When Code Becomes Free: The Organizational Bottleneck of the AI Age | Markus Andrezak

This talk explores the speaker's journey from early AI innovations to consulting on AI organizational transitions. He reflects on his experiences with building one of Germany's first search engines and highlights the challenges faced by businesses integrating AI technologies. The speaker emphasizes the importance of understanding and embracing AI rather than fearing it, noting common misconceptions about its impact on creativity and productivity. He critiques the often slow decision-making processes in organizations, advocating for faster, more agile responses to technological changes. The session concludes by encouraging attendees to engage with AI technologies actively to harness their potential effectively, while also considering the necessary human oversight and strategic understanding required to mitigate risks.

Yeah. So, these guys took my worst confession of the day from me already. Like, yeah, I'm from the dark side, from the business side. Everybody hates it. Uh, anyway, it's my life. What am I supposed to do? And coming from that side, um, I always look at that angle which happily you bridged over to like we build software for a purpose, you know, we we don't just

do it. Yes. It's fun. It was fun to me until 30 years ago. And um actually this conference is kind of uh closing a circle for me. So my master's thesis was actually about AI at the time. So it was a little bit of a lame project. We we did stuff like Siri at the time. So it was about um dialogue sequences that you could talk into

the computer and then he came up with dialogue acts. And one of the things we wanted to get done was like, hey, not Ziri, but the thing at the time. Um, please arrange a date with this guy over there for me around next Easter. And the issue of course and that was beginning of the '9s was like, what is next Easter? Like these were the problems we

are dealing with at the time. And we could integrate the software twice a year. And there was a guy in the European Commission who spent millions on us and he was totally proud to basically point at the file say like this is what the guys are doing but but we were totally unconfident we are doing because it was just not technically success and comparison data sets and

all that stuff. So it it was awkward at the time. So anyways, at this time uh a couple of dumb students like me sat in a room and invented the first German search engine way before Google. That was around 1994, 1995, something like that. And uh we did in Chinese because a Chinese student said like there's no search engine for ch Chinese documents out there. So we

built it and after only half year we had the idea to build it in German would be a good idea as well and it was. So basically we grew as a company from zero to 200 employees in I don't know a year or something but we were dumb students. We didn't know what we were doing. So this was my start and for into all these technologies which

are all basically linear algebra and a little bit on top and um but then as we grew we needed some people to think about the customers the clients who were using the software and it was not nerdy anymore that much but we needed [music] to take care of the business. That's how I arrived at my vocation which you Sebastian mentioned. So I look at organizations all of

the time. Um I'm a consultant. I help them to deal now mainly with how do we transition the business the whole organization into this AI thing. And the first step was always the wow moment like what's happening? What does it mean for us? How do we get people to adopt the technology? all the people who hate it in the beginning or who are amazed and don't [music]

can touch it for the first time. Um so I'm looking at what organizations are going through at the moment in that transition and um that's the last about me like I've seen four revolutions in my time and I was always hyperactive in it like internet I just talked about like what is that thing we don't know about it always open-ended and always you only learn when you're

dealing with it you just lean into it you you don't count on ROI in In the beginning, you just do the stuff to see what's possible in the beginning. Then agile came along and agile was also the thing like the the the time error was always forward like it was always like leave this old stuff behind. Do something new and it's always provocative. It's always polarizing. Always

I get feedback like the stuff you're doing is just not possible. And it always was possible. Then came mobile. Now it's AI. And when always I get this feedback sometimes really aggressive feedback on um that this is not possible that I'm dreaming or anything it was always a good time in my life actually first of all it was fun second there was always a transformation in the

society and technology which leads us forward that's happening right now again so I'm giving courses to product managers on how to use clot code for example >> [clears throat] >> And that's a total abuse of the technology of course like these machines are not that good in creating text but they're really good in understanding and extracting signal from noise. And in all the business stuff I'm doing

there's a lot of noise. So um the human exceptionalism leads to a lot of discussions we wouldn't need and it's only about a lot of noise. So when people like me sit in a room for four weeks fine-tuning the strategy for another wording another wording another wording if you send it into the machines it just doesn't matter and also on the workflow it doesn't matter the signal

stays the same that arrives at the people anyway so I I give these courses on how to abuse the technology for business purposes for setting up company operating systems helping middle management to find out what the next strategy friction is what the next meeting should be about and all these things only augmenting not replacing them. And here's what a participant said in uh week number two of

a course. So came from nothing, no education. This is not about my course. This is about the technology. It's like a drug. If they ever take it away from us, we'll all be standing here looking really stupid. And I on that weekend felt like an expert and I knew I wasn't one. this person had no clue of software development and she ended up with an app you

know whatever it was I don't know that much but this these are the emotions that are triggered in people right now which don't like I have some background engineering this person didn't ever but that's what's happening to people right it is the epidm of democratization of a technology basically enabling and also like the world is talking about this thing that um sim Willis Steve Jagger, Boris Channy.

