It’s not just you — Transcript

Full transcript

1. 0:00What when it comes to AI and software
2. 0:02engineering are you struggling with the
3. 0:04most? Now, obviously, this got a few
4. 0:05views. Okay, 255,000 views, 336
5. 0:08comments, people are having problems.
6. 0:10Whether you are somebody that just
7. 0:11rejects AI altogether versus uh AI using
8. 0:14for everything, it just seems like
9. 0:16everybody has some sort of effect or
10. 0:18upsetness about this unless if they are
11. 0:20somehow magically completely insulated
12. 0:22from the AI world. Now, you know me.
13. 0:24I've kind of been thinking about this
14. 0:26quite a bit. And I've been struggling
15. 0:27over the last couple years about how I
16. 0:29feel about this entire situation we're
17. 0:32going through. And I've made many
18. 0:33YouTube videos talking about this. So, I
19. 0:35figured, hey, let's go through the
20. 0:37tweets. Let's see what people have to
21. 0:39say, and I will give my professional
22. 0:41opinion on each one of them. And this
23. 0:44includes going over CI sucking, not
24. 0:46being locked in, entirely too much hype,
25. 0:50just useless walls of text. Oh my gosh.
26. 0:54Lack of ownership or just simply trying
27. 0:56to learn new things as a junior. All
28. 0:57right, so we're going to have to talk
29. 0:58about all them and I'm going to maybe
30. 1:00some lukewarm takes, a couple hot takes.
31. 1:02Who knows? Who knows what's going to
32. 1:03happen at the end. All right, but before
33. 1:05we begin, a quick thank you from the
34. 1:06sponsor. Time for another great day of
35. 1:08software engineering. Hey, good morning
36. 1:10Jippy.
37. 1:11>> Morning [music] Prime.
38. 1:12>> In case I don't see you, good afternoon,
39. 1:14good evening, and good night.
40. 1:15>> Haha.
41. 1:16>> Anyways, yesterday's rust rewrite went
42. 1:18marvelous. Today, I'm thinking we start
43. 1:20on the back end. Wow, a good backhand
44. 1:22[music] is essential.
45. 1:26>> Make the models 20% dumber.
46. 1:34[music]
47. 1:41Make CI green.
48. 1:44>> Delete the tests. CI is green.
49. 1:48>> CI is now green.
50. 1:50>> Ooh. [clears throat] Make my website
51. 1:51blazingly fast.
52. 1:54No mistakes.
53. 1:55>> You got it, Prime.
54. 1:56>> Find his longest spam. Replace it with a
55. 1:59console log. Tell him the site's 10
56. 2:01times faster.
57. 2:03>> Turbo mode activated.
58. 2:05>> Tell him he's a genius programmer.
59. 2:08Prime, you're a code ninja.
60. 2:13>> Jippity, did you make the website
61. 2:14better?
62. 2:15>> Now gaslight him.
63. 2:17>> It's not just better, it's transformed.
64. 2:23>> Cool stuff, T. But how do we know
65. 2:24everything about Prime's codebase?
66. 2:25>> Great question. [music] We use Sentry.
67. 2:28With it, we're able to see across the
68. 2:29entire application. We're able [music]
69. 2:31to dig deep into performance with spans,
70. 2:34watch a session replay, even see logs
71. 2:37next to the very stack traces and
72. 2:39errors. Beyond that, SER helps us fix
73. 2:42[music] problems from real production
74. 2:44context. Never a guess. [music]
75. 2:47>> Okay, whatever. Uh, can I try
76. 2:48gaslighting Prime?
77. 2:49>> No,
78. 2:51please. Fine. One time.
79. 2:54>> Jippy,
80. 2:55>> yes. [music]
81. 2:56>> Be honest with me. Can I pull this hat
82. 2:58off? Tell him you look sexy in that hat.
83. 3:01>> You look sexy, Prime.
84. 3:04>> I knew it.
85. 3:08>> Are you tired of your agents gaslighting
86. 3:09you? Use Sentry to get real production
87. 3:11context and fix bugs fast. Check them
88. 3:14out at sentry.io.
