The Podcast of the Lotus Eaters - August 18, 2026


PREVIEW： Brokenomics ｜ Will AI Create or Destroy Jobs？


Episode Stats


Length

24 minutes

Words per minute

177.98

Word count

4,406

Sentence count

61


Transcript

Transcript generated with Whisper (turbo).
00:00:00.000 Hello and welcome to Brokonomics.
00:00:25.160 Now, in this episode, I thought I would have a chat about AI because I have noticed that I quite like it.
00:00:32.380 I think it's interesting and I think it's going to be transformational like the Internet or computers all over again.
00:00:38.980 But I have noticed that lots of commenters seem to really hate it and to the point where they appear quite angry that I don't dismiss it.
00:00:51.140 and and it's not that they it's not that they just don't think it's a thing and therefore just kind
00:00:57.780 of don't really care about it it's much more that they are actually angry about ai and i'm trying to
00:01:05.440 get my head around this it's like well why why i mean why would you be angry about it and i can't
00:01:10.160 really remember now if people were just i guess they must have been angry when the internet came
00:01:16.220 along with the displacement that it causes or that well a lot of people i do remember a lot
00:01:21.380 of people thinking that it was going to be massively overhyped certainly the computers
00:01:25.320 and computers are well documented that the back when computers first came along somebody fairly
00:01:30.160 famous i can't remember who it was now but somebody fairly famous said there's demand for about seven
00:01:33.880 computers in the world uh which is wrong uh we have more than seven computers in the world in
00:01:40.140 fact i think i will certainly have more than seven computers myself yes i do especially if you count
00:01:47.080 smartphones which are definitely computers the rest of the family yes i'm sure i do anyway point
00:01:52.200 is i i wanted to address this question of people really hating it and is that because they think
00:02:01.580 it's going to be disruptive it's going to take something away and this kind of feeds into the
00:02:06.720 job debate. Is AI going to come along and destroy jobs and basically diminish us? It's just going
00:02:14.560 to leave us with facsimiles of things as opposed to real things, real creation and so on. Now,
00:02:20.960 that is a broader philosophical point, but I can certainly address the jobs thing and look at the
00:02:26.100 profile of jobs that are likely to be destroyed and the profile of the jobs that are likely to
00:02:31.820 increase now i'll give you my broad view i think ai is going to both destroy and create an awful
00:02:40.220 lot of jobs and on net it will create more jobs but we need to kind of dig into what are the
00:02:48.360 characteristics of the ones that are going away and what are the characteristics of the ones that
00:02:51.480 are going to be more of now you mean and put it very simply i think what a lot of people are
00:02:59.440 thinking is, look, if one worker plus AI can do the jobs of five workers, well, four workers are
00:03:06.280 going to disappear then. And I think that is almost certainly going to happen at the level
00:03:11.860 of the firm. But I think that across the economy as a whole, actually, you're going to see the
00:03:19.160 complete opposite. You're going to see total number of demand for jobs, demand for people
00:03:25.080 increase and what i'm leaning on here is an old economic idea called jevon's paradox so jevon
00:03:33.780 can't remember his name now rest of his name sure he had the first name let's call him bob bob jevon
00:03:39.000 he worked in um steam engines that kind of stuff coal essentially coal-based engines in 1865 sounds
00:03:48.480 about right anyway jevin noticed that as coal-based engines steam engines essentially
00:03:55.280 were becoming more efficient and using less coal per engine per unit of work
00:04:01.360 the demand for coal was increasing not decreasing which you might think well hang on if you if you're
00:04:10.520 using yet less coal per per engine why wouldn't why wouldn't you have total demand for coal go
00:04:15.800 down and actually know it's completely opposite what happened is that as engine efficiency
00:04:21.160 increased essentially as it became cheaper per unit of coal to do a unit of work the demand for
00:04:28.860 work increased as opposed to decrease now this is and and this is something that the economists
00:04:35.560 have found oh that's very interesting and they've studied it ever since and i mean computers internet
00:04:39.700 I mean, a whole bunch of different new technologies have come along, and Jevons' paradox seems
00:04:46.060 to apply.
00:04:46.660 You make doing work cheaper.
00:04:48.400 You get more work.
00:04:49.240 You don't get less work.
00:04:51.560 Although it's not universal, and essentially what it all comes back to is demand elasticity.
00:04:56.940 Is there a fixed demand for the thing, or is there a demand which isn't currently being
00:05:06.680 met?
00:05:07.040 now with the industrial revolution you can see how they would have thought about it at the time
00:05:12.060 right you they would have been looking at this and thinking well there are only so many loom jobs