What they're doing is like opening up these 9 10 12 U terminals, feeding them with short bursts of input. And what they say is at 11 a.m., although I already had a walk with my dog at the beach, I'm done. I'm burned out at 11:00. It's too intense. It's basically Tik Tok for developers. Doom scrolling through terminals and terminals. So there's obviously some side effects going on

with that and um these side effects lead to a stream on LinkedIn for example which I have which is about this like AI kills creativity. We all suffer from brain rot. We will not be able to think anymore. And you know you might remember stuff like I for example could remember all of the phone numbers of all of my friends before mobile phones were around. Now I

can't even remember the number of my son and and that's kind of what what people are seeing. Um the technology is only for green field. You can't use it for my business. It's too serious. You can't use it for B2B whatever. Um and um actually all the tales by all the Boris Chanies and Willis that's all fake. It doesn't work. And that's so funny because like I'm

also always divided a little bit and like is what I do all the quality I'm doing with AI all the quality code I see generated is it actually true and when I come to conference like this like yesterday evening I I shared the table with a person who who totally turned around the business totally trusts the code generation given by AI and he's totally dead certain that

this is the way to go and it is but there's people out there which are like none of this works. So the future is so unevenly distributed right now as it was never before. It's insane. It's like two totally separate bubbles. One bubble is totally storming ahead and the other one is like this doesn't exist. I don't even judge. I just observe that this is what's going

on. And uh I think as technologists it was we were always on this forward-looking time error and now there's divide like one huge part of the tech industry is like this shouldn't happen and there's societal political economical ecological reasons why this shouldn't happen probably but you know it's here but what can we do and uh here's a little thought experiment um take these what these guys are

doing is doing the impossible. That's only for green field playground. As I just said, that's not working for multiple developers in a complex system. Your site will be down a lot. Sounds great for your business, not for mine. Mission critical customers will never accept this. Not here. And now comes the fun part. So of course quality will collapse. None of these statements were according to AI or

reacting to II. It was all on continuous deployment. These are all original quotes from the time. What was it 10 12 years ago or something. The same criticism that comes up now came up at the time and I was still working at eBay at the time. We were totally like this can't be true. Like what we had we had releases one of them was was called tsunami

and then he became a tsunami. Like we had quarterly releases and how it worked was like we had a whole QA department which kind of sensed the field and say like yes this release feels stable now and then we pushed the button and then the tsunami came and stuff happened whatever and then the we all kind of reserve time from home because we knew that after a

release it's going to be horrible. We need to be in place if stuff happens if you have a P 0 incident or whatever. We just needed to stay in place. And then like I among others at eBay fought for, we need to do this stuff. It's a dream come true. It it's killing a bottleneck. We all suffer from. But we couldn't imagine how to do it. And

then we totally put the whole thing on its head and said like we do know how to do bug fixes in no time. So if we treat every release like a bug fix that could help. It it was na eve of course it was more like more stuff was involved but that was the model that drove us and in the end at that this Christmas the all

hands I sat in the back of the room and while I was heavily argued with the whole year the question was like what was the best thing we did this year and what they said [music] is like the CICD thing insane great fantastic I'm like hell And now the same thing is happening again and and like mobile the same this is too fast it's only for images

like all these phases we go through with all these technologies there's always this anger in the beginning because like we need to change how we work process need to change need to change and all that in the end it's always a big sigh of relief that a damn bottleneck is gone when a bottleneck is gone in your organization it's always great because if you're responsible for the

bottleneck you're the guy suffering ing it's gone. Um, yeah, we mastered that obviously. So, uh, nobody today would argue about CI/CD pipelines and was it work? It was a lot of work. Like we needed to understand a ton of things and and a lot of companies got it wrong like they they they got the button to push the release out there, the quality was gone. like there's

some engineering and infrastructure work in the background going on to keep up the level of quality with that like it it's it's not taken for free. It's the same thing now with AI. You need to lean in, engage with technology, learn how to do it, find the right guard rails and all that stuff. It is work, but people are showing you it can be done. It was

