The Human Core of Engineering in the Age of AI

00:00:00: Welcome to Jax and we're doing the Signals podcast live today for your entertainment.

00:00:08: My name is Michael Doudin, I'm pleased to be joined here on stage by Yufa Friedrichsen and Sebastian Mayen.

00:00:26: We are bringing back an apparently very old Jax tradition that i have not been a part of where these two lengthy conversations here at the event, so Sebastian fill us in.

00:00:38: Yeah just a very moment where we were kind of thinking about how to do or to do a few live podcasts Here At The Jacks Conference I referred to this old tradition who when myself sitting together and talking About...I don't know..the world!

00:00:54: About IT And i think it's always so inspiring.

00:00:58: So I just thought, okay why don't we do this in public?

00:01:02: To share it with all of you.

00:01:03: And here we are.

00:01:04: so thanks for joining Uwe.

00:01:06: and also remember that Uwe said what the heck they want to talk about me!

00:01:11: Then when i say well...I just wanted to know what we did every time You said ok im in

00:01:19: That will be embarrassing but anyway

00:01:22: Just present us three of them.

00:01:24: Yeah,

00:01:26: Uwe maybe you introduce yourself quickly.

00:01:28: My name is Uwe Friedrichsen and I'm working in IT for about... Forty-five years meanwhile or longer and So as a profession professional IT person doing that at my for living I'm doing that for about thirty five.

00:01:45: Yes, meanwhile lots of roads and blah blah blah yada yada Yada.

00:01:49: And to make the long story short at the moment i am working as CTO at a software consulting and development company called code centric Should be enough for starras.

00:02:02: Yeah

00:02:05: So as you said that they are already pretty long in the industry I remember a friend of mine who his ex he was But happens to be the founder of essence media our company.

00:02:16: He said well my idea was To miss The next big wave in it.

00:02:23: then came Jenny and said oh no, I need to get in again.

00:02:27: Yeah, so there is this wave of it.

00:02:29: You've seen many of them over the years.

00:02:32: Are you part of this way?

00:02:35: Yes sure I mean in Part of my job i have several roles at code centric but the city all cto role basically means That I have to look ahead a little bit.

00:02:47: How could IT be in two year three years five years?

00:02:52: I'm also guessing off course But Figuring out.

00:02:57: so which kind of options do we have?

00:02:59: what?

00:02:59: Do We need to do now To make sure that, we're not ending at a dead end in two or five years?

00:03:05: and with That it's obvious.

00:03:08: That i also Have to dive into the whole gen ai agent decoding And so on.

00:03:13: movement I mean i have to understand It otherwise is very hard to Make educated guesses isn't it?

00:03:20: uh i think you you're doing or have been doing a session at this conference saying AI is a brand-beschleuniger catalyst

00:03:37: fire accelerant.

00:03:40: okay what do we mean by that?

00:03:43: basically It's not only me who saying that I also picked it up and it would totally resonated with Me, and i just read again in the dora report from last year so this state of devops report which then turned into The Dora Report Which i think google releases every Year And They Looked at AI and how AI influences engineering organizations and what they figured out, but also what I'm seeing a lot is if you have crappy engineering practices.

00:04:19: They amplify them.

00:04:21: And If You Have Good Engineering Practices?

00:04:24: They Also Amplify Them.

00:04:26: So it Makes It A Lot More Visible Which Going good or not so good in your organization.

00:04:32: and it's important always for me to say Be careful, it's not only about the software development department.

00:04:38: It is also about the whole engineering that starts at management level and business-department levels goes into operations in everything.

00:04:45: so its not limited to here because stories or narratives we hear are going to solve software development careful because software development is so much driven from things outside of software development that I'm not sure if AI will simply solve that magically.

00:05:12: So we had this morning in the keynote, i tried to express a certain idea of enterprise AI.

00:05:22: then also Russ Miles added his view on what it means to talk about AI literacy.

00:05:30: So, I got a feeling when i talked with many people that for me it feels kind of contradictory When we say hey!

00:05:39: For doing AI you must practice your engineering Even harder even more consequent.

00:05:47: So things from the past Yeah, which are not considered to be the things of the future must Be done even better than in the past In order to handle the ultra new thing Which is AI?

00:05:59: Mm-hmm

00:06:01: Would you agree with that?

00:06:02: because I feel like okay sometimes for people it's this kind of a mental You know contradiction or something on that.

00:06:08: It isn't contradiction actually and i mean The bigger contradiction Is We have that narratives, which the tech industry and influencers are telling us.

00:06:19: So forget everything you've ever done.

00:06:22: AI is going to do it for you now And then people inside of community who tell your opposite basically saying No!

00:06:31: You don't need to forget anything we're running implicitly they aren't talking about.

00:06:36: We are running on bleeding edge technology, which can give us great results.

00:06:41: Which gives us crappy result depending how you use it.

00:06:45: and in order to get or increase the likelihood of getting great results You have to focus even more on your engineering practices although a good engineer practice as we've learned before.

