WEBVTT

1
00:00:00.160 --> 00:00:03.520
<v Speaker 1>Welcome to the deep dive. We take interesting sources, find

2
00:00:03.560 --> 00:00:06.280
<v Speaker 1>the core insights, and well help you get up to

3
00:00:06.280 --> 00:00:11.199
<v Speaker 1>speed fast Exactly. Today we're diving into something huge, something

4
00:00:11.240 --> 00:00:13.160
<v Speaker 1>that's really changing the game for software development.

5
00:00:13.199 --> 00:00:15.160
<v Speaker 2>It really is. The whole landscape is shifting.

6
00:00:15.320 --> 00:00:17.359
<v Speaker 1>It's not just about writing code anymore, is it. It's

7
00:00:17.399 --> 00:00:22.000
<v Speaker 1>about you know, working alongside AI, AI assisted coding. What

8
00:00:22.039 --> 00:00:23.640
<v Speaker 1>that actually means for programmers?

9
00:00:23.839 --> 00:00:26.839
<v Speaker 2>Yeah, the role is definitely evolving. And for this deep dive,

10
00:00:26.879 --> 00:00:30.399
<v Speaker 2>we're leaning heavily on Addie Asmani's book Vibe Coding The

11
00:00:30.440 --> 00:00:34.920
<v Speaker 2>Future of Programming, specifically the chapters on the seventy percent

12
00:00:35.000 --> 00:00:38.479
<v Speaker 2>problem and beyond the seventy percent really insightful stuff.

13
00:00:38.600 --> 00:00:42.880
<v Speaker 1>Okay, great, So our mission today, let's unpack the reality

14
00:00:42.920 --> 00:00:45.399
<v Speaker 1>of AI and coding. We want you to understand where

15
00:00:45.399 --> 00:00:49.640
<v Speaker 1>it's genuinely good, where it will falls down, and crucially,

16
00:00:49.960 --> 00:00:52.679
<v Speaker 1>how you, as a human developer can use your skills

17
00:00:52.679 --> 00:00:54.039
<v Speaker 1>to thrive with these tools.

18
00:00:54.159 --> 00:00:57.640
<v Speaker 2>Right, it's about finding that synergy maximizing your human contribution

19
00:00:57.799 --> 00:00:59.960
<v Speaker 2>in this sort of new AI powered world.

20
00:01:00.359 --> 00:01:02.920
<v Speaker 1>So let's kick things off with what Osmani calls the

21
00:01:03.000 --> 00:01:07.280
<v Speaker 1>seventy percent problem. Because these AI tools I mean they

22
00:01:07.319 --> 00:01:09.439
<v Speaker 1>are astonishingly good at certain things.

23
00:01:09.480 --> 00:01:13.239
<v Speaker 2>Oh, absolutely, generating boilerplate stand up functions. Yeah, they can

24
00:01:13.280 --> 00:01:14.840
<v Speaker 2>turn that stuff out incredibly fast.

25
00:01:14.920 --> 00:01:18.480
<v Speaker 1>Yeah, getting projects like roughly seventy percent of the way there.

26
00:01:18.760 --> 00:01:20.840
<v Speaker 1>It feels almost like magic sometimes it does.

27
00:01:20.879 --> 00:01:23.480
<v Speaker 2>But you know, it's fascinating how you hit this this

28
00:01:23.640 --> 00:01:27.040
<v Speaker 2>wall after that initial burst. Peter Yang tweeted about this perfectly.

29
00:01:27.159 --> 00:01:28.120
<v Speaker 1>Oh yeah, what did you say?

30
00:01:28.280 --> 00:01:31.159
<v Speaker 2>He said that first seventy percent is super fast, but

31
00:01:31.200 --> 00:01:35.280
<v Speaker 2>that last thirty percent is frustrating, like one step forward

32
00:01:35.319 --> 00:01:38.159
<v Speaker 2>and two steps backward, new bugs popping up constantly.

33
00:01:38.400 --> 00:01:41.599
<v Speaker 1>Hmm. That rings true. It brings up this core idea,

34
00:01:41.640 --> 00:01:45.239
<v Speaker 1>doesn't it. This complexity divide. Fred Brooks talked about this.

35
00:01:45.239 --> 00:01:49.719
<v Speaker 2>Ages ago, right, accidental complexity versus essential complexity.

36
00:01:49.159 --> 00:01:53.120
<v Speaker 1>Exactly an AI. It crushes the accidental stuff, the repetitive, mechanical,

37
00:01:53.519 --> 00:01:54.439
<v Speaker 1>oftent tedious work.

38
00:01:54.480 --> 00:01:55.599
<v Speaker 2>Yeah, eats that for breakfast.

39
00:01:55.719 --> 00:01:59.120
<v Speaker 1>But the essential complexity, you're truly understanding the core problem

40
00:01:59.120 --> 00:02:02.640
<v Speaker 1>managing its inherent difficulty. That's still firmly on us, on

41
00:02:02.680 --> 00:02:03.560
<v Speaker 1>the human developers.

42
00:02:03.840 --> 00:02:07.680
<v Speaker 2>Absolutely, and that leads straight into what Osmoni calls the

43
00:02:07.760 --> 00:02:08.759
<v Speaker 2>last mile gap.

44
00:02:08.840 --> 00:02:11.599
<v Speaker 1>The last mile explain that well.