already done. um yeah on some of that stuff out there all the memes like I actually did the work so that you don't have to do it and look up the research so there's this myth out there that there's going to be brain rot we forget how to think and all that stuff which I just said about like my phone numbers and the one study that is

often cited actually proved the opposite not the opposite but there's no proof that brain rot is existing out there because people use AI. Um it's a bit like um [snorts] yeah what happened is that there's a a paper out there which says they did one out of four experiments which proved some kind of brain rot. The experiment was whatever you're doing start with an AI start with

a prompt. let's produce something and then let's see if a minute later we remember what the outcome was. And what they realized the number of people really remembering deeply what happened after that first initial prompting without a lot of thinking was h they can't remember that much what they did. Surprise. The fourth experiment was totally different. Like it reversed the sequence. They didn't give you a task.

They're like think about a task. start the task, then start using AI, no brain rot detected ever. So, the experiment totally changes the outcome. Basically, how the researchers reacted to all the memes existing out there, it's like, no, it's not true. We did not ever detect reliably brain rot. It's not true. Same for the um for the creativity thing. Like there's all these memes out there that

c creativity is killed by AI and all that. Same there. No research exists proving that. There's dump experiments from a couple of years ago with initial AIs where 300 people were given the same and what take this prompt work with it just one prompt let's see what the outcome is. And what they realize is if you give 300 people the same prompt, the outcome is not that

different over 300 people. Oh great, big surprise. Like if you give me any input, the same input as you, the outcome will be very similar. later on when they went into deeper research on how does create AI work with creativity, first of all they they went into the right experiment where they said like creativity is always a system. It's never a hero that has that great ureka

moment. It's all what they realize is yes the entry hurdle is lower for creativity. So you are more prone to do creative work but the ceiling isn't going lower. So the the outcome of creativity is not reduced by AI. It's just another signal. So um that is interesting that none of that research on all these negative effects exists. I don't even doubt that we have negative outcomes

as well. All technology has that. But the research that is cited is just plain wrong. And um what I'mer observing is and and this is what I hinted at before is that for the first time I think in technology we have this for a long time this separation into two bubbles. one bubble, which to me seems like very defensive and turning around the the error of time

to looking backwards and acting as if we should be protected from something which from the political, societal, economical and so on discussions and based on the fact that actually the whole technology started with a huge theft of intellectual property. I get it or you want to protect against it but the technology is here and I think that the best way to do it is to lean in

to learn what it does and if we think about econ economy even if all the open AIs and tropics will die the technology is here so it's better to deal with the technology even if the big companies are gone and um what you see from the feedback out there is that the bubble that is trying to resist in the best sense You don't see a lot of

competent criticism and feedback on what's going on with AI. What you see is, as I normally describe, you see a couple of who use chat GPT in the chat mode five times a day and prompt a little and basically go to the doctor and say like, I already have the diagnosis and compare their results to what you can do with context engineering. a really good pipeline of

the right [snorts] context at the right time into the small small tiny sub agents which then create a process which do amazing things and they cannot judge the outcome that is produced on the right side by the people leaning in and that is creating the bubbles which is interesting at least for technology we had a long discussion about this yesterday evening and over the years I want

to understand what's going on right now because I've never seen that yes for a year. Yes, for half a year or something. But then normally interest arises, but now it's kind of hardening and the the the bubbles are polarizing somehow. So that's interesting. Anyways, all of you are here because you're leaning in. So that's your workday and you seem to be interested in that and you're on

the right side. So um the right side is probably also the right side given the technology. Um, and here's my why do I think um we should engage the the I'm getting to the point that where even by abusing the technology for all the strategy work I'm doing just to give you an example I I simulate strategy workshops that I will do with my clients before I

do it on my laptop and I'm sitting there and basically the machines are talking to me and like, "Oh, this person's now seeing that and that person is discovering that friction and there's no clarity over here." And I'm like, "Yeah, but I'm seeing this." Hopefully, they will detect that. And yes, lo and behold, the agents will go into that direction as well and say like, "There's a

big gaping hole in that strategy over here." And the outcome is proper strategy documents where I struggle to get the same clarity from humans in a room and the reason is clear like it's not a judgment. People are sitting in a room to align. There's a lot of social friction with best intent. Roles need to be clarified. Politics need to be dealt with. All fine. But the