00:06:57: for me The hard part is that contradictory stories and that you have believers over here, I'm not going to judge who's right at the moment.

00:07:12: And this is more like a religious war than actually something.

00:07:17: well-grounded discussion which will give us good consensus but two parties beating each other nobody ever moving.

00:07:29: It's important for us in having these conversations to help people reframe the discussion, because right now a lot of the discussion is I don't know Mac versus PC when we really want it to be one.

00:07:41: where were constructively trying to help iterate on what as solution?

00:07:47: What was good way thinking about things.

00:07:52: That question so good that Let me try to start from a different angle.

00:08:02: I'm also working with customers will say yeah, we want to introduce AI into our engineering departments because We want to speed them up.

00:08:10: more throughput and so on.

00:08:12: And first thing.

00:08:13: I asked him is okay What's your goal?

00:08:15: More throughput.

00:08:16: why more throughput?

00:08:17: right?

00:08:19: don't you really wanna have more value?

00:08:21: business value i mean If you just build crap faster that doesn't make any difference at the market or even a noise.

00:08:28: your customers What's the point of bad?

00:08:31: And most often I get them thinking at that moment.

00:08:36: I'm not always successful, but more often than not I've success for okay Let's figure out what's your goal.

00:08:42: ok i want to become better at Creating business value or outcome or impact or however You want to frame that and I mean product Development community who talks a lot about outcome and impact.

00:08:56: And that's actually what we want to achieve.

00:08:58: so and say, okay Let's take that as an all star.

00:09:03: how do I getting there?

00:09:04: We have to convince our product managers.

00:09:06: the problem is we have a long-standing tradition in companies that Let's blame the developers because they're too slow.

00:09:15: They are to expensive and so on, etc.

00:09:18: And therefore it's there for that we are not making tons more of money and It's not in.

00:09:26: The problem is that about sixty two ninety percent Of all requirements were implementing.

00:09:31: Do not create any business value?

00:09:33: There just sitting working in the systems Making them or complicated a building basically business debt.

00:09:42: not technical depth but business depth, which is even worse than technically that because you cannot easily identify it as a developer.

00:09:50: Because in there everybody tells us we need to have that while we're not needing to have the ender and so... We are sitting here trying convince our product manager Work more with Hypothesis.

00:10:08: so is measuring actually the outcome The impact?

00:10:11: measuring if their users are picking up these ideas how to respond to that?

00:10:16: That we're not building big epics was one million dollars put on that or euros where?

00:10:21: Why were here in Europe, but basically same story.

00:10:26: But then we Split them up into a series of smaller questions and iterate over them And so on.

00:10:32: The problem is they won't listen say Hey, you're crappy become faster.

00:10:37: And so I say let's start with doing our homework automating the things and becoming better in releasing stuff.

00:10:48: um The automated tested and all this thing.

00:10:51: quick

00:10:51: question when you saying during our home work You mean engineering

00:10:55: engineer software engineer right?

00:10:56: Yeah absolutely

00:10:57: talking about the whole organization yeah.

00:10:59: But your

00:10:59: saying now

00:11:01: hour is yes it's a software engineering.

00:11:03: So that Let's take a simple benchmark that if you get it small requirement, then your able to implement and release that within the same day on demand.

00:11:15: So You don't always have to release several times today but you must be able confidentially To do that.

00:11:24: And then the second thing we have to build in is observability which does not stop at the application level But moves up Business level saying okay, we are going to measure how often features I used How much more revenue goes through the system and all these metrics?

00:11:42: And then We're able to do that.

00:11:44: We have two metrics and then we can have a silent conversation with our product manager say Here, coffee one-on-one.

00:11:52: I can make you hero or the opposite?

00:11:55: And here are the numbers!

00:11:57: The last fifty requirements that we implemented for you and how much they're picked up.

00:12:03: so how much it resonated with your customers.

00:12:09: We will also assist to have everything in place To become a Hero to boost your impact into the company product manager of the company and All we can share these number with everyone.

00:12:25: So okay only using the last part of this sentence if you really have to but so creating a win-win Contra situation for the Product Manager, and then do it.

00:12:37: And here comes AI into play because I can help us With parts off the automation.

00:12:44: It can helps us

00:12:45: too

00:12:47: free the mind of the product managers going back from always doing this, how Constantine yesterday called that transactional work.

00:12:57: So making writing down stuff or analyzing some user surveys and so on and blah-blah-blancet really having.

00:13:07: okay How can I phrase these ideas?

00:13:10: That are you have in ways create a series of hypotheses, and what do I want to measure with that?

00:13:18: To figure out if you're on the right track.

00:13:21: And then I can also use AI because now producing code becomes unbelievably cheap in some way.

00:13:32: Let's not discuss quality exploration.

00:13:34: Exploration is different story than expansion or extraction.

00:13:39: basically If we take this three X model from Kent Beck very quickly iterate and Release code to figure out are we on the right track?

00:13:50: Are you on the ride track?

00:13:51: of course, We have to pay the debt later on And rework that but even their AI camera.