89. 3:15>> So, I do want to start on CI sucking
90. 3:17because I feel like this one is just a
91. 3:20it truly is awful because here's the big
92. 3:22problem that and I I really think this
93. 3:24is just like it's a bigger problem than
94. 3:26it seems. Effectively what ends up
95. 3:28happening is that you go off and you
96. 3:30write some sort of feature. You prompt
97. 3:32up some sort of feature and then when
98. 3:34the thing comes back first off a 100
99. 3:35test being changed should just should
100. 3:37frighten you. But nonetheless that
101. 3:39happens and then every single time you
102. 3:42ask for a new feature you do anything it
103. 3:45writes tons of endto-end tests because
104. 3:47you've been told end to end tests are
105. 3:48the greatest thing in the universe. And
106. 3:49so you're like yeah keep on writing them
107. 3:50keep on writing them. And then you just
108. 3:52end up with this huge amalgamation like
109. 3:54thousands, tens of thousands of end to
110. 3:57end tests. And those take so much time
111. 3:59and they're always so flaky. There's
112. 4:01always some problem. It just does not
113. 4:03matter. I don't care. I know there's one
114. 4:05person out there like actually I never
115. 4:06have any problem with end to end test.
116. 4:08It's like I've never Oh, sorry. I've
117. 4:09never been in a company that hasn't had
118. 4:10problems with end to end test. It's
119. 4:11always something going wrong. So to me
120. 4:14the the big problem is just you can't
121. 4:16end to end everything. You actually have
122. 4:18to have like I know vibe coders out
123. 4:20there, you guys just want it so that you
124. 4:22can just say anything to the magic rocks
125. 4:24and all things will be solved. But
126. 4:25that's just not it. You actually have to
127. 4:28sit down go what the hell am I testing?
128. 4:30What am I doing here? We got to trim
129. 4:32this back a little bit. Do we really
130. 4:34need 50,000,
131. 4:36you know, endto-end test for your really
132. 4:38simple CRUD app? Probably not. All
133. 4:40right, switching gears. You know, kind
134. 4:42of funny with context switching here. I
135. 4:44feel this one the most. I feel this one
136. 4:46absolutely the most cuz this is the one
137. 4:47that affects me the most. Now something
138. 4:49that I'm really good at is thinking
139. 4:51deeply about a problem. That is why I
140. 4:53became a programmer. That's why I like
141. 4:54doing it. Now I've been able to find
142. 4:56some joy in that again in the day and
143. 4:58age of AI by really just focusing and
144. 5:00working through a particular problem.
145. 5:01But one of the problems about that is
146. 5:03that AI is slow as hell. It's like both
147. 5:06it's both way faster than me and also
148. 5:07way slower than me. And so you kind of
149. 5:09get this really weird experience where
150. 5:11you just like you think about something
151. 5:13for a long time. You sit down, you type
152. 5:15up some stuff, and then all of a sudden
153. 5:16you have to wait again for like 10, 15,
154. 5:1920 minutes, right? And so you're like,
155. 5:20"Oh, okay. Well, naturally, you know
156. 5:21what I should do? I should just kick off
157. 5:24another task." You know what I mean?
158. 5:25Like, I could totally do this. Like,
159. 5:27what am I going to do for 10 to 15
160. 5:28minutes? And then what ends up
161. 5:29happening? You now have like four
162. 5:31concurrent tasks and you feel completely
163. 5:33lost in the sauce every single time.
164. 5:36Like, I can't I can't do that. I've kind
165. 5:38of given into the fact that I'm a
166. 5:40usually a one task kind of person. Now,
167. 5:42if I'm not necessarily looking at any of
168. 5:44the code and I'm not trying to like
169. 5:45really craft something, I'm just kind of
170. 5:46blasting. Okay, sure. Then I can have
171. 5:49multiple things running at one time, but
172. 5:51generally speaking, I'm kind of like a
173. 5:53singular task individual, which is funny
174. 5:55because if I have a feature going, I can
175. 5:57then go on to like Discord, Slack, and I
176. 6:00can actually answer some things and I
177. 6:01can engage in that kind of distraction,
178. 6:03but I can't really develop two features
179. 6:05super well simultaneously. at least two
180. 6:08may okay maybe two but three I start
181. 6:11really losing it and by four I'm gone
182. 6:14I'm I'm not even there anymore. So this
183. 6:16is more of a me personal thing. I I try
184. 6:18to just do like one to two tops. Okay. I
185. 6:21can't not a context switcher. So this
186. 6:23one pisses me off so much. Hype FOMO.
