 3:10You bypass a part
  159. 3:11of your brain called
  160. 3:12the hippocampus.
  161. 3:13The hippocampus is your
  162. 3:15brain's learning center.
  163. 3:16It looks a bit
  164. 3:17like a seahorse.
  165. 3:19When you struggle, the
  166. 3:20hippocampus lights up.
  167. 3:21It saves the new
  168. 3:22information.
  169. 3:23When you just read
  170. 3:24AI code, it stays
  171. 3:26quiet.
  172. 3:27You forget the code
  173. 3:28by tomorrow.
  174. 3:29We are already seeing
  175. 3:30the results of this.
  176. 3:31By 2026, the tech
  177. 3:33industry faces a big
  178. 3:35problem.
  179. 3:36We call it the
  180. 3:37seniority cliff.
  181. 3:38We have automated all
  182. 3:39the entry-level work.
  183. 3:41But we also automated
  184. 3:42friction.
  185. 3:43Friction is what builds
  186. 3:44senior-level judgment.
  187. 3:46What exactly makes someone
  188. 3:47a senior engineer?
  189. 3:49It is not typing
  190. 3:50speed.
  191. 3:51It is pure judgment.
  192. 3:52It is knowing what
  193. 3:54not to build.
  194. 3:55You only get this
  195. 3:56judgment through pain.
  196. 3:58You spend three days
  197. 3:59fixing a terrible bug.
  198. 4:01You pull your hair
  199. 4:02out.
  200. 4:02You vow to never
  201. 4:04make that mistake again.
  202. 4:06That pain is your
  203. 4:07tuition.
  204. 4:08You pay it to
  205. 4:09become a senior.
  206. 4:10AI takes away the
  207. 4:11pain entirely.
  208. 4:12But it also takes
  209. 4:14away the tuition.
  210. 4:15If you never struggle,
  211. 4:16you never learn the
  212. 4:17hard lessons.
  213. 4:18Junior developers can read
  214. 4:20code perfectly fine.
  215. 4:22But they cannot map
  216. 4:22the system in their
  217. 4:23heads.
  218. 4:24They lack the spatial
  219. 4:25memory for it.
  220. 4:27Because they never built
  221. 4:27the map themselves.
  222. 4:29They suffer from AI-induced
  223. 4:31neuro-atrophy.
  224. 4:33Atrophy just means a
  225. 4:34muscle shrinks from lack
  226. 4:35of use.
  227. 4:35Your brain is that
  228. 4:36muscle.
  229. 4:37But skipping the struggle
  230. 4:39creates a hidden danger.
  231. 4:40Wait, let's take a
  232. 4:41quick pause.
  233. 4:43If you think this
  234. 4:43video is worth sharing
  235. 4:44with others, hit the
  236. 4:46like button to tell
  237. 4:46the algorithm about it.
  238. 4:48If you use social
  239. 4:49media and have friends
  240. 4:49in tech, please share
  241. 4:51this video.
  242. 4:52If you are not
  243. 4:52subscribed yet, please do.
  244. 4:54So you might think
  245. 4:55reviewing the code is
  246. 4:56the answer, right?
  247. 4:58You read the AI
  248. 4:59output before you merge
  249. 5:00it.
  250. 5:00But that brings up
  251. 5:02a whole new danger.
  252. 5:03Reviewing human code is
  253. 5:05actually fun for your
  254. 5:05brain.
  255. 5:06Humans make predictable mistakes.
  256. 5:08We forget to check
  257. 5:09for blank values.
  258. 5:11We use terrible names
  259. 5:12for things.
  260. 5:13When you spot a
  261. 5:14human error, you get
  262. 5:15a reward.
  263. 5:16Your brain gives you
  264. 5:17a quick hit of
  265. 5:17dopamine.
  266. 5:18Dopamine is the chemical
  267. 5:20that makes you feel
  268. 5:21good.
  269. 5:21But AI does not
  270. 5:23make simple typos.
  271. 5:24AI code is syntactically
  272. 5:26flawless.
  273. 5:27That means the grammar
  274. 5:28of code is perfect.
  275. 5:30The spacing is beautiful.
  276. 5:32It looks incredibly confident.
  277. 5:34But the core logic
  278. 5:35might be completely wrong.
  279. 5:37Think of it like
  280. 5:37tasting soup.
  281. 5:38If you taste a
  282. 5:39terrible soup, you know
  283. 5:41immediately.
  284. 5:42It is too salty.
  285. 5:43You spit it out.
  286. 5:44That is like reviewing
  287. 5:46human code.
  288. 5:47The mistakes are obvious.