45
00:02:11.120 --> 00:02:14.759
<v Speaker 2>The AI gives you a first draft basically a plausible solution,

46
00:02:15.000 --> 00:02:19.439
<v Speaker 2>but that final thirty percent covering edge cases, refining the

47
00:02:19.520 --> 00:02:23.560
<v Speaker 2>architecture so it actually scales, making sure it's maintainable long term.

48
00:02:23.680 --> 00:02:27.000
<v Speaker 2>That needs a human, that needs serious human expertise. And

49
00:02:27.159 --> 00:02:29.759
<v Speaker 2>AI might spit out a function that works fine for

50
00:02:29.960 --> 00:02:33.439
<v Speaker 2>the main scenario, you know, but it won't automatically think

51
00:02:33.439 --> 00:02:37.319
<v Speaker 2>about weird inputs or race conditions, performance issues, or what

52
00:02:37.360 --> 00:02:38.120
<v Speaker 2>the system might.

53
00:02:38.000 --> 00:02:41.159
<v Speaker 1>Need next year unless you explicitly tell it to or

54
00:02:41.319 --> 00:02:43.639
<v Speaker 1>like guide it really carefully with your prompt precisely.

55
00:02:43.840 --> 00:02:45.520
<v Speaker 2>You have to teach it in a way.

56
00:02:45.599 --> 00:02:48.360
<v Speaker 1>And this is where it gets well, maybe big concerning

57
00:02:48.960 --> 00:02:52.039
<v Speaker 1>the AI's confidence often doesn't match its actual reliability.

58
00:02:52.120 --> 00:02:55.120
<v Speaker 2>Oh definitely. Stevie Edge had this great analogy. He compared

59
00:02:55.240 --> 00:03:00.000
<v Speaker 2>LMS to wildly productive junior developers, super fast, super eager potential,

60
00:03:00.000 --> 00:03:03.960
<v Speaker 2>actually whacked out on mind altering drugs, prone to coming

61
00:03:04.039 --> 00:03:06.319
<v Speaker 2>up with crazy or unworkable approaches.

62
00:03:06.639 --> 00:03:09.759
<v Speaker 1>Huh Okay. That paints a picture. So what's the real

63
00:03:09.840 --> 00:03:10.960
<v Speaker 1>danger there for a developer?

64
00:03:11.560 --> 00:03:14.479
<v Speaker 2>The danger is the AI doesn't really get the problem.

65
00:03:14.840 --> 00:03:18.080
<v Speaker 2>It's just matching patterns from its training data. It's stitching

66
00:03:18.120 --> 00:03:21.919
<v Speaker 2>things together. Right, Only a human, someone who deeply understands

67
00:03:22.000 --> 00:03:25.400
<v Speaker 2>the domain the system, can spot if a solution that

68
00:03:25.439 --> 00:03:29.159
<v Speaker 2>looks okay on the surface is actually hiding like long

69
00:03:29.280 --> 00:03:30.319
<v Speaker 2>term land mines.

70
00:03:31.280 --> 00:03:34.560
<v Speaker 1>Land mines, I like that. AI doesn't invent new abstractions,

71
00:03:34.560 --> 00:03:34.960
<v Speaker 1>does it.

72
00:03:34.960 --> 00:03:38.280
<v Speaker 2>It doesn't take responsibility, Nope, it remixes what it seems.

73
00:03:38.319 --> 00:03:41.199
<v Speaker 2>So if it confidently gives you bad code and you

74
00:03:41.240 --> 00:03:43.879
<v Speaker 2>don't catch it, while, that could be disastrous.

75
00:03:43.960 --> 00:03:47.120
<v Speaker 1>So okay, big picture, then what's the takeaway for US developers?

76
00:03:47.240 --> 00:03:51.120
<v Speaker 1>AI is clearly a force multiplier, a turbo.

77
00:03:50.719 --> 00:03:54.520
<v Speaker 2>Boost definitely handles that maybe seventy percent of the grantwork, but.

78
00:03:54.479 --> 00:03:57.439
<v Speaker 1>It's absolutely not a silver bullet. It's not replacing human judgment,

79
00:03:57.560 --> 00:03:58.000
<v Speaker 1>not at all.

80
00:03:58.039 --> 00:04:02.840
<v Speaker 2>That remaining harder thirty percent that still needs trained, thoughtful humans.

81
00:04:02.919 --> 00:04:04.680
<v Speaker 2>That's where the real value add is now.

82
00:04:04.719 --> 00:04:07.280
<v Speaker 1>Okay, good, So moving on. Asani points out, these two

83
00:04:07.319 --> 00:04:10.080
<v Speaker 1>distinct ways teams are actually using AI right now, the

84
00:04:10.080 --> 00:04:11.719
<v Speaker 1>bootstrappers and the iterators.

85
00:04:11.960 --> 00:04:15.199
<v Speaker 2>Yeah, the bootstrappers, they're all about getting a new project

86
00:04:15.240 --> 00:04:18.079
<v Speaker 2>off the ground zero to MVP right.

87
00:04:18.040 --> 00:04:21.199
<v Speaker 1>Using tools like bolt or V zero or those screenshot

88
00:04:21.240 --> 00:04:21.800
<v Speaker 1>to code.

