1 00:00:00,040 --> 00:00:03,120 The current state of the employment market is quite 2 00:00:03,120 --> 00:00:06,840 interesting. The data on junior employment 3 00:00:07,640 --> 00:00:10,880 companies that are adopting AI and really drinking the kool-aid 4 00:00:10,880 --> 00:00:14,360 of AI is is really only moving in One Direction. 5 00:00:15,040 --> 00:00:19,840 And right now across the S&P 500 headcount fell in 2025 for the 6 00:00:19,840 --> 00:00:23,840 first time since 2016. That's a net reduction of around 7 00:00:23,840 --> 00:00:28,800 400,000 jobs ending 8 consecutive years of just 8 00:00:28,800 --> 00:00:31,760 constant workforce growth. This discussion is going to be 9 00:00:31,760 --> 00:00:35,360 one of the most important discussions to have as a society 10 00:00:35,360 --> 00:00:38,920 over the next few years, and there's some really compelling 11 00:00:38,920 --> 00:00:40,520 arguments against worrying about it. 12 00:00:41,200 --> 00:00:42,960 Some of the people that I'm going to mention in today's 13 00:00:42,960 --> 00:00:46,480 episode have discussed the fact that there's been many waves of 14 00:00:46,480 --> 00:00:50,240 technology that's reshaped, for example, programming and often 15 00:00:50,240 --> 00:00:53,680 thinks the doomsday narrative mysteries, history. 16 00:00:54,080 --> 00:00:56,280 But both things can be true at the same time. 17 00:00:57,080 --> 00:00:59,720 We can be an optimistic and pessimistic view of the future. 18 00:01:01,360 --> 00:01:04,160 Today's episode's going to explore those views, the data 19 00:01:04,160 --> 00:01:07,920 points in the history behind them and why the companies that 20 00:01:07,920 --> 00:01:11,880 are cutting junior headcount are going to make a decade long 21 00:01:11,960 --> 00:01:14,040 mistake. This episode brings together 22 00:01:14,040 --> 00:01:17,400 many forms of the latest research and thinking in this 23 00:01:17,400 --> 00:01:20,480 field to try and present the most up to date. 24 00:01:20,480 --> 00:01:23,280 View this in the loop with Jack Horton. 25 00:01:23,880 --> 00:01:38,320 Hope you enjoy the show. So I guess let's start off with 26 00:01:38,320 --> 00:01:41,600 some of the, the context and the data behind, you know, where 27 00:01:41,600 --> 00:01:44,160 we're at today. And, and I think a really 28 00:01:44,160 --> 00:01:47,240 interesting place to start here is, is a fantastic comparison 29 00:01:47,240 --> 00:01:49,480 that I found online. And I, I fell in love with it. 30 00:01:49,760 --> 00:01:54,120 So in January 1971, a billboard got put up on the, I was going 31 00:01:54,120 --> 00:01:57,160 to say motorway, but in the US they're called highways around 32 00:01:57,160 --> 00:01:59,440 Seattle. And it read with the last person 33 00:01:59,440 --> 00:02:01,240 leaving Seattle, turn off the lights. 34 00:02:02,360 --> 00:02:04,720 And, you know, Boeing, the context behind this is Boeing. 35 00:02:04,720 --> 00:02:11,480 It had started 1969 with 134,000 employees and ended in 1971 with 36 00:02:11,480 --> 00:02:16,360 about less than 80,000. So, you know, about 4445 37 00:02:16,360 --> 00:02:19,880 thousand employees have been let go in 18 months. 38 00:02:21,000 --> 00:02:25,120 They wound on the space program. They had a big supersonic, you 39 00:02:25,120 --> 00:02:27,960 know, transport project that had been completely cancelled. 40 00:02:28,520 --> 00:02:31,360 And Boeing claimed they didn't need engineers anymore. 41 00:02:31,960 --> 00:02:35,120 But you look at today, and the aerospace industry is projecting 42 00:02:35,120 --> 00:02:38,400 a shortage of more than a million engineers by 20. 43 00:02:38,400 --> 00:02:42,600 Thirty, 25% of the current workforce is over 55 years old. 44 00:02:43,560 --> 00:02:49,040 And you know that the cohort that would now say be the senior 45 00:02:49,880 --> 00:02:53,240 experienced employees never ever got hired. 46 00:02:53,800 --> 00:02:56,000 And that really sets the scene for today's discussion. 