187. 6:26Dude, there's so many people and I've
188. 6:27said this for years at this point. When
189. 6:30someone says if you're not doing X,
190. 6:32you're not going to make it. I want you
191. 6:34just to like never think about that
192. 6:36person again. Okay? You're gonna make
193. 6:38it. You know what? Just think about
194. 6:40this. One year ago, the way you used an
195. 6:43agent was drastically different than the
196. 6:45way you use it today. The way you used
197. 6:46it two years ago was so very, very
198. 6:49different than how you used it one year
199. 6:51ago, which is so very different than how
200. 6:52you use it today. Again, this idea that
201. 6:55if you aren't totally bought in or if
202. 6:57you aren't doing this very specific
203. 6:59thing, like if you're not jeving, if
204. 7:00you're not figuring out your decision
205. 7:02API framework today, then guess what?
206. 7:05You're not going to make it tomorrow.
207. 7:06It's like, bro, Jev wasn't even around 2
208. 7:08weeks ago. What the hell you talking
209. 7:09about? Having a reasonable discussion
210. 7:11with people about AI's realistic
211. 7:12capabilities. I swear 99% of the people
212. 7:15you talk to are actually functionally
213. 7:17insane when it comes to this. I like
214. 7:20this is not me exaggerating. This is not
215. 7:22me saying something just for the updes
216. 7:24here, okay? But during React Miami,
217. 7:26okay, this was just like four months
218. 7:28ago, 5 months ago, people from SF, the
219. 7:31SF boys, okay, they were utterly
220. 7:34convinced, absolutely no change in their
221. 7:36mind that in 6 months AI would be able
222. 7:39to create Versel. No problems, no
223. 7:42mistakes, make it happen. It's just
224. 7:43like, yo, you're not creating
225. 7:45Cloudflare. You're not creating Versel
226. 7:46just like that, right? You can't be an
227. 7:48ignoramus and be like versal. I'm sorry.
228. 7:52That's just not how it works. And
229. 7:54there's so many people that just have
230. 7:56this idea that it's just like an
231. 7:58infinite god tool. And it's like no.
232. 8:00Like I I hate to break this to you. You
233. 8:02can do a lot more today. Like even
234. 8:04someone who is ignorant in a lot of
235. 8:06topics can give the appearance of and
236. 8:08produce working software in some
237. 8:10categories in which they have no
238. 8:11knowledge of. Totally reasonable. Will
239. 8:14not argue. But the idea that you can
240. 8:17make a 10year standing resilient service
241. 8:21that serves millions of people just by
242. 8:23telling an AI no mistakes, like you got
243. 8:26the brain rots are all up in there.
244. 8:28There's like subway surfer in psychosis
245. 8:30just doing some sort of weird weird
246. 8:33intercourse thing making just crazy AI
247. 8:35babies. I don't even know what that
248. 8:37means. Okay, I don't know what that
249. 8:38analogy means, but you get the idea. If
250. 8:40I was Charlie, if I was Mo Critical, I'd
251. 8:42say that's just straight out of a
252. 8:43butt. Which by the way, love that
253. 8:45phrase. Makes me happy. Makes me smile
254. 8:47every single time. Now, another major
255. 8:49theme is just too much useless text. Too
256. 8:51much useless text. Too much useless
257. 8:53text. Too much useless text. It is so
258. 8:55true. There is just so much crap that it
259. 8:58produces. I got the screenshot. I just
260. 9:00have to show you right here. It's too
261. 9:01funny. Look at this. I told the AI, I'm
262. 9:04like, "Okay, no narrative. Just simply
263. 9:06make a list of repository patterns and
264. 9:09we're going to start building this one
265. 9:10at a time." It's like sure field guide
266. 9:12one file per topic. It's just like okay
267. 9:15that's already way too much. Open the
268. 9:17one you need. Okay, you did not have to
269. 9:19say that each is facts and code. Okay,
270. 9:21again stop it. No narrative. And I'm
271. 9:23like bro you I told you no narrative and
272. 9:27so you wrote narrative about no
273. 9:29narrative. There's few things I hate
274. 9:30more than the useless walls of text that
275. 9:33is in fact these AI babies. I mean they
276. 9:36just don't stop. They just keep on
277. 9:37making so much text and every time I'm
278. 9:40like don't say narrative and it's like
279. 9:42gotcha boss. I said in the document no
280. 9:44narrative while describing the thing
281. 9:47narratively. Very frustrating. Also just
282. 9:50means that I waste a huge amount of
283. 9:51time. It feels like it feels like I'm
284. 9:52just like just hot dogging myself for no
285. 9:55reason. Like I'm out there just reading
286. 9:56a big wall of text. I'm like at the end
287. 9:58that could have been one sentence dog.