00:05:17.180 there are only so many mill jobs or you know whatever jobs that were around in 1865
00:05:23.740 and if if steam comes along and it can replace a lot of the jobs in these industries well there's
00:05:32.160 not going to be any jobs and actually what happened is the ability to do those things
00:05:38.400 became cheaper required less people and then those people were like oh bugger we've got nothing to do
00:05:44.600 now and then they thought something to do and then they did something else and the number of jobs that
00:05:50.520 existed in i mean even even 50 years later 1915 that simply didn't exist in 1865 enormous and
00:06:00.560 And that process has just continued ever since.
00:06:02.720 I mean, the job that I'm doing now didn't exist before the internet.
00:06:05.940 I mean, it didn't really exist until innovations that were built off the back of the internet,
00:06:12.180 freeing up people, led to, okay, social media, early YouTube.
00:06:16.920 Then Carl came along in whatever it was, 2013, I think he started.
00:06:21.320 I mean, he was early.
00:06:22.620 It doesn't seem that long ago to me, but then I'm old.
00:06:26.880 New categories of jobs appear.
00:06:28.500 And now, I mean, my daughter's school, they gave them an end-of-year book,
00:06:33.360 and they asked everybody what they wanted to do.
00:06:35.460 And I remember spotting that 7% of them wanted to become YouTubers,
00:06:39.780 a job that just didn't exist when their parents were born,
00:06:43.420 unless they got their young parents, which is possible.
00:06:46.100 But we were in a nice area, so we didn't get a lot of teen pregnancy.
00:06:49.000 So, yes, anyway, waffled about that.
00:06:52.160 The point is, right, okay, how are we going to apply this to AI?
00:06:56.220 Because we want to try and learn something useful out of this,
00:06:58.500 so that we can look at, well, maybe our own job
00:07:02.020 or our children's job if they're getting into something
00:07:04.400 and say, well, how do you navigate through this
00:07:06.260 so that you have a job, assuming you want a job?
00:07:10.120 You might be in Britain and it actually pays better
00:07:12.880 not to have a job, but you get where I'm going with this.
00:07:15.940 So look, payroll, that's a good example.
00:07:20.900 In any given company, you have a number of staff
00:07:24.900 and if you can process the payroll quicker you don't need as many people doing the payroll if
00:07:32.120 they're doing it manually you obviously need more but if it if it's one person and an ai
00:07:35.920 when you can process the payroll quicker i mean the demand is is very finite is get these number
00:07:41.060 of payrolls done by the end of in fact actually a really good example is um a bit before my time
00:07:49.260 But I remember when I was a young man speaking to an older man who was an accountant, and one of his friends was an accountant at a big firm, and his job was to produce the month-end management accounts.
00:08:04.780 And it took him a whole month to do this.
00:08:08.040 So there was always a month lag in getting your month-end reports from the previous book one month.
00:08:14.920 And it took him a whole month to do this.
00:08:16.760 and he nagged and he nagged and he nagged until he got this software package basically he was
00:08:23.560 doing i mean i guess he had a computer for excel or something but he didn't have a package which
00:08:28.540 sort of just did it all for him and they eventually got him this package and it it lowered the time
00:08:36.940 it took him to produce the month end management accounts from a whole month to one day
00:08:43.640 and he was savvy enough not to tell anyone and so he he used to do his work in a day and then he
00:08:52.080 spent the rest of his time printing out maps of his golf course and then basically sellotaping
00:09:00.000 them together on a table and plotting his game because golf was was important to him for whatever
00:09:07.660 reason anyway at some point that company probably figured out and they realized that they could
00:09:12.200 actually have him do a little bit of management month end reporting and something useful but you
00:09:18.240 know point is you know he he was dealing with a fixed task with a fixed demand and actually what
00:09:24.380 really got me thinking about this was there was this report in american journal of radiology
00:09:29.940 something like that um and radiology is a good example radiology is growing as a response to ai
00:09:36.220 and what's happening is that ai is making scan scans specifically the populating of reports
00:09:45.080 and interpreting the scans afterwards is making it faster and cheaper and there's a huge unmet for
00:09:51.640 for scans okay and it there's there's a real bottleneck in that there are only so many
00:09:58.540 radiologists coming through medical school i presume radiologists are they like are they
00:10:03.980 doctors i guess they are they must be doctors and maybe they do what five years of medical i don't
00:10:08.220 know somebody somebody in the comments will know but i'm guessing you do five years of medical
00:10:12.100 school first three years are fairly generic and then the last two years you know specific to your