00:13:57: But we have a completely different conversation now how we can introduce AI How we can use it how we Can't use it too leverage business value and so on.

00:14:06: I'm not just oh Developers are too slow.

00:14:09: let's make them faster all right.

00:14:11: yeah one of the important things is he said is the concept that you can have a metric, and not necessarily always feel like you have to maximize it.

00:14:22: You talked about... It's important we be able to release multiple times per day?

00:14:30: And yet I've seen organizations say hey!

00:14:33: We released two or three times every single day but if they're not doing so your feeling as development organization

00:14:39: right?!

00:14:41: So how do you recommend That we take some of these metrics, the enablements that get out of AI and avoid pitfall.

00:14:50: Assuming you must do this in high velocity?

00:14:59: I remember Some other colleague here from community who once said Metrics are crucial And they're also our worst enemies, basically.

00:15:17: Because the key point is we want to achieve some kind of quality but quantity isn't a quantity so it's something subjective and I think per definition not measurable in itself.

00:15:36: So what do you have to create?

00:15:40: Proxy metrics, which gives us an indication if we're on the right track.

00:15:45: They are very helpful!

00:15:47: The problem with these proxy metrics then is that they become and end in itself over time so we optimize for the metric For deployments per day or something like this.

00:15:59: If you take the Dora matrix I would say look at your context uncertain environment.

00:16:11: Extremely, maybe in a winner takes all environments where only if I'm the fastest to lead innovation and development of this domain that i will be sole survivor there.

00:16:26: then it's extremely important as many questions are possible at customers who get some kind feedback from them.

00:16:33: If I'm in a highly mature market where maybe cost efficiency is more important than releasing several times per day, then... ...I don't really need to do that.

00:16:47: So always take the context of what you're working into account and make the same decision at least what my typical recommendation is.

00:17:01: You were just touching a topic, I remember we were discussing something similar last year or two years ago.

00:17:08: We are talking about the so-called software crisis.

00:17:12: you were touching this.

00:17:13: So engineering has always been blamed for...we could have had much better performance and made more money if only software engineering would be faster.

00:17:27: Yeah, so bottleneck clearly on the software engineering side being said You were just saying hey finally we managed this software crisis.

00:17:40: This was kind of pre-genii.

00:17:42: yeah in a sense Of Hey with agile With lots of ideas In the sense of automation probably with lots of Ideas which are coming from also the DevOps world and it's whole bundle of things Which we developed Partly pure tech and infrastructure, but also partly let's say mindset people culture process communication these kind of things right.

00:18:05: so with this bundle off measures We came to a level where one could say or I remember your words.

00:18:11: We are kind of on the level now.

00:18:14: Yeah And that's how i understood you words okay?

00:18:18: Now with jnai what happens?

00:18:20: no

00:18:24: i Think jnaI doesn't change them The story, I mean we know how We could be on top of that demand crisis and there are some companies out There That Are But the average company is not there.

00:18:45: I Mean big part Of it Is?

00:18:47: That's They're Still Efficiency Obsessed And Haven't Understood Software.

00:18:53: I mean, as long as i'm still going to confuse software with a physical product like building a car or house.

00:19:00: I am trying to optimize the production cost.

00:19:02: The problem of software is which Maya Liman already said in nineteen eighty so it's only forty five years ago or forty six meanwhile.

00:19:12: So yeah we have forgotten that at least ten times in IT.

00:19:17: but let us assume you will know that software always needs be changed to preserve its value.

00:19:26: And so it's a different story, I mean if young lad full of testosterone and try show off build nice sports car maybe Find so being yet in my place then find a nice girl marry her and Then we start the family, and I will sell.

00:19:47: My race car and Nisha The barcats are.

00:19:55: thanks a lot.

00:19:57: The world was just gone, thanks a lot.

00:19:59: So the children still take attached to their bike heads in the ship and something like that.

00:20:02: I mean That's what we're doing at software every day And therefore production costs is ten percent of the whole story.

00:20:09: But people not understanding software are going there.

00:20:13: Only fix Satan on production cost.

00:20:15: Then trying to minimize production costs by all means because they Effectiveness, which is very different from efficiency because effectiveness.

00:20:25: Is doing the right thing.

00:20:27: so doing what creates value impact outcome you name it.

00:20:31: and Efficiency is Doing something something defined not knowing What?

00:20:37: The value of that is at least cost.

00:20:41: And we just a completely different story.

00:20:43: both have values So you have to look at those off them but most companies only look up one on them.

00:20:47: and then Yeah, the result is suboptimal decisions from management.

00:20:52: From business departments and so on resulting in that we still haven't solved a software crisis but not because you couldn't do it...

00:21:04: How many companies hang onto some cost fallacy where they've invested an approach?

00:21:11: And like you say, they've invested in the sports car.

00:21:14: So they're going to turn that sports car into an airplane because they have already invested it when it's clearly not a rational thing to do.

00:21:20: OK, some cost fellows here strong enough!

00:21:25: That is absolutely... It gives us another rabbit hole.

00:21:34: Not sure if I really want to dive into that one?

