# Nois AG: Spec-Driven AI Development Pivot

**Podcast:** Stories Connecting Dots with Markus Andrezak
**Published:** 2026-10-26

## Transcript

The software part is actually the easiest part.
And to manage the change and outside everything new, that's the most important part.
For me has changed a little bit.
There are the Grenzen away.
We are not so strong in the regulates.
Wonderful.
A little more free.
And yeah, a bigger Ziemark, also it's not.
What's that?
And just as a little have we the power to adjust it faster than any big, heavy, heavy, heavy.
Stories connecting dots.
A business podcast to discover a world of possibilities through stories told.
Yeah, a good day together.
In this episode my guest, Johannes Bosler from the Knowis AG.
And the story is that the noise AG is making software, which helps to create a specific model, and then they generate software.
That's always done.
And now with the wonder of LLMs, they could do everything different and then expand the market.
My perspective, what I heard of, there is a company...
So, you're completely middle-minded.
How many of you have been doing?
70, 100?
So, you're in the area of 50 to 75.
So, a completely normal company, who puts their software production in and up on agent development, builds a product that agent development is deep.
When there is a world that asks, if that all works, what is it?
50, 70 people and make that all just simple.
And we're going to be so very natural, as Johannes just works.
And they say, yeah, it's just normal.
And it's not a drama.
But as I said, in my experience, 80% of the companies there are asking themselves, if that's all right.
Can they produce software, which is productions-reif?
So now comes the Knuller.
The software, which Johannes and Co.
have created, has their origin in high-regulated Umgebung.
Wer so was kann, kann auch andere Dinge.
Ja, wer banken kann, kann alles, würde man denken.
Und diese beiden Aspekte sind für mich super, super spannend.
Also erstens dieses, der Wagemut zu so einem Pivot und gleichzeitig dieses, es gibt so viele große Firmen und Firmen mit unendlich viel Geld und Mut und weiß der Teufel was.
The people still have to wonder if this with this Agentsian-Entwicklung all works or not.
And they all have to go on and that's all okay.
What I also always want to get in the podcast, is to show what it is and show what it is and show what it is and show what it is and show what it is not so marsianish and outer space is, what I always talk about.
And that's why it's today.
Here is a completely normal company with completely normal employees.
You can't work with them.
They can't talk to you.
They can do everything and it works.
That's just a bit crazy.
So bad is it not, as you do it.
Yes, yes, yes.
Really not.
And that will I understand all of you.
Hey Johannes, who are you?
What is your job?
Tell us what you do in life.
So first of all, Marcus, also at the moment.
Thank you for your invitation.
I'm really excited to be here, to answer your question.
Maybe to me.
I'm married 44, I'm married in Bayern in the area of Regensburg and studied ursprünglich Wirtschaftsinformatik.
In one of the last Diplom-Studiengänge was it was the good old Diplom, on which we're in Germany also proud to be.
and then I started to work directly after my study at the Noise AG.
From my personal perspective, I program always very well, am a bit of a Technic-Nerd, interested me for new things.
I can't really get music excited.
I'm going to be outside in the nature.
Whether it's with the camper or whether it's with the bike.
So I'm going to be outside and not only in Germany.
What do you do technically and program your own product?
I play incredibly good and try new things.
My newest project will be Home Automation.
A very new field, I have no experience with the programming.
I'm in Java and love this whole ecosystem and this whole community that is there.
It's a ecosystem that has been like no other programming language for me.
Cool.
Java is also what I want to say.
if I could do something with programming.
But now you don't need so much syntax to do something like that before.
Spoiler.
I had to start with the whole Doku.
You had to go to Google from any blog-eintrags.
And Google was more slow.
The connection to Google was at least.
You had to show them.
You learned to learn to learn to do things.
You know, Java is an Insel.
This book, this online book.
And what for music?
I play a classic, I play a symphony of the blues orchestra.
Wow!
And I play with 50 to 60 other music-beginers really nice music.
Mega!
Really complex?
Yes, especially for those who are on the floor.
Yes, that's mean I.
So, coordination, overview and so.
I play Waldhorn in the orchestra, I played in the orchestra, but it's been a long time ago.
But it makes always so much fun and it's a nice hobby.
Completely contrary to the thing I do with all the technology and digital and everything else, just to make analog music and give concerts and also to go in concerts.
And I played a guitar and I was okay, but in the meantime I didn't even know how to get into it.
One of the hobby-saccesses, which is a lot of things.
And that's getting to get to it, it's hard to get to it, because you come from a level that you don't hold it, and you're always unbefriediging.
But that's high to hold it is always commitment and discipline.
Is it good for a job or is it a challenge for you?
Teils, teils.
Also es macht so viel Spaß, dass es wenig Herausforderung ist.
Klar denkt man sich manchmal an dem Tag, jetzt hätte ich gerne geübt oder hätte ich üben sollen, aber ja, es geht recht gut.
Und das ist ja der Vorteil mit Homeoffice und sonst was.
Man spart sich die Pendelei ins Büro und so weiter.
Und diese Zeit kann man anderweitig auch wieder gut investieren.
Also sei es mal im Haushalt oder sei es beim Üben, je nachdem.
Also die Parallel ist offensichtlich genauso platt und so.
Also Softwareentwicklung klappt ja auch nur im Team, haha.
Ja, doch.
Aber jeder muss sein Bestes geben.
Ja, und was ich noch weiß ist, aus dem bisschen Spielen in Bands ist halt diese, ob man alleine spielt oder in einer Band und was sich da ergibt, irgendwie keine Ahnung.
Bei mir, wir hatten eine Zeit lang so einen unglaublich guten Drummer und schon war die Welt doppelt so gut.
you could have so much freedom allowed, because they had so much taken, because they had so much filled.
What a dream, with someone to play.
For me was always a experience in the band, what can I be able to do, although you are not the best in the world.
For me, until today, when I work at a product, I am not the greatest, the greatest, the greatest, the best decisions, but it is as a team.
And this experience, that you as a team so much or as a orchestra, what is probably much bigger than a 5-Mann-Band or so.
That's I find at music so direct spurtable.
And at software I find it incredibly hard to be spurtable.
You have to go up and down to release and look at yourself and a customer will get it good to find.
A half a year later find you it out.
But music is just like, boom, that's the moment.
So direct and unmittelbar is music.
And it's extremely, what for energy is.
And it's not just that game itself, but you get from the public again something.
And so on.
Yes, it is very special and it is a bit of a bit of a bit of a team sport from software development.
Because together comes something better out, as if you would do it alone.
That's definitely so.
Lustigerweise hat es ja sogar mit eurer Software zu tun.
Ja, okay.
Im entferntesten ja.
Ja, ja.
Tatsächlich.
Ja, doch.
Jetzt habe ich gehört, tatsächlich, kennst du keine andere Firma vom Arbeiten?
Du hast bei Nois angefangen und nach der Uni.
Nein, ich hatte vorher Praktikum und sonst was gemacht.
Dann haben wir auch umgeschaut.
Yeah, but let's end of the day I was hanging out and had to be there in the poor and around the world, from the developer to the architect to the CTO.
And I'm now responsible for the technology development, for the innovation in the product.
And let's end of the design of the product.
Because I'm the user person number one and my job is to make it, me and others happy to make.
I wanted to ask, I come from a consumer-oriented world with the eBay-Based and so.
You have also Consumer, but it's already a very clear B2B-Product.
How are you doing?
Have you done so many product managers or make everything out of the technique?