47 00:02:56,000 --> 00:03:00,320 You know why not hiring junior employees is going to really 48 00:03:00,320 --> 00:03:03,280 hurt the industry in 1015 or 20 years. 49 00:03:04,520 --> 00:03:07,200 And this episode isn't about just whether or not AI is just 50 00:03:07,200 --> 00:03:09,400 destroying jobs, which is often the version of this, I guess, 51 00:03:10,320 --> 00:03:15,120 conversation that generates a lot of big headlines and noise. 52 00:03:15,360 --> 00:03:17,640 It's actually just talking about the realities of talent 53 00:03:17,640 --> 00:03:22,280 pipelines and specifically what happens when the entry point of 54 00:03:22,280 --> 00:03:26,160 a career is closed before people have had the opportunity to 55 00:03:26,160 --> 00:03:30,840 start on their journey. As I said, the S&P 500 across as 56 00:03:30,840 --> 00:03:35,080 a whole in headcount fell in 2025, which was the first since 57 00:03:35,080 --> 00:03:40,000 2016, a net reduction of about 400,000 jobs, which is ending 58 00:03:40,000 --> 00:03:42,080 eight years of basically uninterrupted growth. 59 00:03:42,200 --> 00:03:45,600 And and that's a massive signal. We've had the multiple research 60 00:03:45,600 --> 00:03:48,680 projects released in the last year all pointing broadly in the 61 00:03:48,680 --> 00:03:52,000 same direction, which is a companies that have adopted AI 62 00:03:52,000 --> 00:03:56,680 junior employment drop of between about 7 to 10% within 63 00:03:57,280 --> 00:04:01,360 about a year and a half, while senior employment at those same 64 00:04:01,360 --> 00:04:04,880 places keeps rising. Doesn't take a genius to figure 65 00:04:04,880 --> 00:04:08,000 out that that isn't sustainable. And this decline by the way, in 66 00:04:08,000 --> 00:04:11,600 most places has happened through slower hiring rather than just 67 00:04:11,600 --> 00:04:15,720 redundancies as a whole. So people have stopped bringing 68 00:04:15,720 --> 00:04:19,200 people in at the bottom and and we've had another study for 69 00:04:19,200 --> 00:04:21,560 example that drilled into the specifics of how that's 70 00:04:21,560 --> 00:04:26,240 happening and they found that workers aged 22 to 25 in an AI 71 00:04:26,240 --> 00:04:29,360 exposed job and we can discuss what that is defined as shortly, 72 00:04:30,360 --> 00:04:34,760 are basically down 13% relative to their peers in less exposed 73 00:04:34,760 --> 00:04:40,080 jobs since 2022-2023. For example, 22 to 25 year old, 74 00:04:40,080 --> 00:04:43,120 let's say software engineers, it's around 20%. 75 00:04:43,880 --> 00:04:47,400 So that's like nearly 25% of the developers that would have been 76 00:04:47,400 --> 00:04:49,840 hired and no longer being hired anymore. 77 00:04:50,360 --> 00:04:54,040 That's a huge cohort of people. And a really good telltale here 78 00:04:54,040 --> 00:04:56,920 is that the gap between where employment should be on, let's 79 00:04:56,920 --> 00:05:03,400 say pre ChatGPT employment trends is 500,000 less jobs than 80 00:05:03,480 --> 00:05:06,400 was expected. So there are 500,000 less 81 00:05:06,480 --> 00:05:09,560 developers than there was predicted based on the current 82 00:05:09,560 --> 00:05:11,400 trajectory before ChatGPT released. 83 00:05:12,080 --> 00:05:13,760 And it's a huge mistake for the industry. 84 00:05:14,120 --> 00:05:16,720 Like yes, maybe some job roles are going to be compressed 85 00:05:16,720 --> 00:05:18,920 unless you maybe there are new job roles that will emerge. 86 00:05:19,320 --> 00:05:24,280 But not hiring junior employees and and training people is 87 00:05:24,720 --> 00:05:27,720 obviously unsustainable. So let's explore why this is 88 00:05:27,720 --> 00:05:31,720 happening to junior employees specifically because it's less 89 00:05:31,720 --> 00:05:35,920 obvious than you think. Let's take Luis Garciano, who 90 00:05:35,920 --> 00:05:38,240 who presented a really interesting argument here. 91 00:05:38,920 --> 00:05:41,760 And Garciano is a Spanish economist at the London School 92 00:05:41,760 --> 00:05:44,720 of Economics, and his main work has been exploring how 93 00:05:44,720 --> 00:05:46,480 organization structure knowledge. 