00:21:39: But but you have to do right.

00:21:40: so some cost policy is strong.

00:21:42: So we have invested that much money, so we still have two Double down on this approach and all these things.

00:21:51: That the same thing happens not just with products We've built in Things we've created In our products, but it also happened sometimes in Our approach in our mentality?

00:22:02: We've started Down a road.

00:22:04: How would You get organizations To stop?

00:22:08: You mentioned we could fix this.

00:22:11: How do you get organizations to stop in their tracks and say, okay I'm gonna stop doing things the wrong way?

00:22:18: And think about how to fix it whether We're using new tools to that or even old ones?

00:22:26: Full disclaimer i am a terrible change agent!

00:22:33: I don't have patience to deal with persons who come the tenth time and say, yes but.

00:22:39: And Germans are perfect in saying Yes But!

00:22:43: We're development from the Homo sapiens into the HOMO.

00:22:47: YES BUT as Germans in a nutshell usually I still try... The only thing that i am able do is Here's a wall which you're going to hit and you feel that.

00:23:06: You're hitting your wall.

00:23:07: I show, you adore i tell?

00:23:08: Your what behind the door.

00:23:10: i also explain To you why it makes sense to go through That door.

00:23:14: so i'm trying to Create picture an image What we could achieve what could be first steps And so on.

00:23:21: but i also say very clearly to The people i give a dot-dot-dot if you Go Through that door thats your decision in the end.

00:23:33: and if I'm not going to push you through that, If You are standing like That before The door And Trying To resist to go Through That Door.

00:23:41: Let's again your Decision.

00:23:42: but the next Problem We're Having In Organizations Is?

00:23:46: It's Not Only People it'S the Organization.

00:23:49: the organization is a Higher-Level System Which Basically has its own Mind its own memory.

00:23:57: It's called company culture.

00:23:58: basically is the things that happen between The people how they interact and which patterns had been successful in the past.

00:24:05: Which panel has not been successful?

00:24:06: In the past where you've been punished for as an organization or not?

00:24:10: And all these things shape How it responds, and it even harder to change this thing.

00:24:16: so Basically from the way My personal approach is saying okay we try to get there what is the reasonable idea?

00:24:28: and then, What would be reasonable first steps?

00:24:32: And usually they're not so much tool fixated or something like that.

00:24:35: I mean if you are talking about technical tools.

00:24:39: So i don't think that gyra will solve an agile introduction.

00:24:44: Fool with a tool is still a fool.

00:24:47: There's some wisdom in it.

00:24:52: Another topic I wanted to talk to you is an observation, I made.

00:24:59: And I'm happy if your can comment on this.

00:25:02: talking again about GenEI.

00:25:05: in the software engineering world we know there are let's say at least two different angles one as GenEi being a part of the software-engineering process and The other thing is GenE iBing Part Of The Product Being Featured Of Something Which Does something for the Customer For The End User.

00:25:21: And what one can learn over time is of course, in the end now we got to think more or less.

00:25:30: What does this Jenny I thing doing?

00:25:32: It's doing...what our LLMs are doing.

00:25:35: and We finally need to look at what we humans are doing with respect to This new quality of an agent which does something in our system.

00:25:50: And when we look at on us humans, especially in let's say into software creation process then We got a party which is traditionally well the experts in talking about exactly that.

00:26:02: It's the agile folks.

00:26:05: So I'm really convinced that Agile really changed the game pretty deeply Although we say, ah.

00:26:12: We got a lot of how do you say this cargo cult?

00:26:14: Yeah So kind off.

00:26:16: we just imitate Agile rituals and we're not really doing it.

00:26:20: but in the end I think we both would agree that those principles Those values those habits those experience continue to be continued to be valid more or less.

00:26:32: My question now is i would expect The agile community before front To talk to us about how to deal with this thing called Genii.

00:26:42: But,

00:26:43: and I don't know if Wolfgang is here we also had this conversation... He was he used to be here?

00:26:48: I said why the hell!

00:26:50: I expect you agile people to be the expert to lead that conversations because You were leading This conversation About what it means to Be a human being in this production environment.

00:27:02: Why Don't We see them?

00:27:04: or Is It only me who does not See Them?

00:27:09: Okay, my personal take on that one.

00:27:14: First of all I doubt we really have understood what LLMs are because we're still confusing them way too much with humans.

00:27:23: and LLM's may have some kind of intelligence...I'm not an expert in rating intelligence for people or things.

00:27:32: there is a lot better out here than i am but If they have some kind of intelligence, it's very different from human intelligence.

00:27:42: Also the behavior is different.

00:27:44: so even if... I don't mean that adjuratively but if they imitate how to speak like humans and i think its a great thing human machine interface more towards how humans interact than that we have to Interact with machines.

00:28:06: More on the Machine level and can interact with the machines more honor.

00:28:10: Human like level like be interacting with each other.

00:28:13: That's a great thing basically, but I still think we haven't figured it out about the HR community.

00:28:20: Does it still exist?

00:28:21: I mean Let's be little bit mean.