00:10:17.840 thing and then you go off down the radiology track anyway there's just not enough of them
00:10:21.820 but the demands for scans is really high and if you can process each individual scan faster
00:10:28.560 you can get some sort of efficiency going there well the cost per scan comes down and doctors are
00:10:35.120 like oh i'm going to order more scans and not only that but scans is a bottleneck for the entire
00:10:42.580 workflow of the hospital so if if you're waiting on a scan well not much happens or maybe you can't
00:10:49.500 even get a scan because you're not considered high enough priority for a scan if there's more scans
00:10:53.960 it generates more demand for everything else like because okay we've done a scan and we've
00:10:59.900 identified there is a weird looking lump and therefore somebody somebody and somebody is
00:11:05.300 going to have to do something as a result of that and so as you make scans cheaper and faster to
00:11:11.780 deliver by combining ais with radiologists you then not only do you get more radiologists because
00:11:17.140 the cost is coming down and therefore doctors are ordering more assuming they can train them
00:11:21.160 and the robot now becomes trained as a radiologist.
00:11:23.960 It's leading to not only job increases with radiologists,
00:11:27.260 but it's leading to a job increase of everybody else
00:11:30.660 that would normally sit behind that in the production chain.
00:11:34.700 So I guess what I'm leading to here,
00:11:37.560 what I'm going to be trying to build out,
00:11:39.840 is the core idea is, look, if what you do is valuable
00:11:43.620 and AI lets you do more of it, you will do more of it.
00:11:47.900 and and this notion is this default thing that i think people instantly go to which is if ai
00:11:54.140 you know make makes doing whatever it is you do more efficient and cheaper you will do less of it
00:11:59.940 i that for me just doesn't make any bloody sense because if you're doing a thing and that thing
00:12:05.200 makes you money and now you can do more of it and you can do more of it cheaper you are going to do
00:12:11.520 more of that thing not less of that thing why on earth would you why why on earth would you do less
00:12:17.220 or remain fixed you would of course you would expand anyone in business would know you would
00:12:21.900 you would expand the moment you can do that we've not found ways to really integrate ai effectively
00:12:28.120 here at lotus eaters but i mean there must be things that we we could do i mean i don't know
00:12:33.240 automatically posting to youtube and doing clips i mean that's somebody's job at the moment so
00:12:38.680 hopefully he doesn't watch this but anyway maybe we could replace that guy or maybe we could just
00:12:42.160 maybe we could just get him post on more because at the moment he posts on what is it x and
00:12:47.860 youtube rumble but we're not touching instagram or tiktok or are we doing facebook i don't know
00:12:58.380 if we're doing facebook anyway point is that one guy could then do more and we'd presumably
00:13:03.360 generate more business and then we get on more hosts or more other button people or you know
00:13:07.540 get the idea you get the idea human intelligence is very expensive very rationed and if you can
00:13:16.180 make intelligence cheaper where you get more well everything analysis software personalized
00:13:21.880 services or you know what actually and and what i want to come to you very much is
00:13:26.740 you will get a whole bunch of businesses that had previously been considered too niche
00:13:33.620 or just unviable to offer really as a business,
00:13:38.520 but it's something that everybody would actually really like,
00:13:42.320 those businesses now become viable.
00:13:44.760 But not every job is going to survive it
00:13:46.600 because if you're doing something where the demand for that thing is limited
00:13:50.580 and especially the output is standardized,
00:13:56.040 and if AI can deliver that output directly without having to be mediated,
00:14:03.620 If there's relatively little need for human judgment or trust
00:14:07.260 or responsibility, I guess, then that very much is going to be affected.
00:14:14.000 So essentially, the episode question is,
00:14:16.100 when AI lowers the cost of intelligence,
00:14:19.760 do we use fewer humans to do the same things that we're already doing,
00:14:25.160 or do we attempt to do vastly more things?
00:14:28.400 Is it Jevons paradox all over?
00:14:30.500 is it 1865 if we get rid of the mill workers and the loom workers are we just going to have loads
00:14:37.600 of unemployed people or are those people going to go off and do something else which previously
00:14:43.860 wasn't viable because the unit cost of energy was too high so we've got to look at jobs destroyed
00:14:48.420 jobs expanded what are the common features and then see if we can use that framework to tell
00:14:54.520 there's anything useful going forward and like I say I I don't even though I love AI I don't get
00:15:01.040 to use it as much as I would like to and what I'm going to refer you to is Pete McCormack we've had
00:15:07.740 him on the show a couple of times I did his show a couple of times I think Fraz goes on there quite