  289. 5:49But AI code is
  290. 5:50like tasting a perfectly
  291. 5:51bland soup.
  292. 5:52It is not too
  293. 5:53salty.
  294. 5:54It is not too
  295. 5:55spicy.
  296. 5:56It looks like soup.
  297. 5:57It smells like soup.
  298. 5:59You taste it fifty
  299. 6:00times.
  300. 6:01Eventually, your taste buds
  301. 6:02just give up.
  302. 6:03You say it is
  303. 6:05Your brain just gets
  304. 6:06tired of looking for
  305. 6:07the flavor.
  306. 6:08Staring at the perfect
  307. 6:09code is dangerous.
  308. 6:11It drains a specific
  309. 6:12part of your brain.
  310. 6:14It is called the
  311. 6:14anterior cingulate cortex.
  312. 6:16This is your brain's
  313. 6:17error detection center.
  314. 6:19Think of it like
  315. 6:20a smoke alarm.
  316. 6:21When it sees a
  317. 6:22typo, the alarm rings.
  318. 6:24But staring at flawless
  319. 6:25AI code drains the
  320. 6:26battery.
  321. 6:27The alarm stops working.
  322. 6:29We call this Turing
  323. 6:31fatigue.
  324. 6:32You get tired of
  325. 6:33looking for hidden flaws.
  326. 6:34You see the green
  327. 6:35checkmarks.
  328. 6:37You assume the AI
  329. 6:38is right.
  330. 6:39So you rubber-stamp terrible
  331. 6:40logic.
  332. 6:41You approve code that
  333. 6:42ruins your project.
  334. 6:45And you do not
  335. 6:45even realize it.
  336. 6:47But bad logic is
  337. 6:48only half the problem.
  338. 6:50Let us talk about
  339. 6:50the biggest pain point
  340. 6:51of 2026.
  341. 6:53It is not the
  342. 6:53AI hallucinating functions.
  343. 6:56real problem is the
  344. 6:57AI hallucinating architectural
  345. 6:59That means it forgets
  346. 7:01the main goal of
  347. 7:01the project.
  348. 7:02It is like having
  349. 7:03a really fast intern
  350. 7:05who has read every
  351. 7:05book in the world
  352. 7:06but has zero common
  353. 7:07sense.
  354. 7:08Imagine you hire a
  355. 7:09builder to add a
  356. 7:10room.
  357. 7:11You want a simple
  358. 7:11wooden room.
  359. 7:13The builder works incredibly
  360. 7:15fast.
  361. 7:16But he uses steel
  362. 7:17instead of wood.
  363. 7:18He adds extra doors
  364. 7:20you did not want.
  365. 7:21The room stands up.
  366. 7:22It passes the basic
  367. 7:23safety test.
  368. 7:25But it completely ruins
  369. 7:26the style of your
  370. 7:27house.
  371. 7:28AI agents do this
  372. 7:29to your codebase.
  373. 7:30They silently introduce micro-dependencies.
  374. 7:34They drift away from
  375. 7:35your core design.
  376. 7:37Let me tell you
  377. 7:37a story about a
  378. 7:38developer.
  379. 7:39We will call her
  380. 7:39Sarah.
  381. 7:41Sarah uses an AI
  382. 7:42agent to build a
  383. 7:43payment system.
  384. 7:44She asks the agent
  385. 7:45to add a discount
  386. 7:47feature.
  387. 7:48The agent writes five
  388. 7:49hundred lines of code.
  389. 7:50It takes ten seconds.
  390. 7:52Sarah runs the local
  391. 7:53tests.
  392. 7:54The tests pass perfectly.
  393. 7:56She clicks merge and
  394. 7:57goes to lunch.
  395. 7:58But the AI did
  396. 7:59something sneaky.
  397. 8:00It did not just
  398. 8:01add a discount.
  399. 8:02It changed how the
  400. 8:03database handles currency.
  401. 8:05It imported a massive
  402. 8:07math library
  403. 8:08Sarah did not need
  404. 8:09this library.
  405. 8:10The old code worked
  406. 8:11fine.
  407. 8:12But the AI thought
  408. 8:13the new library was
  409. 8:14better.
  410. 8:15Three months later, the
  411. 8:16math library gets a
  412. 8:18security update.
  413. 8:19It breaks the whole
  414. 8:20payment system.
  415. 8:21Customers cannot buy anything.
  416. 8:23Sarah tries to fix
  417. 8:24it.
  418. 8:25But she does not
  419. 8:26understand the new math
  420. 8:27library.
  421. 8:28She never built the
  422. 8:29mental model for it.
  423. 8:30She is staring at
  424. 8:31a phantom architecture.