00:28:30: agile had its time as a term, basically.

00:28:36: And most Agile transitions failed Basically in terms that if you go to organizations They move for a while To scrumming doing all these cargo calls and then they finally say hey great.

00:28:49: There's safe We can't go back to what we had before because save allows us to do whatever we did Before.

00:28:55: but we have this fancy new names for the roles basically, but we still Levels of indirection between someone's had an idea until it comes down to the developer switch are nicely in a safe diagram put Down left in the corner.

00:29:10: So yeah, let's put them back into the cellar.

00:29:12: this Developers our new to King makers and lets get rid off that fat and so on.

00:29:17: And so basically The mindset is still alive, but the people moved into different directions.

00:29:26: They move it to product development they moved in two dev ops they moved Into other ideas and so on.

00:29:33: So we have we still have that?

00:29:35: And if you go back into the original idea I mean for me agility is inspect an adept period.

00:29:43: That's it.

00:29:44: i mean agility.

00:29:46: Is this re So rediscovery of if we have some level of uncertainty, which you have to deal with.

00:29:58: If you have some level of complexity out there that we cannot perfectly predict maybe it's a good idea not to Have a big plan and run through that And then be surprised at the result does not fit with reality anymore.

00:30:11: But let me care go their step wise.

00:30:14: an half opportunities to adjust.

00:30:17: I mean, that's basically agility in a nutshell Yeah.

00:30:20: And everything else is ceremony or ideas how to implement that and so on.

00:30:26: I think this idea is still alive, but it's.

00:30:30: There are nobody out there or very few people out.

00:30:33: They're who will say we Are agile advocates?

00:30:36: I mean you don't make living anymore.

00:30:38: everybody's age and every company is a jail.

00:30:40: I mean they're not Agile But they of course their Agile.

00:30:44: You can go to the next bus stop on fire asked two people are you Agile?

00:30:47: Of course i'm Agile And so It's over.

00:30:50: basically that the term was over.

00:30:52: yeah Not The Idea Yeah, and then I Think It's just my guess.

00:30:57: That may be one of the reasons why you don't hear a lot from some agile community anymore about this idea Of jen AI and how to adopt their ideas or Jenny I to use that To speed up product development, even here a lot more from the product Community?

00:31:16: Yeah deeply involved in that one.

00:31:18: so So what are the things that I unfortunately do see have been seeing for years with agile rollouts of organizations, is that their misguided implementation in Agile often drives exactly what you referred to earlier on our conversation about how we're building the wrong requirements and that's a big part why software teams are failing.

00:31:43: And so organizations... You've seen memes and jokes for years but agile doesn't just mean build more things faster with less resources.

00:31:54: lot of organizations were using agile to put the burden on developers

00:31:58: for

00:32:00: What it means?

00:32:00: To design and build a product.

00:32:02: they're skipping design phases.

00:32:04: They're just skipping use UX research.

00:32:06: There's skipping all kinds of steps And just trying to put more and more burdened on a developer as part Of the lifecycle building A product, then that doesn't work.

00:32:16: We know that does not work.

00:32:17: you alluded to that earlier.

00:32:19: How do we avoid that same pitfall with the next mentality model that's going to be developing around what it means culturally, a software development team.

00:32:34: To be product-development and incorporate AI into whatever the thing is?

00:32:41: because I think you mentioned the language shift, we're speaking more at a human level now than in machine-level.

00:32:50: The downside of that is it makes it easier for us to make mistakes.

00:32:56: by assuming when we talk to another human... We assume they are going understand context and provide their own intelligence behind things.

00:33:06: And if that's not happening behind the scenes we're open ourselves up to maybe reproducing The same types of mistakes.

00:33:16: We made a vagile.

00:33:17: Do you see kind of?

00:33:19: A path forward for this?

00:33:20: or I have us recommendation on how we navigate This whole paradigm shift?

00:33:25: hmm Give me a few billion dollars to influence the market and maybe The short and harsh answer is it will happen Period.

00:33:39: and the point is What we see, and we haven't seen that in this extreme form before I mean.

00:33:47: We always seem that investors have put quite large bets on technology.

00:33:54: It was a quiet right lucrative business.

00:33:57: basically to bet on Technology.

00:33:59: what will usually save bad especially when The interest rates were low.

00:34:03: it was a good idea too but on technology And therefore we had all these golden years between say And so into two thousand twenty three basically.

00:34:15: But what we see in the for the first time is that?

00:34:18: Basically all companies and All venture capitalists have put all their bets on a single topic, and they're all In some way desperate to win that bet.

00:34:31: Yeah they're absolutely desperate to win that bet, because the

00:34:36: winner will take it all as you just said.

00:34:38: Yeah!

00:34:39: It's a winner-take-it-all big game and there are only few parties who can survive this game.

00:34:47: so if we would have some kind of what Simon Willison calls the challenger moment for AI which is huge failure state something, but maybe everybody would lose of them.

00:35:06: So like the people who bet on blockchain too much and so on But yeah And therefore in their narratives are shaped around Use it don't think use it.