94 00:05:46,880 --> 00:05:49,960 So who knows what who does what? You know, how information might 95 00:05:49,960 --> 00:05:55,000 flow through a group of people. And he said a job isn't a list 96 00:05:55,000 --> 00:05:59,120 of tasks that can be chopped up and automated arbitrarily. 97 00:05:59,960 --> 00:06:04,840 It's a bundle, So a bundle of tasks that are entangled into 98 00:06:04,840 --> 00:06:07,440 each other. The question AI pose isn't can 99 00:06:07,440 --> 00:06:11,480 this task be automated? It's can this task be separated 100 00:06:11,720 --> 00:06:14,840 from the rest of the job without unraveling the whole bundle of 101 00:06:14,840 --> 00:06:16,880 tasks. And basically what he's saying 102 00:06:16,880 --> 00:06:18,800 here is that a job isn't just a set checklist. 103 00:06:19,520 --> 00:06:21,520 You know, some parts are completely independent and 104 00:06:21,520 --> 00:06:24,600 freestanding, so you can lift them out, hand them to someone 105 00:06:24,600 --> 00:06:26,400 else and nothing like breaks down. 106 00:06:27,160 --> 00:06:29,760 Whereas some parts are woven into everything else and the 107 00:06:29,760 --> 00:06:32,440 relationships, the judgement calls the accountability. 108 00:06:32,440 --> 00:06:36,040 And if you try to separate those tasks, you make the entire job 109 00:06:36,040 --> 00:06:39,000 role collapse. And junior work is typically 110 00:06:39,360 --> 00:06:42,280 those who have the the most, let's say freestanding 111 00:06:42,280 --> 00:06:44,440 independent tasks that can be automated. 112 00:06:45,000 --> 00:06:47,600 But the problem here is that the only way to become a senior 113 00:06:47,960 --> 00:06:51,880 advanced experienced employee is to spend years doing the junior 114 00:06:51,880 --> 00:06:53,880 work. You know, the the first year 115 00:06:54,000 --> 00:06:58,080 legal associates document review is separable, so you can 116 00:06:58,080 --> 00:07:00,880 automate it. Same with the junior analysts, 117 00:07:00,920 --> 00:07:04,680 you know, financial model or the graduates, I don't know, 118 00:07:04,680 --> 00:07:08,600 research deck who they produce for, you know, leaders of the 119 00:07:08,600 --> 00:07:10,600 company. These tasks can be just lifted 120 00:07:10,600 --> 00:07:13,120 out and done by AI really quickly and easily. 121 00:07:13,520 --> 00:07:16,760 And then the tasks that resist that kind of separation and 122 00:07:16,760 --> 00:07:20,080 automation are those that are senior jobs reading, for 123 00:07:20,080 --> 00:07:24,120 example, a witness in a disposition and deciding whether 124 00:07:24,120 --> 00:07:27,480 to settle on or pricing, let's say a deal and managing the 125 00:07:27,480 --> 00:07:30,360 board and the stakeholders through that deal. 126 00:07:31,120 --> 00:07:33,600 Choosing which pitch to take, for example, into a chief 127 00:07:33,600 --> 00:07:36,720 marketing officer and the person that reads it in the room once 128 00:07:36,720 --> 00:07:37,800 you're there. These are these are very 129 00:07:37,800 --> 00:07:42,640 entangled tasks that are heavily dependent and have a massive 130 00:07:42,640 --> 00:07:45,800 impact on the relationships or contexts or consequences and 131 00:07:46,560 --> 00:07:49,400 accountabilities that hold the whole job role together. 132 00:07:49,960 --> 00:07:52,280 So your junior work doesn't have that protection. 133 00:07:52,280 --> 00:07:55,080 And the pattern isn't unique to just software engineers. 134 00:07:56,200 --> 00:07:59,160 It's happening in journalism, in architecture, basically 135 00:07:59,160 --> 00:08:01,200 anywhere. The entry level work is 136 00:08:01,760 --> 00:08:04,840 primarily about turning information into more structured 137 00:08:04,840 --> 00:08:09,080 output versus having to exercise, you know, really 138 00:08:09,080 --> 00:08:12,040 difficult judgement calls with real consequences attached. 