amount of time we actually the functional discussion we should have for strategy is minimal compared to what the agents do. So the agents if they discuss with me for one and a half hours they are way further ahead. All I want to say is the the results if you let these agent farms basically run around the strategy problem. The quality is insane and work like this is

not that often documented because all the people are of course into engineering for good reasons. But what I'm what I really want to say is quality of a art AI artifacts is not the problem anymore. Not at all. The problem for example with my documents is how can I explain [music] to anyone the sense making behind it? really complex strategy, really complex product documents. What is the

segment? Why is it the segment? Where's the proof in in all the statements we have from the clients collected over the last eight years in the database and all that stuff like that is hard to explain because it is so good. People just don't believe the quality. And my work is basically translating that hard factual document into human approachable sense making so that you for example could

accept that this hard fact is actually applicable to your company because it's also often very very harsh results in there like for experiments I'm doing are like um I take a strategy document from the CEO I take the reactions from middle management and I take basic big random Slack message and compare what's going on in the intent of the company, how they're executing, what's working on the

ground floor. And of course, the gap is normally insanely huge. If you present the results of these runs to the company, they're just like, that can't be us. But it is them. And me knowing them for 5 years, I'm like, I couldn't write the report better knowing them for 5 years than this shitty thing did it in 20 minutes. It's insane what these things can do. And

again that quality can only be witnessed if you have seen the process. If you sit next to me and see the agents running doing that stuff like yeah sure like that's a valid result. If you don't see the process, it's like hm. And that is a huge difference between software engineering where you have all the tests running, it's green, it's fine, could and strategy documents and more

abstract stuff like this where it's like we never worked on correctness of a PRD of a strategy document of whatever and probably it's not that far possible as it is in software engineering, but that is a huge gap which leads to less adoption of AI in that field. Anyways, here a little uh report from the field client of mine last November. I think it was 6th of

November all hands meeting closely before Christmas and he was like yeah the AI thing really great and all and every engineering job will be affected but uh the rest of the company don't bother about it. Then big U-turn a week later he wrote an email to all the employees like the new models are out there. I take everything back. I saw the quality of the new models.

Every job will be affected. I hope you're on with me on the journey. and then reactions in companies after the watershed moment where you know COOs, CEOs, CIOS they and they start to think about do we need actually a person leading the AI transformation all of that. What they observed was exactly the exactly the split I was talking about before. 50% of your developers don't care. 20%

totally lean in. 20 actually are defensive. You even see those numbers at Google, funny enough. So Google is totally mediocre about regarding AI adoption internally. Same split 2020 and something in the middle. And the defense is like I'm not not doing that stuff. I'm I'm sticking to manufacture. my code will not be touched by these animals. so let's look a little bit in in what conceptually happens

when code is free. specifically talking about code being free, not software engineering because there's more to software engineering than just producing the code. Um and basically what it means is that um the bottleneck moved and you all experienced that probably you the bottleneck you see most of the time is a PR bottleneck like how do we review all that stuff? What is a PR anyway and all

that? Um but here's also first of all the the the speed of moving. So when you look at Boris Chenny I don't know that that might be three four months ago he said like coding is solved. I don't code anymore. 100% of my code is produced, generated. A couple of weeks later, he said like, "Yeah, you know what? Actually, I'm so confident I don't rev my code

anymore." Like, you know, that's kind of a a step change. I don't give a Like, it's going to be fine. And it looks a little bit like a miracle. Like, what is he doing? Like, is he just y totally yolo hands in the air? He doesn't give a about what clo does to me like tomorrow. And sometimes it looked like it honestly, but there seems to be

more behind it. And then like in the last couple of weeks, he's like, you know, who the hell is still prompting their agents? You know, I write loops. Are you still prompting? Like that's basically and the fun part behind it is there is a lot of frigin technology behind all of this. It's it's it's not myth. It's an infrastructure engineering feedback. He always was like the the

daredevil of automization when he was working at MA. And you listen to him how he created automated spreadsheet actions that already pulled automatically buck reports from whatever systems into the Excel because he didn't have another tool. And then he pulled from the axle some things which were automated but not agents at the time. He's just now in the right window of technology where it's like oh my

god and I have kind of a a little pet boy on my side who's doing all that stuff for me. Like this is decades of thinking about automation which is paying off and probably it's what we all need to do over time and need to learn like how is the guy doing it and basically it's all infrastructure. Um, so what's happening to a company? 4,000 lines of