00:35:20: It will solve all your problems.

00:35:21: and they're using tons of money Billions and billions of dollars to send this message.

00:35:29: I mean If a Dario Amodai is out there and says with some fake tears, oh I'm so concerned that AI may take away a lot of white collar jobs.

00:35:42: Basically this message he sent out was like hey investors listen we have some opportunity for investment from you.

00:35:49: We are having the... how's it called again?

00:35:56: Going live on store or stock exchange the

00:36:00: IPO.

00:36:01: I feel right thanks We are doing IPO next year.

00:36:06: Do not forget to invest in us and let's get as high evaluation as possible.

00:36:11: It's okay that Dario is saying that because it's his job of seeing there.

00:36:15: The problem, Is?

00:36:16: That he Influences a lot of people who make decisions with that.

00:36:22: and then we still have the Jenny I thing.

00:36:24: And then Clark's third law really strikes any far advanced technology or Advanced technology.

00:36:35: That is at once far enough.

00:36:36: it's indistinguishable from magic, and for most people in the world Gen AI is magic.

00:36:43: They have no clue how it's working.

00:36:46: That's and that's more than ninety nine point nine nine percent of the humanity with No clue, then we have this oh boy a few people Have some idea about that?

00:36:56: And Then you have very very few People who really understand How it does see how It works in.

00:37:01: therefore I'm afraid So We will go down The road.

00:37:06: the only thing but i think at You an eye and you can do is that we try in our very tiny bubble of influence, the idea say hey good old practices of software development are still valid.

00:37:27: It's still about creating value for customers.

00:37:31: let's figure out how to Automation still has a value.

00:37:37: Let's figure out to do that in reliable.

00:37:40: and also let's figureout two things or less Distinguished two things.

00:37:45: Software, and Jenny I are different tools.

00:37:48: they have different characteristics.

00:37:50: this one software is Deterministic.

00:37:55: sometimes it's little bit stupid.

00:37:57: so if we ask software to do something which Is not in its code?

00:38:01: It gives us Reliably very stupid answers if it doesn't crash.

00:38:09: This Jenny I technology is very different.

00:38:12: It's um, it Doesn't crash.

00:38:14: it always gives you the answer.

00:38:15: its surprisingly often gives us good answers But it's not it's non-deterministic.

00:38:21: So there are chances that it goes wrong and they can also go terribly wrong in its Suggestions?

00:38:29: They're getting better.

00:38:30: but as still this probability in here.

00:38:33: Now, let's go back to the problems that we're trying to solve.

00:38:36: There are problems out there where it is perfectly fine if they were going wrong once in a time but therefore providing answers all of the times even though sometimes and some other situations like maybe depositing money or withdrawing money and so on, I don't want to have some probability if you go through the teller that i get some money or not.

00:39:00: Um...I wanna get the money!

00:39:01: And um..if there's money in my bank account.

00:39:04: ,i wanna make sure that ..i'll get money from the banker.

00:39:07: ...from the tellers .

00:39:07: So maybe....the tether machine may be a little bit stupid If I ask it something outside of depositing or withdraw ingmoney?

00:39:15: That is perfectly fine for me but as long as its completely deterministically reliable does his job.

00:39:22: So maybe that's the new thing.

00:39:24: we have a tool to work with and thats fascinating actually.

00:39:28: I absolutely agree with everything you said, my concern is that the teller example or bank example isn't good one.

00:39:40: And here why?

00:39:42: because i feel like alot of folks for the same reason you alluded to earlier, where they don't necessarily understand how that technology works.

00:39:50: They might think wow it's okay if its not perfect because I want get more money than actually have when i go to the cash machine right?

00:39:58: And so... Because we live in a world were sometimes business value decisions are driven by specifically people who do not really understand the inner workings of this technology.

00:40:13: What kind of responsibility do we have as engineers who do understand this to provide guardrails against that sort of mistaken understanding?

00:40:29: I think... We all, starting with people call themselves architects but also over everyone calls themself engineer.

00:40:47: trade-offs clear, so if I mean it's still the old straight of thing.

00:40:52: I mean architectural work in a nutshell.

00:40:56: So figuring out which options do we have making clear?

00:40:59: The trade awesome help the people to make the best decisions possible and that's basically what I call architecture work at a nutshell.

00:41:06: And this i think We Have This Responsibility Very often when I work with customers, no matter if it's a strategic or technical assignment Or anything like that in the end.

00:41:19: I'm always doing the same thing.

00:41:20: II am showing options and The trade-offs have to people understand What these trade-off are?

00:41:30: Enable them to make good decisions.

00:41:33: I mean, it's about enablement.

00:41:34: We cannot force people into making the right decisions at least as we are Service suppliers in some way if we're not the decision makers and very often we are Not a decision maker.

00:41:45: but we can help People making best decisions possible.

00:41:49: And

00:41:50: It's Often it's not that people Making subpar Decisions or even bad decisions because they're evil people Few of them are but most of them not But because they didn't have all the information that they needed to have in order To make the best possible decision.