139 00:08:12,720 --> 00:08:15,000 But the problem is this creates a massive pipeline problem. 140 00:08:15,560 --> 00:08:18,320 You know, if AI does the junior work, there's no economic reason 141 00:08:18,320 --> 00:08:22,480 to ever hire a junior employee. And if nobody hires them, how do 142 00:08:22,480 --> 00:08:25,120 they become experienced senior employees of a company? 143 00:08:26,360 --> 00:08:27,920 You know, the pipeline didn't just slow down, it just 144 00:08:27,920 --> 00:08:46,210 completely stops. So I guess that's a very doom 145 00:08:46,210 --> 00:08:48,410 and gloom view. And, and there are arguments for 146 00:08:48,410 --> 00:08:50,570 or against that point of view here. 147 00:08:51,210 --> 00:08:54,770 And and let's say let's start on the optimistic side because it's 148 00:08:54,770 --> 00:08:56,330 it's useful and really important to know. 149 00:08:56,330 --> 00:09:00,010 So Tim O'Reilly, you might recognize his name O'Reilly 150 00:09:00,010 --> 00:09:02,610 publications, but you got his books at school. 151 00:09:03,050 --> 00:09:06,560 But he has some really interesting essays. 152 00:09:07,000 --> 00:09:09,200 Now, he's been a developer in his entire life. 153 00:09:09,520 --> 00:09:12,240 He's a very, very unique and fantastic point of view here. 154 00:09:12,600 --> 00:09:15,400 And his argument is, is that he's been watching programming, 155 00:09:15,400 --> 00:09:17,960 say, transition for, you know, 40 odd years. 156 00:09:18,680 --> 00:09:21,560 And a lot of the time these types of panics misread history. 157 00:09:21,960 --> 00:09:26,600 You know, every new programming wave from assembly to Fortran to 158 00:09:27,360 --> 00:09:33,160 C to consumer BASIC to Windows to cloud look from when you 159 00:09:33,160 --> 00:09:36,720 were, I guess, standing in the middle of it like the end of 160 00:09:36,720 --> 00:09:38,720 programming. Every single one of them turned 161 00:09:38,720 --> 00:09:40,560 out to be just an expansion of programming. 162 00:09:41,240 --> 00:09:43,560 But the expansion wasn't just that the bar got lower, some 163 00:09:43,560 --> 00:09:45,720 more of the same people could get involved. 164 00:09:45,720 --> 00:09:49,360 Each way brought with it an entirely new skill set and 165 00:09:49,360 --> 00:09:52,680 population that previously hadn't been involved. 166 00:09:53,160 --> 00:09:56,520 So Fortran didn't make, for example, assembly programmers 167 00:09:56,640 --> 00:09:58,480 pointless. It brought scientists and 168 00:09:58,480 --> 00:10:01,760 engineers into computing. People would have never got 169 00:10:01,760 --> 00:10:05,400 involved with machine code. Consumer Basic didn't, let's 170 00:10:05,400 --> 00:10:08,160 say, replace C program as it brought in teenagers in the 171 00:10:08,160 --> 00:10:11,880 bedrooms, A demographic who couldn't have ever been involved 172 00:10:11,880 --> 00:10:14,280 in the industry. Operating systems like Windows 173 00:10:14,280 --> 00:10:16,520 didn't stop people building applications. 174 00:10:17,200 --> 00:10:20,120 It meant that an entire industry of applications is born as a 175 00:10:20,120 --> 00:10:23,440 result. So the pyramid got wider every 176 00:10:23,440 --> 00:10:25,920 single level, and the people from the previous way didn't 177 00:10:25,920 --> 00:10:29,440 disappear, they just moved up. Their skills became foundations 178 00:10:29,440 --> 00:10:33,040 that basically others just would build on, rather than suddenly 179 00:10:33,040 --> 00:10:35,600 this new ceiling that meant that they were screwed and they had 180 00:10:35,600 --> 00:10:39,400 to leave the job. Now personally, I used to buy 181 00:10:39,400 --> 00:10:42,360 into this belief and recently I've been wondering whether that 182 00:10:42,360 --> 00:10:45,600 