00:15:12.660 a bit Pete McCormack he's got this interview thing and and he revealed an absolutely fascinating
00:15:18.420 fascinating anecdote in that he owns a football club not a big one not like a manual a local a
00:15:25.620 real bedford i think it's called and he was talking about how he got i think it was clawed
00:15:32.140 and he spent best part of a week designing and this is all himself and he's not a program he's
00:15:38.200 not a technical guy at all but just himself he sat down and and he built a whole what we call
00:15:45.180 crm package that runs the football club so i mean maybe i can get my editor to link the bit in maybe
00:15:54.060 i should point him to it but there's a fascinating bit where he's talking about it does oh everything
00:15:59.340 from tickets so so fans now turn up with a qr code and it links directly to who they are picture
00:16:06.620 profile but it's more than that it was every time he built a system where every time a goal a
00:16:15.180 goal is scored they just they just toggle it to the right player and it puts it up on the screen
00:16:21.400 goal scored comes up with a picture of the player updates the tally but also he does things like if
00:16:26.880 that player is on a goal bonus that then goes straight through to payroll so that they get it
00:16:34.140 and and before the game the manager hands over his his his team sheet and that now he does it
00:16:41.060 digitally through this system that he built by himself as a non-techie and then that also links
00:16:46.740 to payroll and it links to stats and it links to automatically posts online or basically a full
00:16:55.300 suite of basically a crn package for a false football club now full small football club would
00:17:01.020 never never normally have a software package i remember you know in in one of my previous jobs
00:17:06.460 I was working for this sort of wealth management type brokerage thing.
00:17:10.120 And we spent 4 million quid on getting a sort of end-to-end CRM package.
00:17:15.480 And that was actually based on some American firm CRM package.
00:17:19.740 So actually, the 4 million quid was integration and adaptation of their system to work in our way.
00:17:25.960 And it took us 18 months.
00:17:27.400 And there was like five of us who were in the firm.
00:17:30.140 I was one of them who got assigned to this 18-month project.
00:17:33.520 the firm that we were using they maybe had well more than five people but they had five like
00:17:39.480 another five people that were signed to this project and then we brought in a couple of
00:17:43.540 consultants maybe another three or four consultants so we had like 13 14 people plus senior management
00:17:50.840 time when that was appropriate plus four million quid to build something which i mean pete did his
00:17:59.320 version of it for his football club by himself in a week and i think he said he spent like six
00:18:07.440 was it six grand or something i can't remember how much he said but he spent a fair bit of money
00:18:11.640 well i say a fair bit of money it's not four million bloody quid is it he spent money on
00:18:16.200 running the ai at a high level high demand lots of tokens but still six grand to get a an integrated
00:18:25.140 CRM package for his football team that does everything from coffee sales to ticket entrance
00:18:31.940 to payroll to posting. Just bloody extraordinary. Absolutely extraordinary. Which then results in
00:18:39.840 a better experience. And the humans he does have, he can then do on the things that humans need to
00:18:44.160 do, which is relationship management and dealing with other people, which is ultimately where you
00:18:49.740 want them directed, not sat around doing some other bloody thing. Anyway, so the point is,
00:18:54.220 right i don't get to use ai for this job as much as i'd like so i forced myself to do some slides
00:19:00.520 for this episode and i i produced them using ai now you might think i was cheating that i was being
00:19:05.820 lazy i wasn't actually it took me a lot bloody longer than if i'd done it myself because the
00:19:10.520 the image generation thing kept failing and it kept getting it wrong and i had to iterate a
00:19:16.180 whole bunch of times and watch it fail and every time if you know every time it processes it to
00:19:20.340 like three or four minutes so i ended up playing a lot of blatro on on my on my second monitor
00:19:26.400 while i was waiting for things to complete and a couple of times i just gave up and thought i just
00:19:30.600 i'll just i just bloody do the slides myself but i didn't i stuck with it because i wanted to i
00:19:35.960 wanted to have ai produced something for an episode on ai i mean more broadly i do find i'm sorry i'm
00:19:42.420 waffling here a little bit but i i do find ai extremely helpful already on some things so i'll
00:19:48.760 i'll tell you how i use ai on brokernomics at the moment is i used to have to read a report
00:19:53.480 and then think oh that wasn't actually very useful i can't get an episode out of that and
00:19:58.340 then read another one or watch a whole bunch of podcasts now what i can do is i can is i can
00:20:05.340 i mean for example when i left work on tuesday there were like three long podcasts that i could