89
00:04:21.600 --> 00:04:24.759
<v Speaker 2>Things exactly imagine like a solo dev taking a Figma

90
00:04:24.759 --> 00:04:27.319
<v Speaker 2>design and turning it into a working web app really

91
00:04:27.319 --> 00:04:30.279
<v Speaker 2>really fast, in next to no time, as oz money

92
00:04:30.279 --> 00:04:32.519
<v Speaker 2>puts it. That's bootstrapping with AI.

93
00:04:33.000 --> 00:04:36.240
<v Speaker 1>Wow, that's powerful for just getting ideas.

94
00:04:35.839 --> 00:04:38.480
<v Speaker 2>Out there quickly, totally great for validation. Then you have

95
00:04:38.519 --> 00:04:42.000
<v Speaker 2>the iterators. These are the folks weaving AI into their daily.

96
00:04:41.759 --> 00:04:46.079
<v Speaker 1>Coding, like using Copilot or cursor, kline, windsurf. Yeah, those

97
00:04:46.160 --> 00:04:47.240
<v Speaker 1>kinds of tools exactly.

98
00:04:47.240 --> 00:04:52.240
<v Speaker 2>They're using AI for com completion, refactoring, generating tests, writing docs,

99
00:04:52.800 --> 00:04:54.279
<v Speaker 2>almost like a pair programmer.

100
00:04:54.439 --> 00:04:57.319
<v Speaker 1>Okay, so two different modes. Yeah, but there's always a

101
00:04:57.360 --> 00:05:00.959
<v Speaker 1>button there. They're hidden costs here, failure patterns pop it out.

102
00:05:00.959 --> 00:05:03.480
<v Speaker 2>Yeah, definitely. One big one is this House of Cards code.

103
00:05:03.600 --> 00:05:05.639
<v Speaker 1>Puss of Cards sounds fragile, it is.

104
00:05:06.199 --> 00:05:10.879
<v Speaker 2>This happens when say, junior engineers just accept the AI's

105
00:05:11.000 --> 00:05:13.680
<v Speaker 2>output too easily because it looks right. The code might

106
00:05:13.720 --> 00:05:18.240
<v Speaker 2>seem complete, maybe even past basic tests, but under real

107
00:05:18.279 --> 00:05:22.920
<v Speaker 2>world stress, it just collapses. It lacks that deeper architectural thought.

108
00:05:23.240 --> 00:05:26.560
<v Speaker 2>They're robust error handling. A senior dev would build in HI.

109
00:05:26.720 --> 00:05:30.399
<v Speaker 2>We stop that vigilance, good mentorship, because otherwise you fall

110
00:05:30.399 --> 00:05:33.480
<v Speaker 2>into that two steps back anti pattern, Right, the one.

111
00:05:33.439 --> 00:05:37.079
<v Speaker 1>Yang mentioned fix one bug, the AI suggest to fix,

112
00:05:37.199 --> 00:05:39.040
<v Speaker 1>and bam, that breaks two other things.

113
00:05:39.279 --> 00:05:43.480
<v Speaker 2>Exactly. It's this frustrating loop of diminishing returns, and this

114
00:05:43.560 --> 00:05:46.680
<v Speaker 2>points to what Osmani calls a knowledge paradox.

115
00:05:46.920 --> 00:05:47.720
<v Speaker 1>Knowledge paradox.

116
00:05:47.800 --> 00:05:50.000
<v Speaker 2>Yeah, seniors use AI to go faster on things they

117
00:05:50.000 --> 00:05:53.519
<v Speaker 2>already understand how to do, But juniors they sometimes try

118
00:05:53.560 --> 00:05:55.279
<v Speaker 2>to use AI to figure out what to do in

119
00:05:55.319 --> 00:05:56.000
<v Speaker 2>the first place.

120
00:05:56.120 --> 00:05:59.000
<v Speaker 1>Ah, so are they actually hindering their own learning? Is

121
00:05:59.040 --> 00:06:00.360
<v Speaker 1>AI creating a skill gap?

122
00:06:00.560 --> 00:06:03.319
<v Speaker 2>That's a serious question and a real concern. It potentially

123
00:06:03.360 --> 00:06:04.639
<v Speaker 2>could if not managed well.

124
00:06:04.720 --> 00:06:07.319
<v Speaker 1>And it gets even wilder with agentic AI.

125
00:06:07.480 --> 00:06:12.000
<v Speaker 2>Oh yeah, the autonomous agents like Devin AI clawd code.

126
00:06:12.079 --> 00:06:15.720
<v Speaker 1>These things don't just suggest code, They plan and execute

127
00:06:15.720 --> 00:06:16.839
<v Speaker 1>whole tasks.

128
00:06:16.720 --> 00:06:21.560
<v Speaker 2>Pretty much minimal human input sometimes, which sounds amazing. But

129
00:06:21.759 --> 00:06:24.639
<v Speaker 2>the risks they get amplified, like.

130
00:06:24.600 --> 00:06:26.800
<v Speaker 1>What just relying on it too much?

131
00:06:27.079 --> 00:06:29.480
<v Speaker 2>That plus not being able to step in easily if

132
00:06:29.480 --> 00:06:32.560
<v Speaker 2>it goes off the rails, imagine cascading errors if you

133
00:06:32.680 --> 00:06:34.759
<v Speaker 2>haven't audited its process properly.

134
00:06:34.959 --> 00:06:37.240
<v Speaker 1>How do you even audit something like that? Effectively?