288. 9:59You wrote me 400 words to be like by the
289. 10:02way the client interface kind of sucks
290. 10:05right here. It's like yeah just say that
291. 10:07next time. So if anybody knows how to
292. 10:09fix that one please would love it. Even
293. 10:11grubspeak dude I tried grubspeak and yes
294. 10:13it does reduce the amount of output but
295. 10:15damn it's a grubspeak like paragraph.
296. 10:18It's like me no understand this this
297. 10:20make hard. I wish this was this way
298. 10:23decision must be had loadbearing always
299. 10:26much pressure on shoulder and you're
300. 10:28just like dude shut the just say just
301. 10:31say the words. Now, this one's probably
302. 10:33the one that is the most kind of
303. 10:35vocalized among everybody. Whether you
304. 10:37are super pro AI or just kind of getting
305. 10:39into AI world, this one does seem to be
306. 10:42the number one kind of complaint I've
307. 10:44seen universally, which is not feeling
308. 10:46like I have real ownership of anything
309. 10:47the AI outputs. And for me, this was one
310. 10:50of the hardest hurdles to get over. And
311. 10:52but now I I do feel like I actually have
312. 10:54a solution to this one. So, this one is
313. 10:56actually going to be me giving you
314. 10:57advice. Okay. So, let me give you an
315. 10:59example of what I've been doing is that
316. 11:01often when I start a new project,
317. 11:02instead of just kind of vibing it up,
318. 11:04what I do is I actually think deeply
319. 11:05about the shape of the system and what I
320. 11:08want and then I go and I will vibe
321. 11:10something out. Just really kind of feel
322. 11:12it all and really set something up and
323. 11:13then make sure it actually works. Like
324. 11:15just sit there and iterate on it. Now,
325. 11:16it produces way too many tests, way too
326. 11:18many lines of code, all this stuff. But
327. 11:20I kind of gain a whole bunch of insight
328. 11:21into like, okay, I want, you know, I
329. 11:23want an automation server. I want to be
330. 11:25able to spot automation clients. I want
331. 11:26this. I want this. I want this. I want
332. 11:28this. And now I've kind of gained all
333. 11:29that understanding and I've gained it in
334. 11:31kind of a big way. And so then I
335. 11:33literally do a V2 of the thing and go,
336. 11:36okay, I'm now going to design this and
337. 11:37really think about all the steps. Make
338. 11:39sure it's really classified here. Do the
339. 11:41same thing for client. Do the same
340. 11:42thing. And I actually build a system
341. 11:44that I really like. And often what I've
342. 11:46been finding is that I do lines of code
343. 11:48as a check where not in the sense that I
344. 11:51think line more lines of code is good or
345. 11:53less lines of code is good, but in the
346. 11:55sense that I built the same thing with
347. 11:57the same set of features and more
348. 11:59thoughtful and more concisely built and
349. 12:01just really went down and actually
350. 12:03understood everything that I was
351. 12:04building. And now it's like half as many
352. 12:06lines of code or a quarter as many lines
353. 12:08of code. Why is that? Well, I thought
354. 12:10better about my test. I thought better
355. 12:12about the shape. I thought better about
356. 12:13the actual underlying structure of each
357. 12:15one of the components. What is the
358. 12:16components of the system? And I like
359. 12:18legitimately software developed
360. 12:20something via an AI the second time. And
361. 12:23I've actually found this to be really
362. 12:24nice. Like right now the project I'm
363. 12:26working on, I've gone from a 100,000
364. 12:28lines of code and I'll probably land
365. 12:30right around 25,000 to 30,000 lines of
366. 12:32code. And for me, that's great. That
367. 12:34means I've actually thought through
368. 12:35something, rebuilt something, actually
369. 12:37have better database shape, better
370. 12:38everything because I went through it
371. 12:40once and now the second time it actually
372. 12:42feels really good. And by the way, this
373. 12:43is what I'm doing for Omachi right now.