00:42:12: and therefore I absolutely agree with you see We need we have this responsibility?

00:42:19: to share This understanding that we might have what other people might not have again We cannot expect everyone to do what we recommend them, but at least... ...we did our share.

00:42:34: To help them know what it means making a decision which options they have What their pro ups and downsides are.

00:42:44: And then in the end I'm as consultant from now on its your job.

00:42:55: That's one of the things I love about consulting and i think it is an advantage that we have being in a position where when were meeting with business or client, decision maker.

00:43:06: We can present them facts to allow them choose.

00:43:10: at this point if we are going to proceed working with them potentially there will be options for us.

00:43:23: Ask them good questions to try to inform.

00:43:27: Them on things.

00:43:27: they expect that from us as a consultant.

00:43:30: Do you have any advice for somebody?

00:43:32: That is in an enterprise situation where there may sit maybe more of uh Here's the thing.

00:43:38: do go, do it type of role and that's let me be a little bit harder But this still kind of want to make these differences.

00:43:47: ask your people.

00:43:48: They know the answers.

00:43:50: I mean very often.

00:43:52: i came two companies And I talked, and some decision-makers said yeah we have to solve this problem x y or z Or something like that.

00:44:04: We had no idea how To do it!

00:44:05: I asked talk to a few people in there.

00:44:09: They exactly knew How to solve It.

00:44:11: they always know how to solve it.

00:44:14: you only Have to listen.

00:44:17: So usually very often my job was simply to write down whatever.

00:44:22: People new bring that knowledge together and share it.

00:44:28: Wow, you must be brilliant to come up with ideas!

00:44:31: Sorry folks I'm not brilliant.

00:44:33: i just wrote down what you told me and put into a consistent form.

00:44:39: so this distributed knowledge which is in there maybe we need more People in companies, but this division of labor idea that's still in the minds of many people Makes it hard for them which brings together This scattered knowledge and bring that into a consistent form something like That.

00:44:59: so there is advisory to the CTO or anything?

00:45:02: Like that The person who listens to the other people Who brings these together and talks with the people and so on saying then okay here We know everything, we don't need to hire Uwe for that.

00:45:15: We can do it ourselves.

00:45:17: I mean bad from my business but good for companies.

00:45:22: After having listening to you and now i'm going to create an agent called uve

00:45:26: Which

00:45:27: behaves exactly like you which does the kind of listening and synthesizing And giving feedback in a similar same way as your just described here

00:45:36: only if you can design that agent to be barefoot.

00:45:42: Otherwise, it's not over!

00:45:46: The reason why I'm asking this question is because we're making a joke.

00:45:49: so then again still what the difference?

00:45:53: You just described as hey i am here and listening yeah... I'm kind of mirroring whatever guys are doing anyway which I totally get.

00:46:03: Maybe I does the same thing.

00:46:05: So what is that?

00:46:06: What is a little extra?

00:46:07: you are bringing in as so-called human being

00:46:11: A different kind of context probably.

00:46:16: So having seen a lot of things I'm not always wrong, right sorry also not over it's wrong.

00:46:25: but Having seen a load of things in the industry about many different domains Gives me A good intuition, whatever that means to say okay here's an important bit of information.

00:46:39: Here is not so important a bit of info and I think That Good consultants are often in this way no matter if they're inside the company or outside The Company.

00:46:54: So And i Think thats this individual advantage that i as a person could offer.

00:47:04: What I see quite often and companies is that they put together some information, but there have a hard time to distinguish Chef from wheat basically.

00:47:14: So what's the important bits in which?

00:47:16: not maybe?

00:47:17: The difference that That I can sometimes make is say okay Here are the important things of all the information that I gathered from you And here other things that you can ignore at the moment.

00:47:30: Basically I think that this kind of understanding and also importantly, understanding organizations software everything is about humans in the end.

00:47:43: Even if people are trying hard to pretend it's not... It's about humans!

00:47:51: And LLMs do not understand humans because again they may be smart But they're not human and cannot understand humans.

00:48:04: They can be very, very useful but understanding human needs demands desires emotions.

00:48:14: most likely you still need a human for that one at least my personal Guess unless we see something very different from the AI sector, but this will not be LLMs.

00:48:26: You could give of course a lot of context again About what does product quality mean for me?

00:48:33: What us good user experience means?

00:48:34: yeah For me you could give lots of theories right about that kind of context.

00:48:40: But still it's you to decide To listen.

00:48:44: on which theory for example.

00:48:45: Yeah absolutely and I mean there They're great for research, for instance.

00:48:52: Saying okay let's bring all that information together.

00:48:55: I mean sometimes you have to cross check.

00:48:57: they brought the right information but still a great tooling But decision making it is not like humans always make good decisions.

00:49:09: we know a lot of counter examples about this one.

00:49:12: Still i think these kind of decisions makes sense in there that make these decisions.

00:49:22: Just to wrap it up a little bit from my side, so one thing is also always in my mind because we were discussing internally maybe... Maybe?