135
00:06:37.319 --> 00:06:39.800
<v Speaker 2>It means you can't just check the final code. You

136
00:06:39.839 --> 00:06:43.399
<v Speaker 2>need to understand its reasoning, trace its steps. Your own

137
00:06:43.519 --> 00:06:47.720
<v Speaker 2>foundational knowledge becomes even more critical to spot the flaws

138
00:06:47.720 --> 00:06:48.600
<v Speaker 2>it might introduce.

139
00:06:48.800 --> 00:06:50.240
<v Speaker 1>It really hammers home the need.

140
00:06:50.079 --> 00:06:51.639
<v Speaker 2>For fundamentals absolutely.

141
00:06:51.720 --> 00:06:53.839
<v Speaker 1>And then there's the demo quality trap. Ah.

142
00:06:53.920 --> 00:06:57.480
<v Speaker 2>Yes, AI builds a slick demo happy path works perfectly.

143
00:06:57.600 --> 00:06:59.240
<v Speaker 2>Oh wow's everyone.

144
00:06:58.959 --> 00:07:00.639
<v Speaker 1>Looks great on Twitter exactly?

145
00:07:00.839 --> 00:07:03.160
<v Speaker 2>Yeah, but then real users touch it and all the

146
00:07:03.199 --> 00:07:07.199
<v Speaker 2>missing polush shows up, weird error messages, crashes on edge cases,

147
00:07:07.240 --> 00:07:09.720
<v Speaker 2>accessibility fails, performance nightmares.

148
00:07:09.399 --> 00:07:12.160
<v Speaker 1>Stuff AI didn't think about because it wasn't explicitly told

149
00:07:12.199 --> 00:07:12.639
<v Speaker 1>to right.

150
00:07:12.839 --> 00:07:17.839
<v Speaker 2>Building truly self serve robust software that still takes human empathy,

151
00:07:18.120 --> 00:07:22.439
<v Speaker 2>real world experience, and that deep sense of craftsmanship. AI

152
00:07:22.600 --> 00:07:23.519
<v Speaker 2>isn't there yet.

153
00:07:23.399 --> 00:07:24.879
<v Speaker 1>Not even close. It sounds like no.

154
00:07:25.360 --> 00:07:28.680
<v Speaker 2>Okay, So given all this complexity, these traps, how do

155
00:07:28.759 --> 00:07:32.680
<v Speaker 2>we actually use AI well day to day? Osmani says,

156
00:07:32.879 --> 00:07:35.720
<v Speaker 2>Vibe coding isn't some weird separate thing, right, It can

157
00:07:35.720 --> 00:07:37.839
<v Speaker 2>fit into agile DevOps totally.

158
00:07:37.920 --> 00:07:41.959
<v Speaker 1>It's about integration. One key pattern is AI as the

159
00:07:42.040 --> 00:07:42.720
<v Speaker 1>first drafter.

160
00:07:43.040 --> 00:07:45.199
<v Speaker 2>So the AI writes the initial version.

161
00:07:45.360 --> 00:07:48.759
<v Speaker 1>Yeah, AI generates the code. Then the humans step in,

162
00:07:49.240 --> 00:07:52.199
<v Speaker 1>refine it, refactor it, test it rigorously.

163
00:07:52.560 --> 00:07:54.600
<v Speaker 2>Makes sense. What are the key practices there?

164
00:07:54.879 --> 00:07:58.720
<v Speaker 1>Well, team coordination is huge, like talk and stand ups. Hey,

165
00:07:58.759 --> 00:08:01.920
<v Speaker 1>I'm gonna use AI for this part. Agree on coding standards.

166
00:08:01.959 --> 00:08:04.240
<v Speaker 1>Maybe even feed those standards to the AI.

167
00:08:04.439 --> 00:08:06.959
<v Speaker 2>Oh interesting, Give it context exactly.

168
00:08:06.759 --> 00:08:09.439
<v Speaker 1>And share good prompts. A well crafted prompt can save

169
00:08:09.519 --> 00:08:12.759
<v Speaker 1>someone else a ton of time and version control. GET

170
00:08:13.000 --> 00:08:16.360
<v Speaker 1>must be vital, more vital than ever. Commit often try

171
00:08:16.360 --> 00:08:20.040
<v Speaker 1>to isolate the AI generated changes. Some teams tag commits

172
00:08:20.160 --> 00:08:21.000
<v Speaker 1>like AI.

173
00:08:20.879 --> 00:08:23.399
<v Speaker 2>Not for blame, but for awareness.

174
00:08:22.680 --> 00:08:25.839
<v Speaker 1>Precisely so reviewers know, okay, give this bit an extra

175
00:08:25.879 --> 00:08:28.480
<v Speaker 1>look for hidden assumptions or edge cases. It's like a

176
00:08:28.519 --> 00:08:30.079
<v Speaker 1>little flag saying scrutinize me.

177
00:08:30.199 --> 00:08:33.279
<v Speaker 2>Got it? What's another pattern AI as a pair programmer.

178
00:08:33.320 --> 00:08:34.600
<v Speaker 2>This is more of a hybrid.

179
00:08:34.279 --> 00:08:35.960
<v Speaker 1>Approach, blending human and machine.

180
00:08:36.080 --> 00:08:40.120
<v Speaker 2>Yeah, human intuition, machine speed. AI does the boilerplate, generate

181
00:08:40.200 --> 00:08:43.279
<v Speaker 2>some tests, maybe the human does the critical review, ensures

182
00:08:43.320 --> 00:08:43.879
<v Speaker 2>the quality.