The product chef, where does it go to you and how does it go to the responsibility as a CEO?
This whole process is pretty fließend.
We have of course product management and we have of course the business things to do.
It is always a business company, even if it's a lot of fun to work there.
But it's going to be a win maximization.
And very clear there is also this business site, but the whole design and product management.
It's just working hand in hand.
It's a nice relationship.
One is the other one.
It's also a good thing.
If you're so technical, you could develop solutions, which can't be solved.
But there are a few regulations in the company that can't be solved.
We need a problem that we can solve and want to solve for the Kunden.
My classic Vorteil and problem is that it's a technical, technical, oriented company that is marketing.
I don't know how that is.
We'll talk later about it.
We have super-geile products.
But marketing is bad and so.
Or sales is bad.
But okay, you have of course investors, which you can't do.
So you can probably sales and marketing not bad.
No, no, that's not.
And it was interesting for me.
I've also been able to do this in these areas and stuff like that.
And that makes it so exciting to work.
To go in this kind of acquisition and anyone from the ground to for the product to begeistern or anything.
It makes really fun and gives so much attention.
And from this point is it has been not been done for 20 years.
Yeah, how cool.
Who has it already?
And who can it already so say?
Maybe you can tell us what the noise AG is and what it is in the end.
But now not the product.
Also the product will get now a extra highlight.
Tell us a bit about the profile of you as a friend, so that we know in which context everything is played, what we will talk about.
Yeah, we are in a software house, in Regensburg.
And only in Regensburg, also in Bayern, is very nice.
Also I can only recommend for a visit.
The Altstadt is currently in the world.
They are also from tourists, but it's still nice to see.
And we do the same thing we do for the customers' business.
And that's why we do the same thing.
We do the same thing we do for the customers' business.
And we do the same thing we do for the product itself.
And we have another area of technology services.
where we help them to help us either with our product to build or to build a product or whatever else is necessary to get the customer.
I understand.
And what is the product that you today offer?
What we now do.
We offer the Cloud Solution Workbench to the Cloud Solution Workbench.
And that is a product that allows you to say it now.
Spec-driven Development zu machen, aber nicht nur mit Texten, sondern mit strukturierten Modellen im Hintergrund und das nicht nur auf deinem Arbeitsplatz, sondern auf Team-Ebene.
Ich glaube, so kann man es am kürzesten umschreiben.
Jetzt hast du ein paar Sachen gesagt.
Zum einen hast du gesagt Spec-driven Development und auf der anderen Seite...
Team, let's first understand for those who heard nothing about it.
Spec-driven development has a few properties.
That's a big word for agent development.
It's funny that...
but also a lot of criticism.
I was very much in the agil environment and many of the old ones said, Spectrum Development, we know that it can work for 15 years.
That's the world.
And that's what I would like to do.
Let's first understand, what Spectrum Development is.
Let's say it is, everything you write, and that it goes into development.
You think first, what you want to build, what are your constraints, what are your requirements, what are your methods, what are your methods you want to have.
You write it off or you bring the information in digital form.
And then you give it to that to the development, which is just to use code in software.
And there is a lot of information.
Behaviour-Beschreibungen, the whole architecture can you describe and in the spec.
And you can even further drive, you go to the implementation design and write specs, how the classes should be structured and so forth.
And there is so 15 years back.
had not the technology and the possibilities we have today.
Today the whole world is different and there are new things.
And before 15 years, I think it really didn't work.
Before 15 years we had to say about the water.
And there was Agile that completely contrary thing.
And I think today we can just mix these two things.
There is for me so, when the world changed, it works.
But, this text document makes no fun.
And just in text writing is kind of anstrengend.
Also for those who write the text and then find out that the text has no longer read, except that someone had this thousand times read.
The idea that you first...
how to write, how to behave, what should be delivered, what is the outcome.
I try to find a little bit, what the agile community is finding.
Kent Beck has it with User Stories.
He describes only the world of the customer, when he uses the software.
That is not wrong, because the user story is a little tiny.
And I'm going to go directly on this AI driven.
How I'm going to look at this today's spec driven development, is not so that the people, that can be done with your product also do not write a lot of Spezifikation for a year and then run down in a half or a half year or a half year or so.
But you can also use a story for user stories back and implement and so forth.
Deswegen, die Kernidee von Spec-Driven ist ja erstmal Gedanken machen, aufschreiben, spezifizieren, Outcome spezifizieren, eventuell, das weiß ich nicht so, wie es bei euch ist, die Eval, wie man heute sagt, also die Abnahmekriterien gleich mit spezifizieren, das kannst du ja vielleicht gleich noch erklären.
Also die Idee, die scheint ja erstmal ziemlich logisch und ist eigentlich unumstritten.
Also vielleicht kannst du ja auch ein bisschen klären, wie weit ihr geht und was du glaubst, was überhaupt...
is an idea?
I think that in the idea is nothing to be said, because it just works.
If you make a thought about it, you get something else out of it.
Or else, shit in, shit out.
That works exactly.
And so, and so on.
For me, I'm also a part of the spec driven.
And so, what works well, is that many people work together on a spec.
and bring ideas together, different aspects together.
You have the Fachbereich, the Techniker, you have someone who thinks about it.
What didn't work, is the whole thing as text to write.
And so we decided to put it into a structure model.
So now we're at a model driven development.
We learned that we have already learned from 15 years, that is not that yellow from the egg, because it's more time to get this model right and right to get it, to generate the code correctly, so that we could write so much faster.
And this model doesn't understand anyone more, the right code is generated.
From this point, it must be lightweight and consumable.
And consumable for the people as for the machines with the LLMs and AI and other things.
And now, to make it back-driven again, I don't think it works on text-to-read, but you need a model and a small portion of it.
And now I'm back at User Stories.
Aus deinen User Stories kannst du wunderbar alles ableiten.
Und na klar, du brauchst die gleichen Informationen wie früher auch.
Und auch früher galt ja schon, je genauer du deine Idee aufschreiben konntest, desto besser war das Ergebnis hinterher.
Und jetzt mit LLMs wirst du dazu gezwungen.
Also diese Präzision muss ja da sein.
Ansonsten bekommst du irgendwas, was du gar nicht wolltest.
Weil erfinden kann das LLM immer irgendwas.
Yeah, and maybe it's also a key point, maybe this one for everyone to explain, because, as I said, my experience is that very many people have so much idea that everything can't work, but also have not seen how it works.
And in my observation is it just so that what people have been able to chat with an LLM can actually imagine is...
When I get a very vague answer, I get a very convincing answer, which fills many layers with confidence, not with truth.
And that's to be able to get on software development.
Je weniger Lücken du lässt, umso präziser kriegst du das geliefert, was du beschreibst im Outcome, in der Wirkung.
Je sloppier du das beschreibst, umso schlechter ist der Outcome, umso mehr machen halt diese wild gewordenen LLMs, die kleinen Biester, was sie wollen und liefern dir irgendwas, aber nicht unbedingt, was du wolltest.
Das Problem ist aber, dass da Fehler auf der Anwenderseite ist, also auf dem, der zu schlecht spezifiziert hat.
Ich bin jetzt nicht der beste Programmierer der Welt, überhaupt gar nicht.
Das meiste, was ich mache, ist eher in so einer Business Ebene auf Strategie und sowas.
Da ist es aber genau dasselbe.
Wenn ich den Outcome nicht gut beschreiben kann, kriege ich irgendwas geliefert.
Und das ist auch viel Kritik, die man so erntet auf der Welt von Leuten, die eben nicht sehr gut beschreiben können, was sie brauchen.