is still the case. Now, one argument is the Javon's 183 00:10:45,600 --> 00:10:47,320 paradox. So I've talked about this a 184 00:10:47,320 --> 00:10:49,720 number of times. And this is when something 185 00:10:49,720 --> 00:10:52,800 becomes so readily available, whether it's affordable or easy 186 00:10:52,800 --> 00:10:55,400 to do, whatever it might be efficient, suddenly there's an 187 00:10:55,400 --> 00:10:58,760 explosion of that thing. At first it looks like the 188 00:10:58,880 --> 00:11:02,680 efficiency gain from the technology that enabled it would 189 00:11:02,680 --> 00:11:04,560 mean that people would lose their jobs. 190 00:11:04,560 --> 00:11:08,360 But just as the as a result of the sheer adoption of a thing, 191 00:11:08,640 --> 00:11:11,560 there is so many more applications and jobs as a 192 00:11:11,560 --> 00:11:14,520 result. So, you know, steam didn't 193 00:11:14,600 --> 00:11:17,720 remove the need for coal, it actually increased it because 194 00:11:17,720 --> 00:11:20,640 efficiency made steam power economically viable in 195 00:11:20,800 --> 00:11:23,240 applications where previously it would have been impossible to 196 00:11:23,240 --> 00:11:25,360 justify the cost. You can apply that to 197 00:11:25,360 --> 00:11:28,640 engineering, which is, you know, making people 1020 more 20 times 198 00:11:28,640 --> 00:11:31,600 more productive might seem like you may need less of them. 199 00:11:31,800 --> 00:11:34,720 But over time the argument is that something there's loads of 200 00:11:34,720 --> 00:11:40,040 new tools that you would build that would never have ever been 201 00:11:40,120 --> 00:11:44,560 justified or or even considered. And as a result, more engineers 202 00:11:44,560 --> 00:11:46,520 needed. Now where I have an issue with 203 00:11:46,520 --> 00:11:48,440 this, to be honest, is the time scale. 204 00:11:48,840 --> 00:11:51,200 This argument is basically about where the industry could be in 205 00:11:51,400 --> 00:11:55,760 2035 or 2040, but the pipeline problem I'm really discussing 206 00:11:55,760 --> 00:12:00,400 here is about 2028 to 2033. You know, if there's no genius 207 00:12:00,400 --> 00:12:05,680 in 20252026, who were the seniors by 20-30? 208 00:12:06,080 --> 00:12:08,680 That requires a specific cohort to have been in the job for 209 00:12:08,680 --> 00:12:12,480 five, 6-7 years, and those people aren't being hired. 210 00:12:12,960 --> 00:12:16,160 You can't suddenly have more software getting built in 2035 211 00:12:16,320 --> 00:12:19,960 and needing more employees and and hires but having less people 212 00:12:19,960 --> 00:12:22,520 to hire because there are just less careers there. 213 00:12:23,160 --> 00:12:27,320 And the Jevons paradox is essentially an aggregate outcome 214 00:12:27,680 --> 00:12:30,640 across the entire economy over a longer period of time. 215 00:12:31,640 --> 00:12:34,760 So it's the culmination of all these things happening over a 216 00:12:34,760 --> 00:12:38,280 10/15/20 year period. But that pipeline problem that 217 00:12:38,280 --> 00:12:42,400 I'm actually discussing is about a very specific set period of 218 00:12:42,400 --> 00:12:44,600 time. To have seniors in the job for 219 00:12:44,600 --> 00:12:47,160 7-8 years, they need to have been hired 7-8 years ago. 220 00:12:47,840 --> 00:12:51,080 Another argument against that I guess positive outlook right 221 00:12:51,080 --> 00:12:54,000 now, and they're doing just seeing somewhat negative here is 222 00:12:54,520 --> 00:12:57,960 that every previous say technology wave, so Fortran 223 00:12:58,680 --> 00:13:03,000 BASIC cloud essentially arrived after the experience of the 224 00:13:03,000 --> 00:13:05,400 previous layer had already kind of been established. 