00:00:00When it comes to
00:00:02companies using AI
00:00:04We are probably here when it comes to the current hype about AI
00:00:10I'm not sure if we reached the maximum the top yet, but we definitely are at a very high point
00:00:17Now obviously, I'm kind of no exception. I create a lot of content about AI
00:00:23I have courses on AI courses on codecs cloud code and much more because it is a thing it is
00:00:31Useful, it's changing how we build software
00:00:35There is no way around it
00:00:36And I mean I have made it very very clear how I feel about
00:00:43AI for coding and that I personally had more fun
00:00:47That I enjoyed writing code before we had AI but it is what it is
00:00:52And of course nothing stopping me from still writing code by hand, but if you are doing serious
00:00:58Development work if you're building software
00:01:01Yeah, AI can make you more productive and not using it is a valid choice
00:01:07But in most cases probably not the right one if you're doing it professionally that is my opinion at least
00:01:15Now that does not change though that we are
00:01:20Really at a top point regarding hype
00:01:24regarding AI in companies and
00:01:27Maybe we'll go to even higher heights now of course when I say AI in companies. I'm fully aware that there is
00:01:36Technically not an endless amount of companies
00:01:39But there are of course many many companies all over the world and the use of AI is not evenly distributed
00:01:46And of course I'm also mainly talking about tech about software oriented companies here
00:01:52But when we talk or when we look at the use of AI in companies
00:01:58there are a couple of
00:02:00Interesting and kind of worrying trends we have to to talk about we have to look at
00:02:08one big trend we saw over the last couple of weeks or
00:02:13We heard about the last couple of weeks is for example
00:02:16token maxing
00:02:19Token maxing of course means that for some companies more AI use is better
00:02:25That's it
00:02:28there are
00:02:30reportedly companies that had or have
00:02:33internal leaderboards that track how many
00:02:37tokens their
00:02:39software developers or employees in general spend on AI and
00:02:45well, that is a bit like tracking just the lines of code you generate as a
00:02:52Measure of how good you're doing. It's a bad measure of course
00:02:57And I probably don't need to tell you that it should be pretty obvious that just spending a lot of tokens is
00:03:03Not a good measure to measure how productive you are I mean for one
00:03:09It's easy to game you can just send pointless prompts to the AI or have it do pointless
00:03:15Meaningless work and that allows you to max out your tokens, but even if you're not doing that
00:03:21Which developer is probably better the one that puts some thought?
00:03:29into a problem that
00:03:32Analyzes the result the AI gives them or suggests and that analyzes the code
00:03:38or the vibe coder that just prompts away and
00:03:42Produces tons of tokens and output for some companies. It seems to be number two for me. It's definitely number one. I
00:03:51Think the the magic thing about the AI and where you really can get a lot of use out of it
00:03:58Right now is if you do combine your expertise with the advantages of AI
00:04:05Which of course are that you can move fast that you have an infinitely patient
00:04:10Mentor where you can ask questions that you can kind of combine two brains though
00:04:16You should really trust your brain a bit more in in most situations
00:04:20Yeah, you can get advantages out of AI but token maxing is probably not it but it is a trend we
00:04:28See what we heard about in some companies though
00:04:31of course now that it got public some of these companies like meta for example already announced that they were kind of
00:04:37Getting rid of that leaderboard or of that strong incentive to max out tokens
00:04:43But it still is a thing for some companies out there as it seems another
00:04:48Trend I I saw more and more over the last week is that there are companies like
00:04:55McKinsey
00:04:56that really like to
00:04:59Push the narrative of AI agents as employees
00:05:04So the McKinsey CEO here said that they would have 60,000 employees. We're
00:05:0925,000 of them are AI agents and only around 40,000 are humans
00:05:15So AI agents as workforce is something I've also seen here and there and yeah
00:05:22Of course, that is one way of looking at it, I guess
00:05:26but then again, I
00:05:29Don't know. I mean for one right now
00:05:32We're at a point in time where most all AI agents are pretty
00:05:38Specialized regarding what they can do. Whereas humans tend to be more
00:05:44Versatile in what they can do and what they can be taught
00:05:48So I'm not sure if that comparison really makes a lot of sense. I mean who would have thought of
00:05:55calling your backup script an
00:05:58Employee that you might have had in 2018 already, right? So we had automations
00:06:03We had workflows before AI and we that makes a lot of sense
00:06:08Obviously we have automated deploy processes backup processes web scrapers data analysis processes
00:06:15We have all kinds of automations and we have them for 10 20 years or longer
00:06:20but now it's the AI agents that we call parts of the workforce and
00:06:26Don't get me wrong. You can do useful stuff with AI agents
00:06:31I'm personally not fully on board with the open claw hype because I personally
00:06:36Still am not able to get a lot of useful
00:06:41Usage out of it. I haven't found those use cases that really are magical to me when it comes to open claw
00:06:49But I am using cloud code and codecs for coding but also for more than that
00:06:55I am using the pie agent which is an open source
00:07:00Independent AI agent that you can for example use with your chat GPT subscription
00:07:05I am using that for doing all kinds of stuff on my machine or on my VPS to have it analyze log files
00:07:12Is that an employee? I'm not so sure though
00:07:16It's a useful tool for sure
00:07:19And I wouldn't even know how to count that if I have two employees and each of them is using this agent
00:07:25Are that then two additional employees or is it the agent still just one employee used by multiple other employees?