Dann kriegst du halt diesen berühmten AI-Slop geliefert.
When I do something like Workshop-Simulation and so, when I very well know how Workshops work, I can describe a simulation and give me a workshop.
When I say, give me a workshop, it's not so good.
If I have to share Rolls and debate, then suddenly very good results.
I think that's what we have to do with the this kind of software development can make, when we context schneiden, when we the specific specific specific specific specific, when we have the individual tasks on small tasks, that we have to get very precise results, where we have to get very sloppy results.
And I think that is also a bit of what you want to help, this precise description to get.
And what I've heard from now is a few of them, where you help.
The one is...
Ah, ein Team kann das machen.
Also es muss nicht einer alles wissen und nicht Word of Mouse aufschreiben.
Der Hans hat gesagt, ich glaube, der Philipp würde dies meinen und die Greta würde auch jenes denken.
Sondern die Leute können um eure Software drumherum sitzen, das kollaborativ erarbeiten.
Und sie können es viel expressiver machen, als einfach Texte zu schreiben, weil sie auch Diagramme, Grafiken und weiß der Teufel, was alles als Spezifikation schreiben können, damit...
can be described as expected.
Is that so grob?
That's very well explained.
What is also important for us is that we want to support the idea in Code to get, this whole workflow with to build.
We have a Design Assistant for example, the can already make some advice, how you can translate a artifact into another.
And it's always the same problem, Teile and Hershey.
You have a big problem and you have a lot of small problems.
You go through different layers, layers, layers, as you call it, and at the beginning you have the idea.
Then you have requirements.
Then you have a architecture and you land at a design implementation or design specification.
And on this whole way we just want to help with these models and not only as text.
Of course, there is also a text-based, where you can just write everything in the same way.
But that can be changed in structured models.
And there comes of course precision, there comes a little bit of standardization.
Because you can also say, a problem must always be in the same components, the same solutions are always there.
The same thing is behind.
You can also play and make it again.
And then it's about the best solution to find, that also graphically or whatever.
And you come always by small pieces.
And these small pieces can LLM very well understand.
Or if it needs more context, you can read the four other connected pieces with and you can then get a good result and you can also generate a good code.
Before I just go ahead and get the Knaller out of my mind, I would...
I would like to ask, what does it mean for a company, if the software is supported, that teams are there sitting there, to write a specific specification?
That is not how every company works.
It is not how your original bank account is.
How my own case gives it to me, how they normally work.
Is it for you, when the software comes, a change?
How are they all so out?
Because it's super cool.
That's original, what we wish for hundreds of years, that people so work, so Spezifikationen work.
It's not what normally in all companies is practiced.
There are these silos, the whole thing is not.
But to bring people together in very regulated environments, I'm not sure.
How is your experience?
Also, it's also difficult, but then you stay in your Silos and that's also supported.
You don't have to work together.
You can still produce your results.
When we're all Git-based, you just have to commit it to Git, you have a version of the stand and with that can a different team work.
And there it's more about the models in the background and they are always there.
Team-übergreifend, muss ich sagen, funktioniert es recht gut, wenn du einen definierten Stand übergeben hast und wenn du natürlich noch die Sprache vom anderen Team kennst.
Also deswegen ist bei uns ganz wichtig, dass diese Anpassbarkeit einfach da ist, dass du mit diesen Dingen modellierst letztendlich oder beschreibst, wo du auch heute schon benutzt.
Also es hilft ja keinem was, wenn wir sagen, das Beste, was du tun kannst zum Spezifizieren des User Stories zu schreiben.
Die schreiben aber seit jeher schon Anforderungsdokumente.
Dann werden sie sie nicht zurechtfinden wollen.
Und diese Anpassbarkeit, diese Flexibilität muss einfach da sein.
Ansonsten wird es gar nicht funktionieren.
Und wie das Team intern organisiert ist, be my guest, wenn die in Silos arbeiten wollen, auch gut.
Hat auch seinen gewissen Charme.
Yeah, yeah, also, wie gesagt, also, ich würde es auch immer gar nicht so werten wollen.
Also, die Unternehmen sind ja im Allgemeinen erfolgreich, sonst würde es sie nicht geben.
Insofern muss man die auch nicht immer evangelisieren, wie sie zu arbeiten haben.
Und solange es für sie funktioniert, ist ja alles gut.
Also, wir müssen ja die Welt nicht komplett retten über alles.
Also, ich glaube, das Wichtige ist, dass du alle Leute mit einbinden kannst.
Also, dass du nicht durch irgendwelche Hürden und sonst was Leute ausgrenzt.
Also, weil es zu technisch ist oder weil sich der, also ich weiß aus Erfahrung, ein Fachbereichsleiter oder ein Business Analyst möchte nicht in einer IDE arbeiten und quasi Code schreiben.
Das ist nicht sein Metier.
Er möchte einen Rich Text Editor haben, der am besten wie Microsoft Word funktioniert und Excel dahinter.
And there you have to have a good thing to have, that you bring to everyone in the tool, so that there can really be a single source of truth.
And at the end we want to let's all their knowledge of their knowledge, so that a better result comes out.
That we need to support.
That's the point.
And the more precise information is, the better results again.
I think with the whole shift AI and so on, we will spend less time in the implementation.
That's now already possible because it's really fast, so a code to create.
But we will spend this time with the better ideas and the better prompts, or the better design to get in there.
And now comes the key, which I already describe and will.
This company is this company.
And if we don't look back, don't know your history, then you're a company, which is today on the market and baut a tool, bietet a tool for spec-driven development at the end.
Actually almost the whole cycle.
And she supports...
this new software development, which works now, maybe since a year or so, and says, yes, that's what we're doing.
We're going to go to companies that already work, so work in order to work.
We work also a big part so.
And, let's say, the world is so.
And, on me, with a relative self-consciousness.
And I bring a total absurd comparison.
Eine andere Firma, die mich immer so überrascht mit dieser Selbstverständlichkeit, ist am vollkommen anderen Ende des Spektrums Amazon.
Wenn ich zu Amazon gehe, da bist du umgeben von Leuten, für die ist diese Zukunft absolut unumstritten da.
Alle werden so arbeiten, sie selbst arbeiten so und dann müssen sie auch noch rausfinden, wie es geht.
Alles egal, aber es sind kleine Holpersteine und so.
Aber dass man so arbeitet, dass fast alle so arbeiten werden.
These tools to use and how we learn this technique.
That's completely unumstritten.
How much that's done rumpelt or not, can now be egal.
But you come into these Amazon-Royages and say, that's absurd, how clear that is.
Because in my world, that really many companies struggle with, whether it's going to be, how it's going to be.
Can it really be, that the code in production is used?
And so on the other side of the spectrum, you have to say, the companies that are just completely self-sufficient.
And now comes the funny thing for me.
The companies that are completely independent, go so far as Amazon on the one end of the spectrum and you on the completely other end of the spectrum, the scale and so on.
And that's for me incredibly unschlossen, mutig, cool, right?
So, we can't even talk about it.
Erklär mir kurz nur den aktuellen Stand.
Wie kommt ihr als Unternehmen dieser Größe und Ausrichtung dazu, euch so stark auf diese neue Technologie zu stürzen?
Letztendlich ist es ein bisschen Notwendigkeit oder auch aus der Historie begründet.
There is where I say, this change is for us not so extreme.
If we make an Ausflug in the Product history, we do always the same.
We modellate something and make a solution.