225 00:13:05,880 --> 00:13:09,000 Fortran programmers had spent years working in Fortran before 226 00:13:09,000 --> 00:13:12,360 it then got abstracted away. C programmers broad our 227 00:13:12,360 --> 00:13:15,480 understanding of same memory management into JavaScript and 228 00:13:16,760 --> 00:13:19,840 each abstraction essentially is being built on the last group of 229 00:13:19,840 --> 00:13:22,160 people who had actually done the thing. 230 00:13:22,440 --> 00:13:25,200 The wave would come after the experience form, not before it. 231 00:13:26,080 --> 00:13:28,480 But this current wave is just happening so much faster. 232 00:13:29,160 --> 00:13:32,560 You know something really is happening to entry level work 233 00:13:32,560 --> 00:13:34,840 faster than any previous transition in history. 234 00:13:35,160 --> 00:13:38,240 Although the Chevron's paradox in the long run, 1520 years, 235 00:13:38,240 --> 00:13:42,000 period, period of time might be true, there might be just such 236 00:13:42,000 --> 00:13:45,800 an explosion of things over this next long period that eventually 237 00:13:45,800 --> 00:13:49,320 there's more, there's more jobs, but over the shorter window, 238 00:13:50,080 --> 00:13:52,560 companies are making a massive mistake by not thinking about 239 00:13:52,560 --> 00:13:56,600 their junior pipelines. So the Gevons paradox argument, 240 00:13:56,600 --> 00:14:00,360 that's very optimistic and the pipeline problem that I'm really 241 00:14:00,360 --> 00:14:02,120 discussing here aren't competing. 242 00:14:02,640 --> 00:14:04,240 It's kind of like a sequence of events. 243 00:14:04,920 --> 00:14:07,640 There needs to be an expansion of demand, but the expertise 244 00:14:07,640 --> 00:14:09,640 just isn't there anymore to meet it. 245 00:14:10,320 --> 00:14:12,840 And you eventually get that shortage that I mentioned about 246 00:14:12,840 --> 00:14:14,480 Boeing facing in the aerospace industry. 247 00:14:14,960 --> 00:14:18,360 And that will happen in 2035 rather than in 50 years later, 248 00:14:18,520 --> 00:14:37,550 but it will happen. So I guess to conclude here that 249 00:14:37,550 --> 00:14:40,430 the arguments to say that each transition of new technology and 250 00:14:40,430 --> 00:14:44,190 say programmers and software development really did expand 251 00:14:44,190 --> 00:14:47,950 the surface area the the amount of potential in the space. 252 00:14:48,800 --> 00:14:52,560 And that is a historical analogy that will probably play out to 253 00:14:52,560 --> 00:14:55,600 be true. But the problem right now, and 254 00:14:55,600 --> 00:14:57,600 This is why I think companies are making a massive mistake, is 255 00:14:57,600 --> 00:15:02,120 that the cohort that will become the leaders of a company in 2032 256 00:15:02,120 --> 00:15:07,160 and 2035 aren't being hired and therefore won't be trained. 257 00:15:07,920 --> 00:15:11,280 And there are companies that are doing the right thing here. 258 00:15:11,720 --> 00:15:15,040 IBM tripled its entry level intake in 2026, one of the very 259 00:15:15,040 --> 00:15:16,680 few companies that had done done this. 260 00:15:17,920 --> 00:15:21,320 Publicist, the marketing group tried to restrict its graduate 261 00:15:21,320 --> 00:15:24,160 training program around AI fluency rather than just getting 262 00:15:24,160 --> 00:15:26,760 rid of it. But these are a drop in the 263 00:15:26,760 --> 00:15:29,920 ocean in terms of the general statistics across the the market 264 00:15:29,920 --> 00:15:32,400 right now. And that that billboard that 265 00:15:32,400 --> 00:15:36,800 people put up in Seattle, it's a joke, but it's a reality. 266 00:15:36,800 --> 00:15:40,440 Now, 50 years later, the aerospace companies who made 267 00:15:40,440 --> 00:15:44,560 those cuts are now probably, looking back, really regretting 268 00:15:44,560 --> 00:15:47,480 some stupid decisions. And many of these big, big 269 00:15:47,480 --> 00:15:50,680 companies especially are going to do the same and are going to 270 00:15:50,680 --> 00:15:52,320 make the same mistake in the future. 271 00:15:53,320 --> 00:15:57,000 Anyway, thank you for listening and I'll see you next week.