00:49:34: A good kind of slogan for what could mean being an engineer in this AI driven world Could be something like Maximize value For your customer whatever By using the least numbers of tokens.

00:49:55: Actually, I don't want to care about tokens.

00:49:59: Well but environment cares for tokens and somebody else will pay For them.

00:50:04: The point is As long as we need To take care About tokens Basically We have a problem with that technology in some way Because it we limit the access to it.

00:50:26: So, this whole narrative breaks apart that tech companies are trying to sell go all in with AI and use as much possible so on.

00:50:37: if they're forcing us into a kind of usage behavior but its so inhibitory expensive You can't use it as you want to.

00:50:55: so do not basically with the pricing.

00:50:58: then they would still tell The story don't go all in.

00:51:01: So and that's the point I'm saying, okay if i really have To think about how to leverage AI?

00:51:10: if possible, I do not want to think about tokens and from the experiments that we made with for instance with open-weight models For software development.

00:51:22: I mean you have to put a little bit more effort into harness engineering for that one.

00:51:26: But basically I have a colleague who says oh i don't see a difference anymore between Sonnet four point six or Kimi K two point six Or something like That.

00:51:36: it's The same for me.

00:51:39: Recognize the differences in results I get and... ...I do not care about a number of tokens from these models basically.

00:51:48: I mean, I built a little prototype with both models And with Sonnet it cost about... I have numbers which i can't see there.

00:51:59: With sonnet It costs me about forty K In the forty case or a forty euros for decay would have been it little bit much For showcase and probably my CEO would have come after me for that one.

00:52:15: But as so forty euros, and with Kimi.

00:52:19: It cost me Exactly the same thing.

00:52:22: no harness engineering know nothing but quite almost this The results being almost as good.

00:52:29: As was on at two point six.

00:52:32: four point six, sorry It cost me about a dollar.

00:52:37: So it's There are some leverage in there that we can still figure out.

00:52:45: and And still Probably the colleagues from T-Mobile or Telecom would tell us then hey, we are also building some kind of hardware and Computing centers where you can run locally all these nice models.

00:53:01: And Then We can use their offerings or from other companies.

00:53:10: there are a lot of companies like Inceptron and Sweden, like Nibius in Netherlands.

00:53:15: And also Oviage in France and the German companies where we can run all these open-weight models.

00:53:26: basically but coming back to your claim it didn't change.

00:53:33: We want create business value.

00:53:35: We still want to do that and we also wanna have an eye on efficiency.

00:53:41: because again, effectiveness without efficiency means you may pay more money for creating the stuff than you might earn from it.

00:53:48: Because effectiveness does not care about how much it costs but only cares about value.

00:53:54: therefore efficiency has a value.

00:53:57: Maybe the slogan is were able balance those two things in best way.

00:54:02: Last question from my side, as we said in the beginning this goes back to our series of conversations.

00:54:08: Talking with you somebody who is a part of this community for so many years looking now and into future.

00:54:19: it's something where your kind of personally look forward when saying that this is exciting because maybe are all stupid at just looking for JNEI I don't know, something happening or maybe a precious thing we should know about in the industry.

00:54:40: It's still early research but what i find much more interesting at the moment if you look at AI developments than generative AI are the world models like Jan Le Kuen currently is building with the JEPA approaches, but it's still quite early.

00:55:02: I mean he got one of the biggest fundings...I'm not sure Cedar A series funding that ever was anybody got in Europe.

00:55:12: He went from META to Europe and France actually, I mean he's French speaking so that is an obvious choice for him.

00:55:23: So there are a lot of interesting things happening outside this saying okay with symbolic AI.

00:55:31: when the people examined Claude Cote after the code leaked they figured out internally, it has built in a lot of symbolic AI approaches to basically put on top of the generative AI approaches.

00:55:53: In order to balance the oddities that you get from Generative AI in some way.

00:56:00: It's still not perfect and there I mean we experience at every day if we work with Claude Code but It's interesting to see these kind of movements.

00:56:12: I think that is very interesting, and... ...I'm curious if we eventually understand this importance of resilience in dealing with all the uncertainty which we face at several levels….

00:56:30: …at market level, at geo-economical and geopolitical levels but also on other levels as well personal levels.

00:56:38: then, if we figure out how to balance this sustainability thing.

00:56:44: Not necessarily only in terms of ecological sustainability but also economic sustainability.

00:56:52: so not looking at the here and now Which is much better in companies, which never went IPA.

00:57:03: Which never went to the stock exchange Never did their IPO But anyway and also how to deal with this exploding complexity And so on if we eventually will learn that.

00:57:16: but it's more like a thing of curiosity If there would be islands or if these will be spread out of necessary necessity.

00:57:24: basically

00:57:29: Well, thank you so much Uva and Sebastian for being here today.

00:57:33: It was fantastic to talk with you.

00:57:36: Thank You all from Being Here Today at Jax either live or online And I'm sure you will find this entire conversation in probably excerpts available on DevMeo and Vickla In the near future.

00:57:49: So

00:57:51: okay?