183
00:08:44.000 --> 00:08:46.759
<v Speaker 1>Ask the right questions, so like I might ask the

184
00:08:46.799 --> 00:08:49.960
<v Speaker 1>AI to draft code for integrating a new library.

185
00:08:49.679 --> 00:08:52.159
<v Speaker 2>Right, and then you take that draft and carefully check

186
00:08:52.200 --> 00:08:55.279
<v Speaker 2>it against the official docs. You don't just trust it blindly.

187
00:08:55.799 --> 00:08:57.799
<v Speaker 1>How's that different from pairing with another human?

188
00:08:57.919 --> 00:09:01.360
<v Speaker 2>It's a different vibe. Human AI is great for rapid generation,

189
00:09:01.720 --> 00:09:07.240
<v Speaker 2>getting scaffolding up fast. Human pairing still king for really

190
00:09:07.320 --> 00:09:11.679
<v Speaker 2>deep problem solving, architectural debates, and just learning from each other.

191
00:09:11.879 --> 00:09:16.320
<v Speaker 1>Okay, best practices for this AI pairing keep AI sessions

192
00:09:16.360 --> 00:09:19.360
<v Speaker 1>focused on distinct tasks so the context doesn't get muddy.

193
00:09:19.840 --> 00:09:24.679
<v Speaker 2>Write clear, focused prompts, review the output constantly, commit frequently,

194
00:09:25.000 --> 00:09:27.399
<v Speaker 2>keep that feedback loop tight. Iterating with the.

195
00:09:27.360 --> 00:09:28.559
<v Speaker 1>AI makes sense.

196
00:09:28.799 --> 00:09:32.440
<v Speaker 2>And the last pattern, AI is a validator, using tools

197
00:09:32.559 --> 00:09:35.240
<v Speaker 2>like deep code or snick for bugs and security issues,

198
00:09:35.399 --> 00:09:38.200
<v Speaker 2>or things like test GTT or Quoto to help generate

199
00:09:38.240 --> 00:09:38.919
<v Speaker 2>test cases.

200
00:09:39.240 --> 00:09:41.000
<v Speaker 1>So AI does the first pass.

201
00:09:40.759 --> 00:09:44.759
<v Speaker 2>Analysis, Yeah, spots the obvious stuff. Then human review focuses

202
00:09:44.759 --> 00:09:48.440
<v Speaker 2>on the tricky bits, the complex logic, the user experience,

203
00:09:48.600 --> 00:09:49.919
<v Speaker 2>any ethical concerns.

204
00:09:50.320 --> 00:09:53.919
<v Speaker 1>It's a collaboration, like a tireless intern, finding common mistakes,

205
00:09:54.200 --> 00:09:55.919
<v Speaker 1>freeing you up for harder problems.

206
00:09:56.000 --> 00:09:56.519
<v Speaker 2>Exactly that.

207
00:09:56.879 --> 00:10:00.559
<v Speaker 1>Okay, This leads nicely into what osminding calls the golden

208
00:10:00.639 --> 00:10:05.200
<v Speaker 1>rules of vibe coding. With all these possibilities and pitfalls,

209
00:10:05.720 --> 00:10:09.639
<v Speaker 1>what are the absolute must dos? The core philosophy.

210
00:10:09.799 --> 00:10:14.080
<v Speaker 2>Rule one basically, be specific, clear prompts, Tell the AI

211
00:10:14.279 --> 00:10:17.399
<v Speaker 2>exactly what you want, like giving instructions to that brilliant

212
00:10:17.440 --> 00:10:21.000
<v Speaker 2>but very literal intern right. And then always always validate

213
00:10:21.000 --> 00:10:23.519
<v Speaker 2>the output. Check it against your intent. Don't just assume

214
00:10:23.519 --> 00:10:25.159
<v Speaker 2>it's right because it looks convincing.

215
00:10:25.320 --> 00:10:29.840
<v Speaker 1>Treat AI like a junior dev. Need supervision, constant supervision, and.

216
00:10:29.840 --> 00:10:31.559
<v Speaker 2>Use it to expand what you can do, not to

217
00:10:31.600 --> 00:10:34.080
<v Speaker 2>switch off your brain. Don't let it replace your critical thinking.

218
00:10:34.120 --> 00:10:35.039
<v Speaker 2>That's absolutely key.

219
00:10:35.120 --> 00:10:37.919
<v Speaker 1>Coordinate with your team before you start generating swaths of

220
00:10:37.960 --> 00:10:38.679
<v Speaker 1>code with AI.

221
00:10:38.919 --> 00:10:42.600
<v Speaker 2>Yes, talk about it. Make AI use normal, transparent, not

222
00:10:42.720 --> 00:10:43.720
<v Speaker 2>some hidden secret.

223
00:10:44.039 --> 00:10:47.360
<v Speaker 1>Isolate AI changes and get separate commits.

224
00:10:47.919 --> 00:10:51.519
<v Speaker 2>Makes life easier definitely, and ensure all code gets reviewed.

225
00:10:51.600 --> 00:10:55.279
<v Speaker 2>Human written AI written doesn't matter. Code review is non negotiable.

226
00:10:55.399 --> 00:10:58.879
<v Speaker 1>And the big one. Don't merge code you don't understand, period,

227
00:10:59.320 --> 00:11:01.039
<v Speaker 1>no matter how cool the AI made it look.