code are produced not in a weekend, not in a week or whatever. It's produced in 5 minutes, 20 minutes, and then have fun. You're assigned the PR. What are you going to do? Surrender, take a break, go on vacation. I don't know. And and basically, that's because building stopped being the hard part. the the irony behind all of this and and I know this might be a

little bit insulting if you produce the code. Benedict Evans a couple of weeks ago said like the industry is not now that this is getting in that bubble a little bit routine. What we realize is producing code was never the hardest task in a software company. And and the thought experience I'm having is like if I talk to any of you like when a crazy person in

your company comes to you and says like I have the following idea and this is a feature and it should go to the client. How long does it take from that first conversation of an idea to getting the idea to client and finally also billing the client getting any money any value anywhere in that chain and normal answer I get is something between 3 months and 12

months from idea to value at the customer. Now how long did a person code in those three to 12 months? probably a week. So what's happening in all the other time? What [snorts] are we doing? Sales, marketing, lead times with customers, all that stuff, waiting times, because you know it can't be done in one team, you know, all of this. Like I need that team, but they're

in a sprint already. So I can only talk to them in two weeks and then this and that. Yes, it's just a week of coding, but the week takes actually four weeks. You know, all that stuff is happening. So that's how difficult actually coding is and that why I'm saying is because I want to explain to you um how huge the impact of coding is free is

but how little is it is on the other hand funny enough and there's a lot of talk about you know we don't know what the future will be like and we don't know how rolls we that's all like we have that was 2022 202 we know a lot of what's going to happen the rest is just myth And why do we know it? Because we have stuff

like theory of constraints. We have good predictability of where the bottlenecks moved, where they will move and all that stuff. And that's basically where we are a little bit. you have the inner loop like you know get the done and uh have the code produced and anything and that's kind of bordering to the outer loop when we do um CI and then come CD and then CD

actually takes care of that getting shipped to the customer and then something and basically most of the time we're just talking about this when I talk about the six or 12 or whatever months I'm talking about the outer cycle of the whole thing which is um I'm done I'm lost um how long does it take that the market will actually adopt my and that takes weeks to

years and you only have lagging indicators and you said before the purpose is to do the right thing the right way. What is the and the signal if it was the right thing is hidden in here. We cannot know it. It's unknowable until it hits the market. is, you know, the the old Muhammad Ali thing, you know, you have a strategy until the opponent hits [music] your

face and all that. Um, and that has funny consequences if you think it from from that way. Why the inner circle was solved so easily now even with AI why it's so easy in comparison to to kind of speed up in the inner circle and outer circle is because software engineering has worked on tests and assertions since 60 years llinters type checkers compilers you know all that

stuff in in [snorts] even architectural checks and all that stuff um a lot of how and why we can trust code is because we have all these checks. You have an agent producing code. You have an agent checking the code all of the time. That's how you get from level two to level three. As a developer, you trust the thing doing the right thing because you see

it 100 times doing the right thing. And and the actual problem with trusting the code that's generally is actually seeing it 100 times working. Otherwise, you cannot create that psychological trust in it. Behind that is all the engineering work, all the infrastructure work on having these tests running. If you don't have the test running, probably you will burn your hands on the stove one of those days.

So that's why the inner circle is kind okay. Yeah, I want to skip that. That's boring. Like you know that's the difference between vibe coding and like real coding like it's also merging a little so I don't want to talk about this. So trust is actually the the task in that inner circle. So that basically also means code is free software is not because there's also a

lot of stuff going on with like are the machines that good with architecture decisions right now automations or I don't know you know a lot of people I meet are not that convinced about that as I say we know more things and people who really know a lot of things about working with shifting bottlenecks is the car industry Toyota and they have a concept it's called hjunka

and What it does is when also in car production in any production always bottlenecks are moving and what you need to do is you don't stress every step with 100%. You know that like if if every step of coding is always stressed by 100% you minimize output. It's basically a traffic jam. It's old good old flow theory behind that. Um I'm losing it again. So the the

whole imperative is smooth the flow, don't maximize it. No, it's fine. It's fine. Fine. Um, so don't keep every station 100% busy. Level the flow. And we know how to do it. It's basically if a bottleneck is too constrained, you will create the traffic jam. Which means, and how you can visualize if the autobond already is stuffed, you shouldn't use the outbound. If you are in a