And at that point, 15, 18 years ago, had a product, where you could already modellate something.
And we had this modellate artifact then interpret.
and ultimately the solution was built.
And that worked wonderfully.
You built very quickly some solutions with domains in the background and other information.
It was a bit difficult to get this whole situation, because it was interpretable with Scripts or Snippets had some difficulties.
But it worked really well.
And that's the same idea.
One generation was written, Cloud-native development came on.
Everyone wanted to go into the Cloud, Kubernetes, Deployment, OpenShift, and how you always name them.
And then, at the point of time, came the idea that this behavior problem could be very simple to solve, if we can solve our models.
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
And it was not just a good idea of Codegeneratoren to standardize, for all possible cases to think about it.
And there something to do.
The one will in JavaScript, the other will in Java, the third one will go and .NET then also to do it in different variations.
And the one will do it at the beginning, the other at the end.
You have just changed templates.
And templates are not so the best thing you can do in your life.
Then came the whole LLM-Chief on and that was for us the conclusion, okay, with LLMs we can do it, that we don't have the Code-generator in the product to us, but that these LLMs can do the code from the models.
That was the step, the one that was really needed, um this flexibility in the product to be able to do.
And then it was so...
Now it's nice if you have models, but the models must also be flexible.
And that will we do now again graphically make and not just in the tabellae properties for entities, but now with graphic support.
And so is our whole product entstanden.
And so far from being a big change.
And now is for me again so a thing of so a sense.
So, warum das für euch leicht gefallen ist, diese Technologie anzuerkennen und direkt zu benutzen in eurem Sinne ist, weil Code-Generierung für euch kein Wunder war.
Ihr macht das schon immer.
Ihr lasst Code generieren.
Und ihr wisst sagen, es gibt Ebenen, auf denen man Code generieren kann, der vollkommen okay ist.
Die Parallele ist, jetzt wieder ein anderes Ende des Spektrums, ich versuche seit...
I hear more than 1000 of Boris Charny interviews, um behind the marketing speak to find what they actually do.
What behind there is that Boris Charny is...
Automatisierungsfreak seit Ewigkeiten.
Früher hat er über Excel-Tabellen jeden Scheiß automatisiert und sich die Bug-Tickets automatisch reingesetzt.
Heute hat er LLMs, die diese Übersetzungsschritte für ihn machen und sagen, jetzt kommt das Automatisierungswunderkind in so eine Umwelt, wo Automatisierung über LLMs viel einfacher ist.
Da kann eins und zwei zusammenzählen.
Deswegen kann er da halt für andere absurd wirkende Systeme ...
Bauen, den er stark vertraut, weil er sehr genau weiß, was er tut, weil er seine Erfahrungen da letztendlich besser ausnutzen kann.
Bei euch war die Erfahrung Code-Generierung.
Das sind Schritte, denen ihr vertraut, wo andere jetzt noch hingehen müssen und sagen, what the hell, generiert der Code, was soll das denn?
Das wird doch mit der Hand geschrieben.
Ich gebe mal ein einfaches Beispiel.
Eins meiner letzten doofen Projekte noch als Studi an der Uni war, Java came on, I'm older than you.
A little.
Java came on and I wanted to be able to be able to be in a SQL-Datenbank.
What I did was, I had to generate code.
the objects, also, let's say, an adapter between objects and the SQL data bank.
I'm not doof, but I've always been generated this adapter.
And it was me completely understandably, that this generated code works, because it wasn't a big black box.
And I think, if you have long enough experience in Code-Generierung...
we know that code can generate and work.
People who have never tried it, said how can a code work.
And I think that's today a bit of the world with the trust in how you always generate a code.
Yeah, I must say that this deterministic generated code has many benefits, because you have tested him one day and then it worked.
And you can be quite sure that it's for other things again.
Insofern war es sogar einfacher an der Stelle.
Und was wir jetzt aufgeben mit LLMs, ist dieses Deterministische.
Und da muss ich sagen, liegt es aber auch sehr an dem Input, den es du gibst und an den Guardrails, die du nebendran aufstellst.
Also ganz klar muss es irgendwelche Skills geben, die beschreiben, wie dieses eine Modellelement am besten in Code übersetzt werden soll.
Und da darf nicht viel Interpretationsspielraum bleiben.
Und dann kommen relativ gut vergleichbare Ergebnisse wieder raus.
If you don't let Lücken and don't describe how to translate it, then you will get something to find and it will always be different.
But with this technique you can achieve a certain standardization over all the LLMs and something else in the company to get.
And I think there is another important point for people to understand how to use this whole technology.
What I always say is that the...
The Code-Generierung is not the Wunder, but the harnessing and the guardrailing, the engineers' performance, the ensures that the Code does what it should do.
That is not a Wunder, that is a lot of wisdom, discipline, work, which you can do in the infrastructure.
Maybe you can describe a few elements that were you forced to be given, because the Code-Generierung not deterministic.
Also, you said already, the parts that are supposed to be generated, are very precise written, but still you have to do the other side of the code to check, that the quality is there.
Can you describe what you do on the other side of the code?
Let's do it.
Also es gibt immer dieses Problem, dass du mehrere Coding Assistants im Einsatz hast im Team.
Und bei uns gibt es jetzt schon mal einen Weg, wie du zentral einfach deine Skills und deine Einstellungen für das Projekt in dem Projekt verwalten kannst.
Und das funktioniert dann für alle Coding Assistants relativ gleich.
Also es ist einfach so eine Convenience-Funktion.
Und damit packst du eben die ganzen Skills und sonst was, die beachtet werden sollen, eigentlich schon mal zu dem Developer in die IDE rein.
And that's already the first step.
You can't build up small pieces, which you can build up and so on, to generate.
And then it's also possible for people to do this review process.
Because of course a small, small, small part of the entity XY, this command or this prompt will be able to do the same result, which you can do relatively easy to review.
And it's also going to continue to do it.
You can also specify tests and so on.
So everything you can think about, you can do so.
Or you have said, you should make the JUnit tests and do it again with your Acceptance criteria.
So, such things go down.
Now, just a little bit of a quick experience.
That means that your product has a team-based environment.
Viele Firmen haben ja heute das Problem, die sozusagen schon alles weiter treiben, dass mit den neuen Modellen fing alles an, dass Leute ihre individuellen Skills gebaut haben, ihren individuellen Kontext gebaut haben, aber Zusammenarbeit untereinander oftmals schwer ist bis heute noch in Unternehmen.
Alle suchen, wie können wir das in Git vereinheitlichen, in welchen Kontext teilen wir, welche Skills teilen wir, was sind für Prozesse um Skills wieder und dann abzugleichen.
Das heißt, so eine Ebene habt ihr letztendlich schon drin bei euch oder ist es nur für euch intern?
No, it's a product.
It's not written, how you can use the skills.
There's no method behind it.
But it's at least the mechanism there, that you for a project or for a project type, that's a Java service as Microservice Deployed on Kubernetes, you have the same skills and can be done with it.
We call it baseline.
And a baseline is basically the Modellierungsartefakten, what you can do all of them and how these things are translated, so skills or commands, so that it's a bit easier to use.
And finally, we have also found design recipes, how you can do the elements from the one's and the other's.
And that's what we're going to do together and as a verwaltable artifact.
That can tell you to tell us about the teams or at least team with the same project type.
Cool.
So, and now comes a killer insight, sozusagen, from your own experience, that you're code generating and infrastructure for you, based on a good specification, which can be done in a team, so that everything is relatively precise.