228
00:11:01.039 --> 00:11:06.519
<v Speaker 2>Couldn't agree more. And finally, prioritize documentation comments ADRs, especially

229
00:11:06.519 --> 00:11:10.600
<v Speaker 2>for complex AI generated bits. Share good prompts, reflect on

230
00:11:10.639 --> 00:11:12.399
<v Speaker 2>what's working what's not. Iterate.

231
00:11:13.519 --> 00:11:16.440
<v Speaker 1>These aren't just about getting code out faster, are they?

232
00:11:16.559 --> 00:11:19.519
<v Speaker 1>They're about keeping quality, clarity, and control.

233
00:11:19.240 --> 00:11:22.919
<v Speaker 2>Exactly, maintaining sanity in this rapidly changing field.

234
00:11:23.159 --> 00:11:26.000
<v Speaker 1>Okay, so that brings us to beyond the seventy percent,

235
00:11:26.159 --> 00:11:29.480
<v Speaker 1>that crucial thirty percent where human skills are just irreplaceable.

236
00:11:29.519 --> 00:11:31.600
<v Speaker 1>This is where you really add value or future proof

237
00:11:31.600 --> 00:11:32.120
<v Speaker 1>of your career.

238
00:11:32.159 --> 00:11:34.919
<v Speaker 2>Absolutely, this is the zone of maximum human contribution.

239
00:11:35.200 --> 00:11:37.120
<v Speaker 1>So for senior engineers, what does this look like?

240
00:11:37.399 --> 00:11:39.960
<v Speaker 2>You become the architect, the editor in chief. Let the

241
00:11:40.000 --> 00:11:43.440
<v Speaker 2>AI be the fast typer generating the first draft, but

242
00:11:43.759 --> 00:11:46.120
<v Speaker 2>you are the brain. You define the architecture. You've vet

243
00:11:46.159 --> 00:11:48.080
<v Speaker 2>every line, you maintain the high standards.

244
00:11:48.679 --> 00:11:51.480
<v Speaker 1>You push back if people are just throwing raw AI

245
00:11:51.519 --> 00:11:52.279
<v Speaker 1>code over the wall.

246
00:11:52.440 --> 00:11:56.879
<v Speaker 2>Exactly. You also use AI as that force multiplier for bigger,

247
00:11:57.039 --> 00:12:02.080
<v Speaker 2>more ambitious projects. Steve yeggs these chat oriented programming or

248
00:12:02.480 --> 00:12:04.279
<v Speaker 2>chop idea fits here.

249
00:12:04.639 --> 00:12:07.759
<v Speaker 1>Chop iterative prompting, Yeah, refining prompts.

250
00:12:07.840 --> 00:12:10.960
<v Speaker 2>Collaborating with the AI. It lowers the barrier to entry

251
00:12:11.039 --> 00:12:14.840
<v Speaker 2>for those wouldn't be nice if projects that maybe seem too.

252
00:12:14.679 --> 00:12:18.799
<v Speaker 1>Big before and mentoring becomes even more critical for seniors.

253
00:12:18.600 --> 00:12:22.399
<v Speaker 2>Hugely coaching less experienced folks on how to use AI effectively,

254
00:12:22.480 --> 00:12:26.000
<v Speaker 2>how to self review how to test properly, championing diligence

255
00:12:26.039 --> 00:12:28.039
<v Speaker 2>and critical thinking about the AI's.

256
00:12:27.639 --> 00:12:32.000
<v Speaker 1>Output, Cultivating domain mastery using your history and business knowledge.

257
00:12:32.120 --> 00:12:35.639
<v Speaker 2>Right, that's how you catch the AI's subtle mistakes. Trust

258
00:12:35.639 --> 00:12:38.840
<v Speaker 2>your gut. When code looks off, your experience often spots

259
00:12:38.879 --> 00:12:40.759
<v Speaker 2>things the pattern matter misses.

260
00:12:40.559 --> 00:12:43.360
<v Speaker 1>And with AI handling some routine coding, seniors need to

261
00:12:43.399 --> 00:12:46.360
<v Speaker 1>double down on soft skills leadership.

262
00:12:45.879 --> 00:12:49.840
<v Speaker 2>Totally communicating with stakeholders, leading design discussions, making those tough

263
00:12:49.919 --> 00:12:53.519
<v Speaker 2>judgment calls. AI can't being the indispensable technical guide for

264
00:12:53.559 --> 00:12:53.919
<v Speaker 2>the team.

265
00:12:54.240 --> 00:12:57.919
<v Speaker 1>Okay, what about mid level engineers. They might feel squeezed.

266
00:12:57.960 --> 00:12:59.639
<v Speaker 1>Maybe tasks automating.

267
00:13:00.039 --> 00:13:03.559
<v Speaker 2>It's definitely a shift, but it's an elevation, not elimination.

268
00:13:03.919 --> 00:13:07.639
<v Speaker 2>The key is adapt and specialize. How so deepen those

269
00:13:07.679 --> 00:13:13.559
<v Speaker 2>CS fundamentals, data structures, algorithms, distributed systems, databases, networking. This

270
00:13:13.600 --> 00:13:17.639
<v Speaker 2>stuff helps you manage the tricky integrations, the boundaries between systems,

271
00:13:17.960 --> 00:13:20.320
<v Speaker 2>the edge cases AI flubs.

272
00:13:20.320 --> 00:13:22.120
<v Speaker 1>Become the integration expert.