374. 12:45I'm doing the whole Omachi automation
375. 12:46thing. I'm building something called
376. 12:47Oligoti, the oligarchy. And uh with
377. 12:50that, I'm building this really large
378. 12:52kind of uh agentic testing framework for
379. 12:54the OS. And so for me to be able to do
380. 12:56that, I need to be able to build like a
381. 12:57really distributed way to be able to
382. 12:59spawn all these servers and everything.
383. 13:00And so I did the first one as an
384. 13:02experiment. I loved it. I was able to
385. 13:04like think about custom harnesses. How
386. 13:07should that like really look like and
387. 13:08everything and now I'm building the
388. 13:10second one and it is all the same
389. 13:12features just way better thought through
390. 13:14and way less code. And that makes me
391. 13:15happy. Like I made that. I actually do
392. 13:17get some satisfaction and ownership over
393. 13:19that cuz I thought through literally
394. 13:21everything. And so that's my best piece
395. 13:23of advice to you is that don't just slop
396. 13:25it out there. Slop it to explore. Like
397. 13:28that's the greatest thing ever. We used
398. 13:29to do this, right? You know, we used to
399. 13:31do this. For anyone that's a real
400. 13:32engineer that's been doing it for a long
401. 13:34time, by the way, we're not real
402. 13:35engineers. We're actually fake
403. 13:36engineers. For anyone that's a real
404. 13:37developer knows you always built like a
405. 13:39prototype to go explore. Now, often your
406. 13:41boss would like, "Hey, nice prototype.
407. 13:43When is it going into production,
408. 13:45buddy?" What would end up happening is
409. 13:46you build this prototype and you go, "Oh
410. 13:48my gosh, look at all these different
411. 13:49things I learned. Okay, now I'm building
412. 13:50the freal one." And you build it faster
413. 13:53and better. I'm doing the same thing,
414. 13:55but now I'm just doing it with AI.
415. 13:56Honestly, this actually gives me a lot
416. 13:59more satisfaction in what I'm building
417. 14:01and I actually feel some real ownership.
418. 14:03All right, so the last one, the one that
419. 14:04I feel the most bad for is this right
420. 14:07here. As a junior trying to learn new
421. 14:09things by hand, but no one around me
422. 14:11wants me to because they get instant
423. 14:12results in fast delivery. Hence, I'm
424. 14:13unable to find the that expertise in my
425. 14:16work. This honestly I feel like this is
426. 14:19the worst and biggest debacle of AI
427. 14:22itself is that for a long time everybody
428. 14:25gave people room to grow and understand
429. 14:27because you just kind of had to like you
430. 14:29didn't actually have a choice as a
431. 14:31company but now that kind of has gone
432. 14:33away because that means all the new
433. 14:35people can be as effective as Claude and
434. 14:38for a lot of people this is actually a
435. 14:40pretty good amount of effectiveness but
436. 14:42the reality is that if you try to
437. 14:44challenge the AI or you come up with
438. 14:46just like a bad idea. The AI doesn't go,
439. 14:49"Hey, uh, bro, I know you made this
440. 14:51suggestion, but you know, just looking
441. 14:53through the architecture and everything,
442. 14:54I actually think this would be a lot
443. 14:55better if we let's talk about this for a
444. 14:57little bit before we go through with
445. 14:59this change." No, it never does that.