hurry, you should take a bypass. actually in the southern parts of Germany and southern and and Austria they're blocking you from going off the outbound right now because you would also block all the other ways. So you need to level the flow and for that you could create something that's why you if if you keep stressing the thing the more work you have in the system the

longer it takes to do the thing. So the normal instinct we have as humans is like, "Oh my god, I'm in a rush. I push more stuff into it." If you see company that's really calm, the manager's instinct [music] is like, "Push more things into [snorts] it because people are calm." Like more must be done. People are not, I don't know, have no sense of urgency or

anything. The opposite is true. A calm organization reaches more throughput because there's no traffic jams. That's why you always have utilization at roughly 80%. That's kind of an ideal system state and that's what we need to do now. If the PRs are clogging your system, don't push more into the PRs. It it doesn't help. It doesn't help to create more stuff that just is blocking even more

PRs in your company. Have kind of a traffic light system which says like PRs are full. We don't need to create more software. It doesn't help. It's just going to block even more. It's because it's take everything I'm producing today takes even longer to get It's totally hard psychological exercise, but it's sadly rationally what you need to do. So coming back to what is the organizational bottleneck

on on the outer loop, we have the problem that we as business people never worked on correctness. What is a correct P product requirements document? There are aspects we could actually check for correctness and all of that, but we slept on it for 20 years while you were all, you know, building your llinters and checkers and whatever and all the gentic loops now which check for correctness

and all that. We didn't do We're basically and what we're saying now I'm a product manager. My community now is like, oh, now that AI is coming, the last frontier of defense for us is judgment and taste. and also some software developers like what the hell what is it and and that's coming from people like me that who for 10 15 years if a boss came to

me and said like you know Marcus I don't like your research I'd like your product to have this background I'm like what the hell that's opinion who needs your opinion and now we call it judgment taste and that's like we were asking for being datadriven all that now now that other people and the machines are data driven like oh judgment taste and that's because we never did

the work basically and why I think it's difficult like Sebastian you come to me with a p and like mine is best and I'm like no mine is better why because my judgment is better and my taste what why your taste like what's the criteria it's mine like is it hippo like highest paid income person dominates and all that again like we've been there 20 years ago.

It's but we're going there again as product people. Whyever I don't know. Um and we are really bad as product people like a person Ron Kohjavi the whole career he devoted and basically proving how bad human intuition is. If you read Nate Silva's books, the guy who predicted the first presidential thing at the time and totally changed his business from sports betting and prediction to politics because

he basically found out that experts don't know stuff. So if you look in experts predictions, they're always wrong. Like it it's like throwing a dice and experts are not better better than non-experts. It's really funny if you look at the statistics. We as product managers, that's our outcome on average. One in three features that we release have positively tested outcome. Another third has no impact, no measurable

impact whatsoever on value creation. A third has negative effect on value created. That sounds like rolling the dice to me. The excuse is intuition and taste. Well, I don't know if that should be the state of the art. And now I I want to tell you one of the the core problems where we see like how big and small the revolution with creating code is at the

same time. So when I was working at e eBay and that's how you see how old I was like eBay was not that hip anymore. The problem we had at the time was that we explained to people for 10 years that auctions are just the best. Forget about everything else. Like we came from a niche and you know longtail businesses and whatever. And then Amazon came along

and said like why shouldn't you just buy stuff rather than wait for the auction? Surprise, it worked. So we like hm why is that working? Why why are people just buying stuff if you can auction it? So the response was of course like we also want to sell stuff to people right away. We didn't have the infrastructure. We the quality of logistics and everything which of course

Amazon invested in Anyways, it took us five years for the customer to understand [music] that buying stuff at eBay, just buying stuff makes sense. Five years of explaining it to the customer. And then we had the funny idea that classified ads would also be fine, like a third option and you know, cool stuff. Another five years. My question is would more software faster have reduced to five

years? Probably not. What does that mean now? Like you're a software software driven business like eBay and creating more software doesn't help in your core problem. What the hell is going on? But but that's what it is out there. like creating just more software over time isn't the right thing. It's more going into your direction. Doing the right stuff at the right time matters much more which

is totally value conservative and boring and you know not embracing the new technology and all of that. I get it. But there is that outer loop which just what it says is customers don't absorb change as we coming from internally like it. They take time to anchor your new business idea to your name and all that. Anyways, because it sounds so boring, I want to give you

kind of a counter thing for you to play on. What I explained before with the loops actually has a chance for us. And if you look closely to what the Channies and all the guys are doing is exactly that they're using the inner loop and they decouple CI from CD. Basically, what what they're doing they're having a system where they're looking at stuff internally. you know that

