The new AI world: aspects of LLMs

E3: Three years ago, at the E3 Steampunk and BTP Summit event in Heidelberg, you gave a much-noticed presentation that can be summarized as follows: You praised SAP's individual functions and activities very highly and said that it all had a solid foundation, but that SAP BTP was not a platform because it lacked a common interface, a common license model, etc. Interestingly, not much has changed in this respect. How do you see it?

Wolfram Jost, CEO Scheer IDS: Yes and no. I think that SAP's administration interface has already been standardized. On the other hand, however, the individual tools or products or systems that are offered in BTP for the various views, for data integration and process automation and for other functionalities in terms of the method UI are not yet very harmonized. You can see that they have been developed from different angles. But there is also a lot of movement. The topic of UI is currently undergoing a lot of change. AI agents and coding agents naturally have a major influence on what BTP will look like in the future. Especially when it comes to development.

E3: And if I look at your company now: How are you positioning yourself in precisely this environment? In which Agentic AI is becoming more and more important?

Jost: Everyone talks about AI agents and agentic AI and many people understand it differently. So, for us, an agent is a piece of software. And it works autonomously. So it's about execution. It's not a chatbot. It's not a system where you ask something and get an answer. For me, it's not an agent, it's a chatbot. An agent is a piece of software that uses IT technology to carry out a given task as autonomously as possible. Sometimes with a human touch. We come from the process side. Our positioning with our platform is agent-based business processes. We are able to develop agents that can be integrated into existing business processes or even replace entire business processes. We always come from the business process. We have been saying this for over 40 years: It all starts with the process and it all ends with the process. And that's not going to change. Yes, many platforms lack a link to the business process. And we don't believe that these agents can just buzz around and do anything on their own, because they also need to be orchestrated. That's why we call our platform an agentic orchestration platform, because this orchestration of the individual agents leads to end-to-end processes. For us, this is the focus of our positioning. It also sets us apart.

„MCP servers are the measure of all things. It will be exciting to see whether SAP will offer a model content protocol server.”

Wolfram Jost,
CEO,
Scheer IDS

E3: Is Scheer a supporter of SAP business processes? Do you support what takes place in an S/4 system with your Agentic Platform or are you one of the killers of SAP systems?

Jost: Yes, that is the question of all questions. Will these SaaS applications survive, as they were built many years ago, or not? And we can see that in the share prices. There have been a lot of reports that these very agents could lead to these SaaS providers no longer being the user interface, but just the backend - and the user interface being run by agents. It's a bit of an exaggeration at the moment, but there is some truth to it. There are processes that change quickly. They have a higher rate of change. And wherever I cannot deterministically predetermine the process, where it is subject to frequent changes, classic hard programming is not the right way to go. I think that agents, with their stochastic properties and flexibility, can better meet customer requirements. In other words, I think we will have a hybrid world in the future. We will still have processes that are relatively deterministic, that run the same way every month, that hardly change.

E3: And where do we find Agentic AI?

Jost: It makes no sense to shoot sparrows with cannons and say that this should all be agentic AI now. On the contrary, it would also be counterproductive because these agents still don't have the reliability we need. But in other areas where it's about innovation, where it's about speed, it can make sense. And then the question is: who builds these agents? Will SAP build them with its platform or will there be other platforms that can be used to build such agents? And I think SAP will certainly offer standard agents, i.e. predefined agents out of the box that can be customized a little. But then there will certainly also be a platform from SAP that can be used to develop customized agents that are not predefined as standard by SAP, so to speak. In my opinion, as a SaaS provider, you have to open up. The best platform or the one that fits best simply has to win.

E3: So is the future hybrid?

Jost: I don't think there is an all-encompassing platform that covers all customer requirements. We have seen this in the past with data, integration platforms, process automation, BPM platforms and so on. In other words, SAP has always had its own offerings. But you can't assume that you are the best everywhere. There will always be specialists who simply do certain things better. And we need this diversity. And that's why I believe that there will be a hybrid landscape. An SAP customer will use SAP for certain standards. But they will also have custom agents. And with custom agents, it depends on the customer's focus, from which side they approach the topic of agents. Depending on this, one platform may be a better fit than SAP.

E3: But when I look at the current development with the new paper from SAP on API policy, I see a monolithic trend, don't you?

Jost: On the one hand, I understand that SAP says we can't let just any developer with any platform develop against our APIs. You can also get up to a lot of mischief because an SAP ERP is an integrated system - it has integrated governance, it has an integrated data model. We can't play Wild West here and everyone does what they want. But the second step has to come. And the second step is for SAP to think about governance. Then the door has to open again, because the closed store for the first phase is okay. A „closed store forever“ would be an indictment and would be contrary to the customer's interests. You still have to say: the customer pays for the party!

E3: Many SAP users are committed to BTP - the pricing model is complex, isn't it?

Jost: I think, technically speaking, the SaaS providers will not disappear, because they have a huge customer base and this base has an integrated backend. What happens at the front end is a different story. But to the question „Will they survive or not?“: They will survive. Will SAP still have a 28 percent margin or will it only have 12 percent? I think that's the question of all questions. How strong is this ERP model once the agents are at work? How brand-relevant is this ERP business model?

E3: How do you sell your platform? Which license model do you use?

Jost: Our platform is used to develop process agents. In other words, we sell to software developers who build agents. We are one level lower. In other words, it doesn't affect us in this way, but primarily affects the providers of standard software, especially cloud-based SaaS applications.

E3: First of all, you mentioned a very important buzzword: Compliance. How do you deal with this so that AI agents don't go crazy?

Jost: There's a lot you can do. And I think that's also the main issue I see with these agents at the moment. How do I get them under control? If I don't know what the thing is doing, it's not feasible. And that's why there are many new approaches that serve to put these agents in their place. People are trying to make these non-deterministic systems more and more deterministic. That's a bit contradictory. There is context engineering. And a context window is limited. The more there is in it, the worse the result will be. The aim is to keep the context window as small as possible. It is one step towards improving the performance of the agents. The other is called harness engineering. This is a new term where we say that the models are not so important, but the ranking of the models is the important thing. (See also the box on this page.)

E3: What can be achieved with it?

Jost: MCP servers are the measure of all things. (Editor's note: A model context protocol server (MCP) acts as a standardized connection for AI. It enables agent-based LLMs to securely and dynamically access external tools, APIs and data sources without the need for a separate, manually programmed integration for each individual system). It will be exciting to see whether SAP will offer an MCP server, that is a big question for me. And then there are skills, normal text files where instructions are described. These text files can be loaded and unloaded. In other words, the skills: context engineering, harness engineering, etc. All of these things are used to capture the models so that they still retain their flexibility, but ensure that they don't do things that the ERP user doesn't want them to do. This may look fun, but nobody will tolerate it in their company. We need to get the AI models on track. But we don't want them to just run on rails, because then we might as well leave it alone.

E3: When I ask Dr. Wolfram Jost as a scientist: What chances does he give the methods of finding algorithms to be on track without danger?

Jost: I mean, what is most advanced is the code generation. These agents are the best agents there are at the moment. And they have managed - via harness engineering, via context engineering, via these skills - to run as if on rails, but still retain their creativity and their generation. You can also define governance in these skills. You can define company rules and tell the model: These are your limits. You can add a human touch, and when the time comes, the system will also ask. That's a spectacular development. We no longer have a developer who doesn't use a coding agent. I don't think that a bad developer becomes a good one through a coding agent. But I do think that good developers become even better developers thanks to the coding agent. If you have the talent, you can use these agents to become faster, more efficient and better.

E3: Thank you for the interview.