Now comes with the LLMs The insight, the way we can't build the generators and determine the deterministic system and that we can't get into it.
We've heard that it's not really a gift, but a relatively natural step for you.
Now comes the idea on the next level, that you say, we're not going to be in the business anymore, which we're not going to be in the business.
Genau, also diese Idee ist auch schon lange gereift, die kam auch tatsächlich vor den LLMs schon oder ziemlich gleichzeitig mit den LLMs, bevor wir die LLMs überhaupt angewandt haben, haben wir festgestellt, dieser Ansatz funktioniert auch nicht nur für Banken, sondern erstmal für alle regulierten Industrien.
Es gibt sicher ein paar Ausnahmen, also Automotive oder sonst was, wo es ganz andere Standards gibt und so weiter, aber für so...
General Purpose Industries would I call it relatively straightforward.
And also for software development.
It's clear that there are different approaches, methods and so on.
But it doesn't just work for banks, but it can also work for other industries.
And that idea has been taken.
With this change, where we said that not only the bank market is relevant, but also the other, came the change in the process from this I sell a solution to I sell a product.
That was the extreme change as this industry-wechsel, but really this product-centered sell.
But the general Witz is that you...
from outside perspective, you have learned to be able to experience it.
And the experience was, with the knowledge that you have done, if you have everything generalize and open, you can use a lot of market to use, because you have the function in this small, small, hard Niche bewiesen.
And it has been found that everything, what you do, is much more and more usable.
Right technologically, there was no change in the product.
Technologically, there was changed with LLMs and so on.
But now from the idea of the kernel, it's always the same.
And it was more than the environment that was completely linked.
You need new messages and so on.
You have to go on the market and so on.
Selbst hin zu den irgendwelchen Fotos auf der Website und sonst was.
Also heutzutage brauche ich nicht mehr im Anzug irgendwo stehen und früher war ohne Hemd in der Bank aufzutauchen, ja, das No-Go schlechthin.
Da hat sich also auf vielen Ebenen einfach viel getan dann.
Vielleicht um alles zu verstehen, hilft uns ja auch nochmal zu verstehen, warum eure Lösung gerade in den regulierteren, komplizierteren Umfeldern einen guten Fit hatte.
Also das muss ja nicht sein.
Das wisst ihr und so, alles cool, wir wissen es nicht.
Vielleicht kannst du nochmal erklären und sagen, wenn ich in einem regulierten Umfeld bin, was war der wesentliche Vorteil, der mir in diesem schwierigen Umfeld sagt, ich sollte so eine Lösung wie eure verwenden?
Also der größte Vorteil war letztendlich schon immer, dass wir verschiedene Leute miteinander zusammengebracht haben, ein Ziel zu verfolgen, also verschiedene Rollen eingebunden konnten.
And then of course, that Code generated a lot of effort.
At the beginning was also deterministic, what also for trust was used to be.
You have to accept that Code generated and that the there is and that the maybe not look like you would like to.
But as soon as this one was done, you had a great speed.
And that was the special thing.
That was the special thing.
That was the acceptance.
Is there not a gemeinsame Ebene between, verrückterweise, the complicated, regulated Umfeld and agent development, that both these precise specifications need?
Also, both the Agent, the just anything else, as well as a regulated Umfeld, that says, where is it going to go?
Is it really specific?
Is it exactly specific?
Can I do it back?
That's the key is...
that you can use both cases to be able to use on the specific level.
Exactly.
And the only thing that's now gone away is a bit of these big processes, these big processes, where the most important part of the team was with four eyes and what we know.
That's not yet so hard and I'm not sure what's in the bank's hand is.
Aber diese Freigabeprozesse sind jetzt nicht mehr so streng.
Aber die Präzision ist genau die gleiche.
Und tatsächlich, das ist eine schöne Parallele eigentlich, weil wenn du dem Agenten nicht in seine Schranken weißt und ihn regulierst, wird er halt irgendwas machen, was du vielleicht nicht willst.
Also für mich reicht da gerade so eine Erkenntnis.
Und du musst mich zurückhalten, wenn ich durchdrehe und so.
Aber das Schöne daran, finde ich ja, ist wegen diesem ganzen...
a completely different trust against this agent development.
The Witz is just that you can make it all the way to make it all the way and everything is furchtful.
What you must understand in the moment where agent development works, is she just disiplined.
and not Vibe-Coding, when you generate prototypes, that's also a good point.
really precise, high quality agent development is on both sides, on the Input and on the Output side with the test harnesses and so on, high disiplinier.
I think that goes a little bit verloren, that the great stories we hear about agent development, they just happen in a high disipliniering environment.
I need to process, I need to infrastructure, I need to be a certain discipline behind me.
And so from, I'll let it develop, let's see what happens, that's not what happens.
And I think that's relatively much...
From what we see, I personally get a lot of personal messages on the articles I write, I have tried it, it has not been done, but for us is one of the testing-crumbs.
Of course it doesn't work.
What's the quats?
If you're a sloppy software-exploration process and have it on AI, it's sloppy and bullshit and it's just a bit worse.
That's what you're doing.
If you're sloppy and have it on AI, then it's a bit of fun.
If you want to have AI, you have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to But this precision to create, and that makes this model, this structure, the erfaring, light-of-the-day language, that enables you to do it.
And so you get precision.
You could theoretically even put it into the field, which should be filled, without it it doesn't continue to go on.
And that creates a different level, as if you put something in your prompt.
Let's say it's also about structure.
Yes, without end.
It sounds so long.
Yes, it can be.
It can be.
It can be.
It can be.
It is actually not.
We don't have to think about it.
When I speak of Models, it's like for 15 years with these high-formals models.
That's not so.
The LLM can also a bit with other models.
Also für mich war es ja bei der Arbeit, die ich mache, ein relativ, oder der größte Lernprozess war für mich, herauszufinden, auf welcher Ebene ich mit den Modellen arbeiten muss.
Also wo muss ich was vorgeben, wo muss ich nichts vorgeben.
Und so ein Learning war, dass wenn ich es auf einer zu tiefen Ebene vorgebe, geht es schief, weil das wissen die Modelle besser.
Wenn ich es zu abstrakt vorgebe, geht es auch schief, weil es dann zu viel füllen.
For me was a lot of practice about how to interact with which I interact with you.
How is that with your tool?
Let's say you have to always find out where you want to go.
Some want to have more freedom in generating output.
What we have found is that this in Phasen or in Layern Denken helps always very well.
And so that you have to go from requirements on your architecture and there are already the capabilities maps.
And in the description of what this capability must be done, must obviously Precision.
It has but with the code still nothing to do, because I don't give the capability to generate, but I set another one in the middle of it.
Also zum einen über die Architektur und dann sagen wir noch, okay, so eine Design-Spezifikation macht auch noch Sinn, dass du mal beschreibst, welche Entities du hast, welche vielleicht mit Domain-Driven Design ein bisschen angehaucht, welche Commands gibt es denn und so weiter und so fort.
Und auf der Ebene kann man dann den Algorithmus wieder...
or that LLM, the algorithm to implement it.
Because it's not about the exact syntax, I don't want to give it to you, but I want to give it to you, I want to give it to you, I want to give it to you, I want to specify what errors are going to throw and so on.
So, now it's a tool.
Could you use your own tool to build your own tool?
Yes, we had at the beginning really problems with it, because it just didn't have to pass on.
These all the tools have been done with us.
We can also use them to use them.
It's interesting to see what happens.
One of them, we get these effects of the same way.