00:07:34I don't know. But yeah, this is something you also see here and there and obviously there is a lot of marketing involved here
00:07:41obviously you wanna be the company that knows how to do AI and that is on the on the
00:07:48forefront of getting the most out of AI and obviously that is a narrative that would benefit a consulting company like
00:07:55Kinsey so I get it where they're coming from. But that is yeah, it's it's it's a weird trend. I will say that
00:08:03for sure another
00:08:05Interesting trend I saw is kind of related to the token maxing. I guess it's
00:08:11Mandated use now with that. However, I don't mean that you are required to use AI as much as possible
00:08:19But you are forced to use it
00:08:23Often or in some cases also specific models or agents
00:08:29so that is something that goes hand in hand with that mandated use in in some companies out there at least and
00:08:37This is a point I do get to some degree because I do
00:08:41Understand that as a company you want to make sure that your employees
00:08:47experiment with this new technology and try to find out where it's useful because I
00:08:53Think there is a decent
00:08:55amount of people that
00:08:57Used chat GPT the free version a couple of months back or anything like that
00:09:04Or they use it here and there and it's okay, but it it doesn't seem super impressive and
00:09:10Especially amongst the normal people that are not in this tech bubble the share of people
00:09:17That are in that bucket that don't regularly use cutting-edge models is very likely
00:09:24very very high so I get why companies want to incentivize or push you to
00:09:30actively use AI to give it a try to try to use it in your day-to-day work and
00:09:35That works best. Of course if you then also get access to the more capable models
00:09:43I mean if I force you to work with an AI model, that's two years old that doesn't have any extra
00:09:48Capabilities because it's not running in any harness that would give it those capabilities
00:09:53That would be pretty meaningless
00:09:55But I do understand why companies want employees to play around with AI now needless to say some companies are
00:10:03Definitely pushing too far and there is no sense in
00:10:08forcing employees to do everything with AI and I think you should also
00:10:13Try to trust your employees when they tell you that a certain task can't be done with AI or at least
00:10:20Can't be done better with AI. But of course, I understand
00:10:24That companies can also doubt whether employees really engaged with AI and it's easy to push back
00:10:32against the new technology obviously also because many people are afraid and how wouldn't you with Dario Amodei basically
00:10:40Going on to a talk show or interview every week telling people that the vast majority
00:10:45Of white-collar workers will lose their jobs. I get it many people are afraid so it's easy to push back against AI
00:10:53But I've said it before in other episodes and I strongly believe in it
00:10:57The only way to deal with this new technology
00:11:01Just as with all new technologies in the past is to really
00:11:07Embrace it and try to get the most out of it. That does not mean that you want to blindly trust it
00:11:13That does also not mean that you want to use it for everything
00:11:16but it means that you should
00:11:19Seriously try to use it to see where it can help you that you want to push it to the limits and especially with AI
00:11:25Where everything is changing so rapidly all the time
00:11:28You really also want to reevaluate
00:11:32Regularly every few months or so because things change the models change but more importantly
00:11:38and I've said that in other episodes too, but more importantly the the harnesses the tools around these models also evolve the
00:11:46agents the agentic harnesses in which these models runs out to say those also evolve and therefore it's it's very
00:11:54Likely or it's possible that you can do something today with AI that you couldn't do a couple of months back
00:12:00so I get this but it's of course clear that some companies are taking this too far and
00:12:06Forcing employees to do something with AI that just doesn't work with AI or where you're more efficient without AI
00:12:14of course is kind of
00:12:16overshooting that target of of
00:12:18Incentivizing the workforce to use AI but yeah, this is where we are
00:12:24That is the weird state in which we are right now. And I I have not really been part of earlier
00:12:30Technological revolutions we went through in the past
00:12:33I mean sure the the Internet the mainstream adoption of the Internet happened whilst I was alive
00:12:40But I was a child back then when the Internet became a thing. So I wasn't part of the workforce
00:12:48it's very likely pretty normal for things to be rough and weird whenever such a
00:12:55Transition or evolution happens, of course with AI
00:12:58it is probably especially rough because everything is going so fast and
00:13:03Due to it going so fast and the way it works and what people like Dario tell you all the time
00:13:10It's also super frightening. So yeah
00:13:14All these things combined. It makes a lot of sense that things are very very weird
00:13:20And again, as I mentioned before
00:13:22for me
00:13:24Definitely - I had more joy in the coding part before AI was a thing. But here we are and
00:13:31yeah, definitely right now many companies are here and we'll see if we continue like this or if you find a more
00:13:40meaningful reasonable path forward where we can really try to get the most out of AI instead of just using
00:13:47the most AI
00:13:49without the use part