273
00:13:22.320 --> 00:13:26.600
<v Speaker 2>Yeah, and build deep domain expertise. Finance, healthcare, and real

274
00:13:26.639 --> 00:13:30.879
<v Speaker 2>time systems, places where nuanced human understanding is vital. Bridge

275
00:13:30.879 --> 00:13:33.080
<v Speaker 2>that gap between code and real world impact.

276
00:13:33.200 --> 00:13:34.559
<v Speaker 1>Master performance and DevOps too.

277
00:13:34.679 --> 00:13:37.919
<v Speaker 2>Oh yeah, As generating code gets easier, making it perform

278
00:13:37.960 --> 00:13:41.960
<v Speaker 2>well and deploy reliably becomes super valuable. Understand the whole stack,

279
00:13:42.039 --> 00:13:43.919
<v Speaker 2>monitoring security costs.

280
00:13:43.679 --> 00:13:46.240
<v Speaker 1>And focus hard on code, reviewing QA relentlessly.

281
00:13:46.720 --> 00:13:49.519
<v Speaker 2>AI often gives you functional but horribly optimized code. As

282
00:13:49.519 --> 00:13:52.879
<v Speaker 2>Osmani says, you need to be the quality gatekeeper. Develop

283
00:13:52.919 --> 00:13:56.279
<v Speaker 2>a strong testing mindset, diagnose the complex bugs.

284
00:13:56.039 --> 00:13:58.639
<v Speaker 1>Learn systems big picture crucial.

285
00:13:58.960 --> 00:14:02.360
<v Speaker 2>AI doesn't currently get the user needs or business goals

286
00:14:02.440 --> 00:14:05.120
<v Speaker 2>unless you tell it very specifically. You need to hold

287
00:14:05.120 --> 00:14:07.080
<v Speaker 2>that context, see how all the pieces fit.

288
00:14:06.960 --> 00:14:09.840
<v Speaker 1>Together, and just keep learning right, be adapt.

289
00:14:09.879 --> 00:14:14.279
<v Speaker 2>On, stop, learn the fundamentals deeply, stay curious about AI tools,

290
00:14:14.600 --> 00:14:17.159
<v Speaker 2>maybe even do those AI detox days to keep your

291
00:14:17.200 --> 00:14:19.919
<v Speaker 2>own coding muscles strong, smart idea.

292
00:14:19.679 --> 00:14:22.919
<v Speaker 1>Communication skills too. Translating business needs.

293
00:14:22.720 --> 00:14:25.879
<v Speaker 2>More important than ever, and prompting AI well is actually

294
00:14:25.879 --> 00:14:29.840
<v Speaker 2>a form of precise communication itself. Asking good questions both

295
00:14:29.840 --> 00:14:31.200
<v Speaker 2>of humans and machines.

296
00:14:31.320 --> 00:14:33.200
<v Speaker 1>Learn system design and architecture.

297
00:14:33.279 --> 00:14:38.240
<v Speaker 2>Yes, designing robust, scalable, secure systems that needs an experienced

298
00:14:38.320 --> 00:14:41.679
<v Speaker 2>human hand on the wheel. AI offers limited help there

299
00:14:41.919 --> 00:14:42.960
<v Speaker 2>right now.

300
00:14:42.679 --> 00:14:45.080
<v Speaker 1>But still use the AI right, don't avoid.

301
00:14:44.759 --> 00:14:49.399
<v Speaker 2>It absolutely, use it daily, scaffolding prototypes, pair programming, repetitive stuff.

302
00:14:49.759 --> 00:14:51.639
<v Speaker 2>Let it handle that while you focus higher.

303
00:14:51.399 --> 00:14:54.759
<v Speaker 1>Up and maybe venture into UIUX thinking.

304
00:14:54.559 --> 00:14:58.159
<v Speaker 2>Yeah, develop better product sends, Collaborate better with designers. Understand

305
00:14:58.159 --> 00:15:01.480
<v Speaker 2>the user because coding itself might not be the bottleneck anymore.

306
00:15:01.639 --> 00:15:03.919
<v Speaker 2>Understanding what to build and why that's key.

307
00:15:04.240 --> 00:15:08.840
<v Speaker 1>Okay. Finally, junior developers lots of anxiety. Maybe headlines about

308
00:15:08.879 --> 00:15:09.399
<v Speaker 1>the death.

309
00:15:09.240 --> 00:15:13.120
<v Speaker 2>Of the junior death understandable anxiety, but the role is evolving,

310
00:15:13.240 --> 00:15:18.000
<v Speaker 2>not vanishing. The bar might be rising. You need stronger fundamentals,

311
00:15:18.039 --> 00:15:21.879
<v Speaker 2>faster perhaps, but you're needed to effectively use and validate

312
00:15:22.000 --> 00:15:22.720
<v Speaker 2>these tools.

313
00:15:22.799 --> 00:15:24.960
<v Speaker 1>So number one piece of advice.

314
00:15:24.600 --> 00:15:28.399
<v Speaker 2>For juniors, learn the fundamentals, don't skip the y. Use

315
00:15:28.480 --> 00:15:31.279
<v Speaker 2>AI like a tutor, ask it to explain things, but

316
00:15:31.440 --> 00:15:35.159
<v Speaker 2>build your own mental model of data, structures, algorithms, memory.

317
00:15:35.399 --> 00:15:37.759
<v Speaker 2>You need that to know when the AI is wrong.