We have a product-sourced, that we create a single source of truth.
We collaborate together with other models.
Our diagram grows and grows.
The discussions are shorter.
We can quickly get to the point.
We can better focus.
We can understand ourselves, the mistakes are gone away.
That helps us, only for our speed.
And the other thing is, if you do eat your own dog food, you test it, also, at the same time.
very nice to see how the colleagues can experience their own product.
They can experience it and tell them to tell them to tell them.
And they wonder why they have not tried it before.
And the other thing is that if something doesn't work, then the way to the colleagues is very short and there is also a feedback that the function is okay but I need it so and so and so.
And so the tool is also very similar.
It's much easier and much easier.
And a positive effect, the colleagues wish to be more than ever releases.
At the moment we release it, so, grob gesagt, once a month.
But I needed it much earlier.
And, you know, it's a customer being, it makes so much out.
It has positive effects on all the edges and ends.
Who knows, what's going on next to the release cycle?
Yeah, yeah, also...
And, yeah, it's nice.
Total spannender Rückkopplungseffekt.
Ihr braucht sozusagen eine gewisse Öffnung, um das Tool selbst benutzen können für euch.
Jetzt ist die Öffnung da und jetzt treibt es sozusagen umso mehr an, wie sich das Tool öffnet und anwendbar ist und Qualität und alles drum und dran.
Verrückt.
Genau.
Nee, eigentlich nicht verrückt, eigentlich logisch, aber sehr schön zum Sehen.
Ja, ja.
Und ich kann es jedem empfehlen, der selber ein Tool herstellt, also setzt es selber ein.
Also, was erstmal so war, ist...
that our product completely left left, when we said, we're going to go away from the statics of the generators.
That was already the big thing.
And in our work is now so that we get a higher standardization, if we generate something with LLMs.
We had it also used before we had the tools used to use.
But we got this standardization again.
We got exactly these problems that others have.
Everyone has their own skills, their own prompts and their own ways.
And we didn't bring them together.
And what was difficult at the beginning was, was the first to build a product for ourselves.
And of course, every change is a way of changing.
It's not always been greeted with the open arms.
And now, regarding the LLM, there are many issues that you can ask me kritically.
Like, ethics, what's going on with my data?
Are the business practices really okay to finance them?
Let's do it through my business and through my business?
But there is the reason why there is the reason why there is this possibility of swimming, to swim, to swim.
You have to lose.
And that was at the beginning difficult, this Akzeptanz itself to set up.
And now what has changed is this shift left.
We use more time in the Spezifikation.
We use more time.
first to think, to design, to look at the diagram and then first to generate the code.
Because the code-generated part of the code is really quick.
And there are now new tasks.
There are people who make skills and make them available for others.
Früher, in two years, nobody thought about it.
In a job description should be, you should be careful, that AI can work for us.
So things happen new.
Have you noticed that you have to explicate for yourself?
Can they describe a job or are those kind of jobs that someone has?
It's like a job that people have with.
But they get the job, let's say, we need a baseline for this project, so we can do it better.
So we can do it better.
And that's just jobs that it's before, I say, a year ago not so much.
How do you go with the fears that people have like the development of the token use?
Also in this case we're still comfortable in this case, because for us it's always so, everyone has their own budget.
And since we've still no automatized agent in the use of it, is always the developer who sits in the car.
He says, when something generated will be generated and holds the token in the way.
We have a lot of advantage of the product.
The token-based cost is low.
Especially because it doesn't have five different routes, where it's all over again.
When you need to buy a little more tokens, then it's also a good price.
The rate, which you have generated, is very high.
And that's also very high.
If someone needs more tokens, then you'll get them, but only on the way of asking and with a better interest.
So it's always interesting to buy a budget for everyone, but also to use it as a useful tool.
Auch die Konferenz bei dir, wir waren, jede Konferenz hat es gerade tausend von Nebentracks über lokale Modelle betreiben.
Ist es was, worüber ihr nachdenkt oder nicht, was ihr einfach beobachtet?
Bekannte von mir bei Zipgate, die experimentieren da sehr stark.
Sie sind auch Telco, die wollen jetzt...
auch strategisch sozusagen ein bisschen unabhängig sein von irgendwelchen Schocks, die es gibt über wir dürfen gewisse Modelle vielleicht nicht mehr benutzen und sowas alles.
Also gibt es da 20 Gründe, über die man nachdenken könnte.
Wie ist es bei euch?
Seid ihr komfortabel mit dem, wie die Geschäftsmodelle jetzt sind oder denkt ihr da auch über mehr Unabhängigkeit nach und wenn ja, auf welcher Ebene?
Also wir haben zum einen, haben wir uns schon mal geöffnet, weil wir jetzt mehrere Provider können und nicht mehr auf einen spezialisiert sind und du hast jetzt schon mal die Wahl und kannst natürlich dann auch verschiedene Modelle hinterlegen.
Die müssen natürlich verschiedene Gütekriterien erfüllen, sonst wird es wieder nicht funktionieren.
Das ist so der eine Teil und wo wir uns komplett herausgehalten haben, ist die Geschichte, mit welchem Coding Assistant du den Code letztendlich übersetzen wirst, weil es passiert eh auf der Entwicklermaschine.
We put our specifications as YAMLs in the repository and then the coding assistant is not our main topic.
And if the customer speaks with your local model or whatever, be my guest.
What is interesting is that IBM plan with Bob.
So that's the whole thing on-premise and not in the cloud.
So I think we will see that the development of the development is that we want to just the bigger and the gold-given local things to have and not with other parts, because they have fear of their data.
And for your product is it so that it is token-agnostic?
Where the customer takes his token, is you regardless.
Yes, it's not a problem.
You can also write it with bring your own key.
Yes.
And so we are relatively open.
And if someone local the interface can fill the interface, then it will work with the local model.
If the local model is less than anything.
Now there was...
Letztens hat Benedict Evans das so schön geschrieben.
Im Moment finden Softwarefirmen raus, dass Software schreiben gar nicht das Schwierigste in einem Softwareunternehmen ist.
Und jetzt baut ihr sozusagen an eurem Pivot hin in diesen neuen Markt.
Die Software ist da, aber die Software ist halt nicht der ganze Pivot.
Was müsst ihr noch machen als Unternehmen, dass dieser Pivot which is also possible, except for writing software, which is what you can do and the skill of the skills.
You said before, even the photos on the website have been different.
You can't go on the hand, super cool, but that's probably not the whole pivot.
No, not yet.
I think it goes in all the areas of the industry.
Messaging, what do we tell you about the product?
We have to adjust the words and other things.
We have been doing the change from the solution to the product-centered sales.
It has a different impact on how you sell sales.
What do they tell you about the product?
And then it's also called Enablement for these people.
This Enablement is on a different way to set up.
The people don't want to visit a schooling, which was before it was a good time.
He's got to be a certificate.
He's going to be a video.
There is a lot outside.
The software part is actually the easy part.
And intern the change to manage and outside everything new to make, that's the most important part of the thing.
Also I know the story of you not, almost almost not in detail, but maybe it's just to explain it to people.
We say that so well, that it's also a change for sales.
But if you sell a solution, which is a lot of service, to be able to do it, or not so much service, the three words, the famous sales words, and the promise they give them, is so different and that sales process must so different and that if you develop software, it's not clear how the world is at all, because you think that you have to think that you have to do the same thing.
That's totally weird.
to observe how a team must work and work and the processes work, what must be shown, what are the assets that I have all been doing, what is the first conversation, what are the three questions in the first conversation.