as well like staging systems and what they're doing is still Boris Chenny he doesn't care at all about what he gives his machines as long as it's in his huge frame [snorts] of reference which is given to him by Armod who says build the best develop tool you can imagine and whatever Boris sees around and say like oh that could help me building that tool he just

picks it and gives it and shs it into his agents and they're doing the work and then he comes back and he's looking at the thing. It's yeah, doesn't look like it helps me kill So, he's basically storming the inner loop like crazy. And I swear to you, out of the 10 terminals he's having open at every time, he's throwing away the results of at least eight

of them. My metaphor is analog versus digital photography. same thing. technology discussion was like, "Oh my god, digital photography will kill quality, aesthetics, you know, whatever, and people will be flooded because it's so simple." Yes, like all of you are basically probably doing 200 photos of every vacation. It's insane. What you're not doing is inviting people over for an evening where you show them all of the

200 photos. Hopefully, [laughter] some might. The point is curation. Just like 20 years ago, you show all your friends 10 photos, the best ones, because you want to represent. You don't want to be the fool on the hill who shows all these stupid photos of your feet, of you know, an empty lake where I don't know like you wouldn't do it. And now you can do the

same because the inner loop is so cheap. You can play around and it totally turns around the way of thinking. Until now, we were totally overthinking what we put into the production line. Guys like me, product managers were overthinking and overthinking and coming up with priorities. Priority zero, then priority 0 a priority 0 A1 and so on just to think about like what's the most precious thing

we can give those really scarse resources the developers we have because it's so expensive the whole process. Now we can just about take everything which might make sense because we know my intuition is really terrible. I have no good intuition. Obviously giving the numbers I can basically tell developers do all of that stuff work on five features in 10 variants. Then we have a look every 20

minutes and select kill it, stop it, go on, more of this, more in this direction, go on and we postpone the judgment of what we show to the customer to the end because now we have a better glimpse. It's basically mood boards in life but better because we see the concrete outcome. So it turns everything around. Don't overthink the beginning, but overthink in the end what you

actually push to the customer. Because what's coming after is really, really expensive. It's hard for us to measure outcome from the market because all the metrics are all over the place. Even AB tests are a mess. Like how many impressions do you need to get any statistical relevant results from 20 experiments? It's insane. So that's that point. And what I want to come to is like a

totally overlooked point which is of course my work um we need to shift focus to how can AI help us run businesses because I want to get out of the intuition taste whatever loop and what I think is that that the business is held so tacid and and you know we can't touch it and is your intuation really better than mine I think that's just privilege It's

just a privileged positions which can allow themselves to kind of work this way. Any developer that comes to management like oh I don't know three weeks, six weeks. What the hell are you doing? But business like you know it stops there. Like I once built a a bank account didn't take n months took one and a half months uh years. Nobody said anything but every developer postponing

the thing for a week like what the hell is going on you know because my work is so tested so not touchable not um haptic and what Reed Hoffman I don't know like nobody needs to like him but what he said is most companies have their AI strategy backwards they focus all on engineering but actually AI is really good at business processes at the glue at keeping

things together, finding signals and noise and all of that. And you know what what companies are doing of course is like all the transcripts of all the meetings, find the strategy behind it, undocumented decisions, all that stuff. Like most of that is really super simple. Honestly, that's what I'm But what's blocking the progress of this, you know, it's it's in the transcripts and so on is the

decision cycles. the calendar is blocking the whole company. You have people deciding in glacial speed basically like yeah we treat every 3 months maybe 6 months because we need a little turn in that strategy piece and you know there's an upcoming question from that direction. Oh, you might discover offsite sailing in the evening. Cool stuff. meanwhile, you have a production machine running on steroids, option storming, as