318
00:15:38.000 --> 00:15:40.120
<v Speaker 1>Practice problem solving without the AI.

319
00:15:39.919 --> 00:15:43.559
<v Speaker 2>Sometimes yes, do those AI free days. Debug AI code

320
00:15:43.600 --> 00:15:46.279
<v Speaker 2>yourself first before asking the AI to fix it. That

321
00:15:46.320 --> 00:15:49.840
<v Speaker 2>builds real confidence and understanding. Don't outsource all the learning.

322
00:15:49.960 --> 00:15:53.559
<v Speaker 1>Focus on testing and verification. Challenge the AI's output exactly.

323
00:15:53.600 --> 00:15:56.320
<v Speaker 2>Write tests, for its code. Prof it works. Catching an

324
00:15:56.360 --> 00:15:58.919
<v Speaker 2>AI bug that's pure human value right there.

325
00:15:59.039 --> 00:16:02.000
<v Speaker 1>Build an eye for maintainability. Go beyond just making it work.

326
00:16:02.080 --> 00:16:05.840
<v Speaker 2>Yeah, look at code structure style, refactor messy AI code.

327
00:16:05.879 --> 00:16:07.679
<v Speaker 2>Internalize good design principle.

328
00:16:07.720 --> 00:16:10.360
<v Speaker 1>Develop prompt skills but wisely it's useful.

329
00:16:10.399 --> 00:16:14.000
<v Speaker 2>Sure, but good prompting often just reflects good understanding of

330
00:16:14.000 --> 00:16:18.600
<v Speaker 2>the problem itself. Use AI like a supercharged stack overflow,

331
00:16:18.960 --> 00:16:21.279
<v Speaker 2>a powerful assistant, not an oracle.

332
00:16:21.519 --> 00:16:25.039
<v Speaker 1>Seek feedback, mentorship from humans absolutely critical.

333
00:16:25.480 --> 00:16:28.080
<v Speaker 2>Learn the nuances, the trade offs, the soft skills from

334
00:16:28.120 --> 00:16:31.600
<v Speaker 2>your team. Be open to feedback on your AI assisted code.

335
00:16:31.440 --> 00:16:35.240
<v Speaker 1>And remember communication and collaboration software as a team sport.

336
00:16:35.039 --> 00:16:38.919
<v Speaker 2>Always has been, always will be. Humans clarify requirements, coordinate,

337
00:16:39.320 --> 00:16:43.399
<v Speaker 2>Even prompting is communication. Shift your mindset. Don't just consume

338
00:16:43.480 --> 00:16:46.879
<v Speaker 2>AI answers, Understand them, dissect them. Learn how you could

339
00:16:46.919 --> 00:16:48.799
<v Speaker 2>have gotten there yourself. That's how you level up.

340
00:16:48.919 --> 00:16:52.080
<v Speaker 1>So pulling it all together. The durable skills for the

341
00:16:52.159 --> 00:16:55.360
<v Speaker 1>future in this AI world seem to be this in

342
00:16:55.440 --> 00:16:59.960
<v Speaker 1>design systems thinking, critical thinking, domain expertise, quality is sharing

343
00:17:00.080 --> 00:17:03.600
<v Speaker 1>and communication, adaptability, continuous learning.

344
00:17:03.720 --> 00:17:06.279
<v Speaker 2>That's the core list. These skills are durable because they

345
00:17:06.319 --> 00:17:09.440
<v Speaker 2>don't become obsolete with the next AI model. If anything,

346
00:17:09.480 --> 00:17:11.079
<v Speaker 2>AI makes them more important, right.

347
00:17:10.960 --> 00:17:14.640
<v Speaker 1>As Simon Wilson says, AI makes strong programming skills more valuable,

348
00:17:14.680 --> 00:17:15.240
<v Speaker 1>not less.

349
00:17:15.640 --> 00:17:19.319
<v Speaker 2>Exactly, think of an experienced engineer like a seasoned pilot

350
00:17:19.359 --> 00:17:23.559
<v Speaker 2>with a super advanced copilot. Faster journey, maybe farther, but

351
00:17:23.599 --> 00:17:26.960
<v Speaker 2>the pilot still needs to navigate, make decisions, land the

352
00:17:26.960 --> 00:17:27.720
<v Speaker 2>plane safely.

353
00:17:28.000 --> 00:17:31.480
<v Speaker 1>Software engineering was always about solving problems really, not just

354
00:17:31.559 --> 00:17:33.240
<v Speaker 1>writing code lines Precisely.

355
00:17:33.559 --> 00:17:36.559
<v Speaker 2>AI doesn't change the fundamental goal. It just challenges us

356
00:17:36.559 --> 00:17:39.359
<v Speaker 2>to solve problems at a higher level, maybe tackle things

357
00:17:39.359 --> 00:17:40.799
<v Speaker 2>we thought were impossible before.

358
00:17:41.079 --> 00:17:44.680
<v Speaker 1>So the future involves less typing, maybe more directing and

359
00:17:44.759 --> 00:17:45.240
<v Speaker 1>curating

360
00:17:45.400 --> 00:17:49.160
<v Speaker 2>Seems likely, but it will always always need human developers

361
00:17:49.160 --> 00:17:51.839
<v Speaker 2>at the center, people with the wisdom, the judgment, the

362
00:17:51.880 --> 00:17:54.799
<v Speaker 2>deep understanding to actually do it right.