That's so weird.
And when you're in there, so to do something to do...
how much the world changes in which you live.
Although you think, you do not make the same thing.
That's so weird, I think.
Yes, exactly.
And that's good.
From my technological perspective, for me has changed a little bit.
There are no limits.
We are not so much in the regulates.
Wonderful.
A little more freedom.
Toll.
And a bigger Ziemark.
Also, it's not just what's going on.
And what do you say to the salespeople?
Do they have a lot of contact with you?
Do they have a lot of interaction with you, so that you can have the stories and so?
Also, we have a lot of contact with me and my colleagues and we think about what is the best thing that we can show.
Because you want to show that you have a complex business.
the customer, the first time it's done, you want to try it.
You want to try it, that it's easy, that it's easy to find.
And there this balance to find and some products highlights are just in complex terms.
There are always in a direct exchange and we try to always try to get out.
Now I want to try to find one another one.
On the one side you said, So, for you is the development relatively natural.
We have trotzdem heard that there were already a lot of changes in the product, but still more.
So, now I still have to keep with what I was saying before.
Also, if you play it for yourself, it's a totally okay.
I say that's a incredibly easy step, which you go to, you totally on this technology to support and so on.
So, for that there must be Gründe.
What are the reasons in your DNA, which allows you to see all the doors and options, which are on and through the door?
You can also stand and fear and not make it.
The change is too big, that's too radical, we know.
Crystal Ball has no one, what's next, let's do it.
Let's do it, what's next.
Let's do it, what's next.
It's the way we do it.
It's the way.
Irgendwas muss euch ja Mut geben auf der Welt.
Ja, definitiv.
Letztendlich ist es in erster Linie die Chance, die da ist.
Wir sehen die Chance und die ist höher als das Risiko.
Wir wollen weiter existieren.
Wir glauben nicht, dass wenn wir jetzt nichts mehr tun, dass wir dann noch erfolgreich sein werden in zehn Jahren.
Sondern wir müssen uns den Gegebenheiten anpassen.
Und gerade als Kleiner haben wir die Kraft, uns schneller mal anzupassen als irgendwelche großen, schweren Dampfer.
And there is a company that is always been a company that is always technology-affined and visionary.
He was asked to do the customer's question, where he is now going to be parked, where he is now going to be stopped and so on.
And that is part of the DNA in the company, just this vision, this vision, this idea of thinking.
And then we have a great chance to see the investors in full line to us, support us and so on.
That's a good setting.
And then it's like we have a big team.
And this team has been growing over many years.
And this DNA is just there.
And then we have all the fun, new things to adapt and to make something new.
And so a little Startup is still there.
And that's, I think, the important thing.
We are still etabliered, but this Startup-Geist is still still in us all.
And we want to live it.
So, now, you don't have to listen to it and so, but that's exactly what I'm talking about.
Because everything, what you're talking about, is so hard to say.
Also, 20 years to keep up front, 20 years to keep up the development as you, the idea of the technology going forward and always keep up front.
Or what you say, the investors hold it.
That's not a sense.
That's a sense of communication from morning to morning, with the story telling, the validation to bring, the evidence to bring.
That's what I'm trying to do.
Everything is hard to do.
And so many companies do not do that.
And what I find so nice is, you don't want to get the world out.
You want to live healthy.
You want to not want to be 35.000 employees.
Maybe you want it, but everything you do is doing well.
Everything you do is healthy, sustainable, you go on and so.
And it lies all around that you at the end, where you say, for us is not normal, nothing is so normal.
There are not so many companies that have this discipline, that they work on the models.
The models will be more lightweight, super, and then support.
That's a tough thing.
And it's cool that it's so much fun to make.
I want to emphasize my experience for that in your size, these complex areas of your size.
the hard way to support hard-regulated Umgebung with such an idea.
You could find it easier.
I find that super cool.
And that's until today the story of this beer garden, where Johannes sits and everything is clear.
But actually it's also a great performance.
I find that so cool.
And I think the world needs such stories.
Not the story, but the company...
This self-sufficient in the way of the mood and also this way of the mood to find and not to be outside of the way, so cool.
I'm full of awe and not to be a fan-wonderful and bullshit.
How cool it is, that it's so cool.
I would wish that 90% of the companies are so cool.
Each of its own way and so blah, blah, but yeah.
Deswegen, großer Respekt.
Ich werde es auch weitergeben, wenn Sie es nicht direkt selber hören.
Ja, ja, genau.
So, noch eine Frage an den so halb beruflich, halb privaten Johannes.
Also jetzt diese ganze neue Technologie da.
Und wir haben schon gesagt, so ethisch zum Teil super schwierig.
Alles fängt an mit einem riesen Schwindel und Diebstahl von Intellectual Property.
Not every Berufssparte is good.
And as always, when a huge, huge technology, innovation and schub comes, there is also a phase where there is a lot of difference.
Also, no idea, Ruhrgebiet has it until today not 100% has been ruined.
The shift from the heart of the body to a service to a service.
And sometimes you need to generate generations.
Probably you need to generate generations.
We can't do that all.
What my thesis is, and it was made for me, that the argument is that the technology is there and undisputable.
I will not have so much position.
But anyway, this technology is there.
And even if I would die, I would die, I would die, I would die, I would die, I would die, I would die.
I would die.
I would die.
I would die.
I would die.
if I don't understand them.
That means in the moment where I understand and use it, it doesn't mean everything that comes from me as okay is taken away.
Even the same thing.
And the answer I have to do is I want this damn technology, which is probably the next couple of years and probably years old, I want to understand them, um them to use perfectly and responsibly.
That's my point.
I don't want to be a machine-stirmer, who says, you're going to use the technology so far, as it goes.
And everyone needs to use it.
Everyone needs to use it.
But I say, if you want to deal with it, you must understand it.
And it's important to use it, whether you want it or not.
It brings you as a person always in this weird, weird, weird, weird, that there's this Reibung there.
The is there, the we don't get away.
What I don't like is people who use technology in a so weird and unantwortlich Ecke to put in.
The support is all like Elon and Altman and so.
And the talking about this technology, Faschos, that word and so on.
That's a point of view.
I don't like to support you and I all living in the world like it is.
We have not chosen the world.
I and you have not chosen the world.
this AI-UFO now.
And no one knew that.
But Ignorieren is probably not smart.
So the question to Johannes, not so much on the moral level, but where do you see the challenges of technology?
Where do you see the consequences of the challenges?
How do you see the consequences of the challenges of the technology?
How do you see the consequences of the challenges of the technology?
Where do you see the consequences of the challenges of the technology?
What I can't beurteiling is what happens if the AI can learn from themselves or from themselves themselves.
If she herself can't be considered what would happen then.
There is a bit of a bit of a problem because it can actually not better.
The things are not new to be found.
They are just new to be combined.
And that must not be a sense of sense.
There is a bit of respect for it.
The technology is there.
We have to beherrschen.
I think that's the important thing.
We have to be able to use them well.
We have to know how we can use them well and efficiently.
Maybe even think about them not always and everywhere.
For the search for the web, I don't need AI to use it, even if it's very convenient.
But of course, it's first time with costs.
If we don't pay directly, then pay maybe with data or pay maybe with the environment and other things, because energy is used to be used.
But she is there, she is also in the everyday life and we can't talk about it.
And maybe we can find it again, maybe some of the ways to find.
I think just when music generated, as a musician.
It's already grenzwertig, that the whole music files are just so used to be used.