00:20:13.000 of listened to on the way home and what i did is i fed each of them into into ai who knows me quite
00:20:19.340 well by now and i said look um read the transcripts of these three tell me which one i'm going to be
00:20:25.320 most interested in and why and it spat that out and it turned out to be an interview with
00:20:30.000 jason yuong of nvidia and it said you'll find this one most interesting because of these points
00:20:36.240 actually put another one first but i disagreed with the points and thought yeah i'm going to
00:20:39.680 watch the jason you want one so that's really helpful so in terms of input coming in that's
00:20:44.560 really good but also whenever i encounter report now again i can do the same thing i can drop a
00:20:49.520 200 page pdf into it and say what are the key points from here and what are the things that
00:20:54.340 you think knowing me that i'd be most interested in it spits it all out and so i used i i've went
00:20:59.100 i went from reading or skim i used to skim maybe a one report a week because i just didn't have
00:21:06.160 time to do anything more than that now i can load like seven or eight in a week and drop them in
00:21:12.280 and say which of these are going to move the dial for me and it will tell me and and the bits and i
00:21:19.520 can get it to draw out and i used to then go to the report and go to the bits and read them and
00:21:24.240 think that was interesting increasingly i can just get it to produce summaries and pull out extracts
00:21:29.100 and say go and go and extract that bit and i've got various segments i think i must have used it
00:21:34.740 on Brokonomics at some point,
00:21:36.520 but I definitely used it on segments.
00:21:38.520 So I was like, yeah, okay,
00:21:39.200 dig in and find me the relevant stuff.
00:21:41.180 What it's not useful for yet
00:21:43.100 is producing an episode
00:21:45.220 because, I mean, obviously,
00:21:46.160 I'm sat in a camera
00:21:47.240 and if I tried to get it to produce a script,
00:21:50.480 it wouldn't sound like me.
00:21:51.380 So there's no point in doing that.
00:21:53.620 But I am looking for ways to simplify
00:21:55.800 and speed up the process
00:21:57.040 so that I can do more content elsewhere.
00:21:59.520 And so I insisted on doing these slides.
00:22:01.400 So we have a look.
00:22:03.280 I'm going to have to...
00:22:04.740 i don't know why we don't have a screen in here i'm gonna have to pull up a screen
00:22:07.940 right so you'll notice if you've ever because i do sometimes slides for lads out you you've
00:22:12.820 probably noticed already that it's laid out better and designed better and there's probably fewer
00:22:19.540 spelling mistakes so so there's that but anyway so oh and and i did actually i did actually put
00:22:27.220 quite a bit of work of telling it what to do and all that kind of stuff anyway so i went off and
00:22:32.180 I found these examples of jobs and firms that have been damaged by AI and firms and jobs that
00:22:40.500 have been expanded. So I loaded into it articles about customer service and white collar, and there
00:22:46.200 was a PWC report I loaded in, and I got it to produce this. So anyway, Cheggs, that's an example
00:22:52.060 of all the jobs that have been lost. Customer services, there was a clan, there was an article
00:22:59.500 on clan that well we get into it anyway free anyway let's just go on to the next slide right
00:23:04.680 so so case case study one chegg so this is a this is a company that was doing
00:23:11.320 basically teaching help so if if you had if you had a student and you wanted to give them help
00:23:20.680 you sign them up to this service and they could get answers to questions and it was a lookup
00:23:25.180 database and study aids all that kind of stuff now essentially that that entire stack has been
00:23:33.600 taken by ai instead of going on chegg and saying you know i want to learn about whatever subject
00:23:40.500 and and getting an answer which isn't particularly bespoke you could just go on ai and say well tell
00:23:47.140 me about the i don't know the bloody spanish civil war whatever it is you're working on at
00:23:51.580 the moment and it will give you an answer and it's much more adept so this chegg company so they
00:23:57.020 once ai started to get really in there may 2025 they had to cut 248 jobs which is 22 percent of
00:24:03.540 their workforce and then a short while later within the year they then cut another 388 jobs
00:24:09.660 which is 45 percent of their remaining workforce and i think they're just they're just they just
00:24:13.440 go presumably the only people still using this service are people who have just forgotten to
00:24:17.880 cancel it but you can you can replace the entire thing with AI so yeah hundreds of job losses
00:24:24.080 company off it goes don't need you anymore AI is better if you enjoyed that content and of course
00:24:30.760 you did because you are a smart person then why don't you go over to lotuseaters.com where you
00:24:36.160 can watch the whole episode for as little as five pounds a month which really is not much money at
00:24:42.360 all and you get loads of really good content.