  425. 8:33A phantom architecture is
  426. 8:34a system nobody understands.
  427. 8:36The original design is
  428. 8:37completely gone.
  429. 8:39This is agentic context
  430. 8:40drift in action.
  431. 8:42It creates massive
  432. 8:43Technical debt
  433. 8:44Eventually, the project becomes
  434. 8:46a mess.
  435. 8:47Nobody can fix it.
  436. 8:48So how do elite
  437. 8:49teams actually fix this?
  438. 8:51They use a new
  439. 8:52method.
  440. 8:53It is called the
  441. 8:54adversarial audit.
  442. 8:55You have to change
  443. 8:56your job title.
  444. 8:57You are no longer
  445. 8:58a code creator.
  446. 9:00You are an adversarial
  447. 9:01scientist.
  448. 9:02Adversarial means you play
  449. 9:03against the AI.
  450. 9:04You do not trust
  451. 9:05it.
  452. 9:06You treat every AI
  453. 9:08output as a guess.
  454. 9:09Your job is to
  455. 9:10break that guess.
  456. 9:11Think of the adversarial
  457. 9:12audit like a car
  458. 9:13mechanic.
  459. 9:14A bad mechanic just
  460. 9:15turns the key.
  461. 9:17If the car starts,
  462. 9:18he says it is
  463. 9:18fixed.
  464. 9:19A good mechanic does
  465. 9:20not trust the
  466. 9:22car.
  467. 9:22He revs the engine
  468. 9:23hard.
  469. 9:24He listens for weird
  470. 9:25noises.
  471. 9:26He hits the brakes
  472. 9:27suddenly.
  473. 9:28He actively tries to
  474. 9:29find the hidden flaw.
  475. 9:30You have to be
  476. 9:31the good mechanic.
  477. 9:32Elite teams use a
  478. 9:34three-step routine.
  479. 9:35First, you use a
  480. 9:36concept called evals as
  481. 9:38code.
  482. 9:38Eval is short for
  483. 9:40evaluation.
  484. 9:41You spend most of
  485. 9:42your time writing tests.
  486. 9:43You write the tests
  487. 9:44before AI writes the
  488. 9:45code.
  489. 9:46You define the exact
  490. 9:47rules.
  491. 9:48You build a strict
  492. 9:49cage.
  493. 9:50Then you let the
  494. 9:51AI work inside the
  495. 9:52cage.
  496. 9:53If the AI steps
  497. 9:54out of line, the
  498. 9:55compiler yells.
  499. 9:57This stops the AI
  500. 9:58from drifting.
  501. 9:59Second, you hide the
  502. 10:00AI output.
  503. 10:01Do not look at
  504. 10:02it immediately.
  505. 10:03Force your brain to
  506. 10:04rebuild the logic from
  507. 10:05scratch.
  508. 10:06Write down the steps
  509. 10:07on paper.
  510. 10:08This creates synthetic struggle.
  511. 10:11It forces your brain
  512. 10:12to work.
  513. 10:13It keeps your neural
  514. 10:14pathways strong.
  515. 10:15It prevents that neuro-atrophy
  516. 10:17we talked about.
  517. 10:18Third, you review the
  518. 10:20AI code like an
  519. 10:21attacker.
  520. 10:21Do not look for
  521. 10:22typos.
  522. 10:23Look for logic gaps.
  523. 10:25Ask yourself how you
  524. 10:26can break it.
  525. 10:28Run it with weird
  526. 10:29inputs.
  527. 10:29You want the code
  528. 10:30to fail.
  529. 10:31When you break it,
  530. 10:33you win.
  531. 10:34Let us recap.
  532. 10:35AI gives you incredible
  533. 10:36speed.
  534. 10:37But it takes away
  535. 10:38the cognitive friction you
  536. 10:39need.
  537. 10:40If you just merge
  538. 10:40perfect code, your brain
  539. 10:42shrinks.
  540. 10:43You lose your spatial
  541. 10:44memory.
  542. 10:45You suffer from Turing
  543. 10:46fatigue.
  544. 10:47Your project suffers from
  545. 10:48context drift.
  546. 10:49To survive, you must
  547. 10:51become an adversarial auditor.
  548. 10:53Write tests first.
  549. 10:54Hide the output.
  550. 10:55Break the code on
  551. 10:56purpose.
  552. 10:57That is how you
  553. 10:58stay a senior engineer
  554. 10:59in 2026.
  555. 11:00If you enjoyed this
  556. 11:01breakdown, please hit the
  557. 11:03like button and subscribe.
  558. 11:05It really helps the
  559. 11:06channel.
  560. 11:07I will see you
  561. 11:07in the next one.
  562. 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.