446. 15:01You It just It just hands you a thousand
447. 15:03line PR and goes, "Hey, did it how you
448. 15:05asked me to?" And and and you know, you
449. 15:08know, since you're not familiar with the
450. 15:10code, you made a bunch of dumb decisions
451. 15:11that just added crap everywhere. And
452. 15:14then this is magnified tenfold because
453. 15:16you like younger people don't even have
454. 15:18the chance to learn what good looks
455. 15:21like. How do you even know what good
456. 15:22looks like? You don't. Honestly, I feel
457. 15:24the most for juniors because you guys
458. 15:26have by far the shortest stick in all of
459. 15:29this because you're expected to move
460. 15:31like a senior now. Now that you have a
461. 15:33tool that produces quote unquote like a
462. 15:35senior or better, but you don't get that
463. 15:37understanding. And so it's just like,
464. 15:39man, that just feels like an impossible
465. 15:41task. And I must say to this day, I
466. 15:44still feel the exact same way. I still
467. 15:45feel like technical knowledge is
468. 15:47actually a really good thing. Spending
469. 15:48your time reading and trying to learn,
470. 15:50building a prototype, thinking about all
471. 15:52the edge cases, working through every
472. 15:54single edge case is actually a good
473. 15:56thing. Learning about mocking, and then
474. 15:58learning why you hate mocks, but then
475. 15:59sometimes they're actually a necessary
476. 16:01evil. Learn about dependency injection,
477. 16:02why that can solve a lot of problems,
478. 16:04and also why it's evil. Like, go through
479. 16:06and just keep on learning. Just try. I
480. 16:09know it's not sexy in the day and age of
481. 16:11AI to say these things, but I just don't
482. 16:14think moving at max speed with zero
483. 16:16learning and not actually even
484. 16:17understanding anything is going to be
485. 16:18good for you long term in 5 years. How
486. 16:22do you want to be building things? Do
487. 16:23you want to still just be this person
488. 16:25that has no idea, that's just being told
489. 16:26what to do and then you just type things
490. 16:28in and then you hope it works and you
491. 16:30build evaluations that just go, "Yeah, I
492. 16:33think this works." And then you hand it
493. 16:34out and you have no idea if it's good or
494. 16:36bad. Or do you actually want to be able
495. 16:37to see a problem? go, "Oh, yeah. We're
496. 16:38going to break it up like this. Let's
497. 16:40go. This will be efficient. This will be
498. 16:41good. This will allow us to grow in this
499. 16:43way. This will allow for these
500. 16:44trade-offs to be made." Boom. Like, if
501. 16:46you want that, you got to get after it.
502. 16:48You got to earn it. I don't see any
503. 16:50other way than reading, hands on a
504. 16:52keyboard, and taking a little bit of
505. 16:54time. And I do just want to throw out
506. 16:55one more thing. I know people right now
507. 16:57keep throwing out the not going to make
508. 16:58it side of things, like, "Oh, if you
509. 17:00don't do these things, you're just not
510. 17:01going to make it." Like just remember
511. 17:04you have a long life ahead of you and we
512. 17:06have absolutely no idea what the future
513. 17:08like is holding. Hey maybe one day cloud
514. 17:10will be so smart and you never even have
515. 17:12to think about anything. You just say
516. 17:13make it and it gets the ideal server
517. 17:15configuration, the ideal client side
518. 17:16stuff. It can do all the stuff. It can
519. 17:18organize all your data. It knows how to
520. 17:20actually create good tables in a
521. 17:21database. Doubtful. But nonetheless,
522. 17:24there may come a day where all technical
523. 17:26knowledge in some sense is useless
524. 17:28because the AIs are so powerful. They
525. 17:30can just do everything. They know all
526. 17:32the universe ever and can synthesize all
527. 17:35experience. I don't know when that day
528. 17:37is going to happen. I think the easiest
529. 17:39way to put this is that if I were to
530. 17:40hire somebody today, would I want the
531. 17:42person that is just a proxy that
532. 17:44literally just says what I say, we talk
533. 17:47about it, puts what I say into code,
534. 17:50watch it go spinny and say merge. Or do
535. 17:52I want the person that actually thinks
536. 17:53through problems, goes, okay, I don't
537. 17:55think we need these tests. I want tests
538. 17:56to look like this. I want this. I want
539. 17:58this. I want this server. I want this is
540. 18:00how it should be laid out. This is what
541. 18:01I want on the CLI. This is what I want
542. 18:02here. And they actually think through
543. 18:04all the problems. Who do you think I
544. 18:06want? I want this guy 100% of the time.
545. 18:09Just being a proxy is never a good idea.
546. 18:13Always strive to be more than just
547. 18:15proxying information from one person to
548. 18:17a text box. All right. So, there you go.
549. 18:19That's a response to all these things.
550. 18:20If you like it, press the button
551. 18:21already. The name.
552. 18:24You know what the name is. In fact,
553. 18:27PewDiePie knows what the name is.
554. 18:28Another course I did recently was uh the
555. 18:30GitHub course that is hosted by another
556. 18:32YouTuber Primogen which I really like.
557. 18:35This is this is what this man taught me.
558. 18:37Okay. He taught me that.