As orchestra costs money from the composer's music, so we can play from the composer's music.
Let's say there are some mechanisms to give this intellectual property at least to be able to do this.
That a certain Austausch can happen.
But we can't do the time to go back and we have to take it so.
And we are probably too small.
We can only try to be able to do it best possible.
And so to constrain, so that it can not be able to harm.
Let's say I have a question about the story of Automatization.
It's very exciting, if you look at it.
That's the banal example is Automatization of Verkehrssteuer, also Ampels.
What were the first Ampels?
Had man thought that the Ampels could?
The goal was always to be money, dollars, also Effizienz and cost to be built.
The first Ampels were built because they didn't want to be on every Verkehrskreuz.
They wanted to have a job.
That's always a bit of a furchtable time.
As I was a little kid, were we a huge family debate.
Of course, that's right.
Because of the good reasons.
Because so technology is not.
And everything we discussed, And I say not that it's true, it's just a observation that it's society so happens.
The Denkenträger will never be taken seriously at the beginning when the technology comes.
What will be taken seriously at the end of the day.
And I think that's also a phase in the end of the day.
The value is there for anyone.
That stuff is taken and the regulation comes back.
That's the society's story of Automatization and Technological Revolution.
That comes.
And then it's regulated.
It's not just regulated and then it's langsam.
And that's what we all hate.
And I hate it for example.
But it's just there.
And the other thing is that the economy of such a solution is not equal to the same.
It's not profit.
in the first wave of so many of these, every one of them.
That's a horror and all that.
But that will make me change.
The only thing you can do is try to learn to do the same thing, that you don't just have the Nachteile of it.
I think that's what I want to do with.
So everything we're just doing, this overwhelming and where does everything is, is totally typical for so many phases of the...
and look at new technologies.
Everything is not helpful.
It's just there.
What I'm a bit surprised is the speed of this whole development.
So I've never experienced this speed.
Two years ago.
We tried something with LLMs and other things.
That was my Eclipse IDE had already been with the Autovervollständigung.
And nowadays we don't think about this thing anymore, when we talk about Co-generating.
There are really good things out there.
And I think the speed is also what the polarization of the topic brings.
When it's so fast, you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can feel you can Yeah, vor allem, also, da kommen ja dann persönliche Ängste um Arbeitsplatz und sonst was ja alles mit rein.
Also, wenn man hört, dass KI alles automatisieren soll und sonst was rationalisiert, ja, was passiert denn mit mir als Menschen dann?
Also, wo arbeite ich morgen?
Was mache ich denn morgen damit?
Ja, so, jetzt noch eine allerletzte Frage.
Die Gretchen-Frage, ne?
Also, der wirklich...
Also für mich, das ist so für mich so verblüffend.
Also das ist fast schon die Suche nach persönlichem Trost.
Was für mich, ich verstehe das alles emotional, was wir gerade beschrieben haben.
Was ich erstmal beobachte, ich habe das auch mit unserem Freund Herrn Main von der Konferenz, auf der wir waren, besprochen.
Die Polarisierung ist super krass, gerade von totaler Ablehnung, aber auch so ein Das, was die da reden, kann nicht funktionieren.
Aber wir können rational wissen, dass Code-Generierung in einer unheimlich guten Qualität funktionieren kann, produktionsreif ist und so weiter und so fort.
Und gleichzeitig gibt es Leute, die das komplett ablehnen und sagen, das geht nicht.
Und das zu erzählen ist komplett unverantwortlich.
Und ich glaube, das hat auch mit der Geschwindigkeit der Verteilung zu tun.
Was lustig ist, ist, dass Technologie und Tech-Community had always had to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to have to What is the quality you see?
You personally, what is the quality you see in the code?
The quality can be very good if the input was precise and small enough.
The greater the more the quality is actually the quality.
The more the prompt is the more the more the more the code is generated.
the more impressive, the more impressive.
But I find that the quality of the quality is not available.
And the smaller, the more precise, the better I find the Code-Quality.
That's just from my practical use, which I saw.
Klein und sonst was funktioniert jetzt wunderbar.
Für die Großen ist es erstmal beeindruckend, aber ich würde noch nicht unterschreiben wollen dafür.
Und ich glaube, das ist auch, was ich beobachte.
Und was auch so ein bisschen ein Branchenproblem ist.
Fast so wie bei Siri früher.
Die erste Version von Siri war so beschissen.
Also kurz mal beeindruckend.
Da hast du rein beim Nuss.
Oh Gott, bringt ja gar nichts.
Was soll das?
And then you can't use it again.
And I think people, the first experiment with what was used to be used, one shotting and there is the prototype, forget the stuff.
That was not good, that was not good.
If you want to get results, you have to learn a little bit about techniques.
You have to know what works, how they work, how you cut the right context, how you do it.
And then the results are and on a different quality.
And I think that much of what you see on the position of the game, comes from the...
I try to try to do a chat-through with a big prompt and I make prompt engineering and so on.
So, je größer and geiler the prompt is and if I just describe the role correctly, we know the title.
Nothing works out of it.
That's what was at the beginning propagated and where it was also the Gerüchte that...
Prompt-Engineers, die unheimlich viel Geld verdient hat und so Bullshit.
Nichts davon funktioniert.
Der Break seit einem Jahr ist, dass man nichts davon mehr macht und ganz viele Vorurteile entstehen darüber, dass Leute das versuchen in so einem Chat-Window und sich auch teilweise Code generieren lassen, der in seinen Reihen copy und pastet und nichts davon ist halt sinnvoll heutzutage.
Also man muss schon versuchen, ein bisschen zu lernen, wie das heute funktioniert.
I think a good case.
Then you can always find it.
But the case, what doesn't work, because you're a little bit of a chat, that's just not working.
Yes, and so is it.
So let's say you have to know what you want to have.
And the smaller, the better.
And it's always the same.
If you have a problem, you're in small Detail problems.
And then you're so...
Atomar, let's say, now we're almost philosophically with the atomar.
Are you so atomar?
But you're on an Ebenen, where you can have the LLM.
And you could theoretically also write these little things again statically.
Why do you write not just statically for these elements, if you know what comes out of the way?
It's much easier, faster and probably cost-eating.
But at that point, it works well, if it's a bit of a limit and so on.
Cool.
So, Johannes, what are you and Nois on the agenda next?
What are the short-term plans?
The short-term plans are a few cool features to build.
where you can deal things between teams and build new systems.
You will see your requirements from CIRA or somewhere else and not manually tap on it.
On the solution, our general approach is very energy in the collaboration with IBM.
We have a strategic OEM partnership with IBM.
and also for us, there will also be a little bit of something.
Yeah, that's been a bit of a trip.
And you personally, what are you doing in the next time?
In the next time, also that's very short-term goal is now Urlaub.
And I have a new building project for my life.
And then after the school, it's going to be a little bit with music and the normal life.
It's not going to be a bit.
Yeah, how cool.
Johannes, thank you.
Thank you for a nice person in a happy company.
Yes.
All of the best reasons.
But it's great, so a positive light on things to see.
And so, as you do it, so a sense of sense.
Also, thank you.
No more respect for you and the company.
You said last time, that you...
gerade jetzt auch wieder weiß, warum du schon so lange dabei bist.
Ich wünsche dir, dass das noch lange so bleibt für dich.
Ich hoffe auch.
Sehr cool.
Vielen Dank fürs Gespräch.
Ja, danke dir auch.
War mir ein Vergnügen.
Cool.
Bis bald.
Ciao.
presented to you by Markus Andretak.