I just described. You know, the inner loop on steroids, everything really fast. The thing is, if this production machine is running for two weeks on based on the wrong information, you're just running havoc basically as an organization. And then you wonder what the outcome is. Like your machines need to know what the strategy is. Honestly, So that's a sea level task. Anyways, here's one example. Developer asks

for 20 um API access. Takes a month for decision. Well, you know that's how you can waste money waiting for decision like this that for a month. The problem is the agents had no buffer. Like if I emit and want people to work after my strategy document, there's a human buffer. You might have a discussion with me like what the hell are you talking about? The machine

can't ask me what the hell are you talking about? But the machines need to understand that stuff. They need to have verbose direct machine readable information about the information and the decisions being made up there. Otherwise, they just go wrong. So what I say is the company needs to come into radically fast decision cycles. Management needs to meet weekly. You think that's foolish? That's what Apple did

since decades. Every Monday, top management meets for an hour, discuss all the important things. Every company I talk to is like, "Yeah, that's cool. Let's do it." Then I want to do it with them like, "Oh, not this week." discipline is not easy but that's the discipline people need and then if a decision is taken somebody needs to transfer it into infrastructure strategy MD goes straight into

the repo repo gets pulled by all the developers they have it if they need information on the strategy it's there what product am I working on who is the customer what is the purpose of this feature it all needs to be document because that's otherwise what developers need to make up on their own and then the sewave of documents which get into the infrastructure is really bad

compared to the original. So that's not what we want. So decisions need to be infrastructure basically. Well, you know and then when I discussed this on conference mostly even by technologists the thing is like oh but you can't expect these people to use git. I don't care you know basically printed his emails and haded them printed for him. Fine. If the CEO needs someone to push it

into git, who cares? But you need to have that process somehow. This stuff needs to be infrastructure. And the excuse, oh, git is so complic we mastered worse. We know how to use Jira. That's worse. We know how to use Google Analytics. That's worse than Jira is git is basically three commands. Give me a break. So a company needs to get closer to the infrastructure level or

stay slow. I'm fine. Stay slow but don't complain and then comp don't complain about the developers and all the process behind and all that. If you don't come up with the decision speed, leave me alone. I'm fine. Um what the COOs who are currently hanging over those layoff lists for all the developers and the support guys, whoever they want to lay off, what they don't realize is

they're going to be the next bottleneck because they created that glacial company OS that's not going to help. and they lay off people before they even actually realized the productivity gains in development and support. They don't even know like who here knows how to measure developer productivity like it's not easy. They don't have a baseline. They it's just a bet. It's a stupid bet to lay people

off. So last couple of minutes and that's kind of a summary of where are [snorts] we with the technology what does it psychologically mean and philosophically actually and it's actually a h highger all over the place again so he came up with that word of gnight throness in English what he basically says is none of us chose that I AI will come in the way it came

we were all handed this And if we made money off it or if we are suffering from it, it's not our merit, positive or negative. It's the thing we're thrown into. Cool. So, you didn't choose that you're now in that industry that is totally impacted. Boris Chney didn't choose that. He was hired by Entropic. In a way, it was kind of a fate as well. He was

at the right place at the right time just like I was as a stupid student with my search engine. It it was just out there. 100 people around me worked on that. It worked for me. It it was not my merit. I was dump as at the time. It's like today's weather. It's really hot. It's going to be 40° on the weekend. You can suffer from it.

It's You will not change the weather on the weekend. The question in the concept of throness or gw night is like these options were never your choice but the question is what are you going to do with it? Are you going to cave in? It's like oh my god it's terrible. I don't know what's going to happen or are you dealing dealing consciously and explicitly with the

options you were given. Um and then the task is and and Ken Beck said that already three years ago and I hope he's more competent on that than me. 90% of my skills um reduce their value to zero. 10% though went up a thousandfold and that for [snorts] all of you for society for universities for schools for employers for employees the what are the 90% of skills

that lost the value which I talk about at breakfast at lunch with my colleagues in the evenings in the beer garden or like you know coding styles dependencies of course where should it be depend where not all that stuff which defined me for 20 years which now doesn't matter that much anymore like how do I deal with that and how do I find the 10% which elevate

me above the technology like you know is it intuition taste I don't know but there's some skills which we have for example agents don't want anything as far as I know my agents start when I start them and they do what I tell them and they're totally going wild at the bound agents need us like we are steering them. We're at the driver's wheel and and that

is what we all need to do now. What are the 10% lean on engage tinker along? That's how we find out what these 10% are and how we dominate the technology and make the best out of it for society and for all the people. Even if someone hates the technology, that person should lean in totally to learn the technology to actually be able to conquer it. Everything

is just like emotionally understandable but rationally stupid. And with that, it's a real breakthrough. It's not a transformation with all strings attached as always. Open outcome, but not that much anymore. And that's the end. And with that, I wish you a happy great day. Um, more days, a whole week at this conference. and thanks for having me and uh thanks for your patience. I hope it was

a good start into your day with uh a lot of heat.