Agentic AI Platform for SAP

The reality in the data centers and IT departments of existing SAP customers often resembles an archaeological excavation site in which a massive network of technical legacies has accumulated over decades. In the golden era of the R/3 and ECC 6.0 systems, the Abap programming language gave companies unlimited architectural freedom, allowing them to dig deep into the standard code of the SAP ERP software and adapt the system precisely to their individual business processes using Z programs, modifications and highly specific in-house developments.

What was celebrated by CCC managers, CIOs and board members as a competitive advantage to stand out in the global market has turned into a leaden and toxic legacy in the face of the unstoppable digital transformation and the approaching end of maintenance for old SAP systems. These historically grown SAP monoliths are interspersed with millions of lines of customer-specific code, up to eighty percent of which is no longer used in day-to-day operations, but still causes immense maintenance costs, jeopardizes system stability and poses serious security risks. The mandatory switch to the new, in-memory-based ERP generation SAP S/4 Hana and the associated strategic direction towards the cloud now require a radical paradigm shift that presents IT managers with financial, technical and organizational challenges. In this tense market environment, the software group SAP has proclaimed a new development doctrine that dominates the entire market under the term „Clean Core“ and is forcing customers to fundamentally rethink their IT architectures that have evolved over decades.

„SAP landscapes are never finished,“ explains Professor Alexander Zeier, co-inventor of the SAP Hana database and co-founder of Nova Intelligence. Regulatory requirements, new business models, functional enhancements, release changes - the question of how to keep custom code clean, maintainable and close to the standard does not arise once, it arises continuously, every quarter, for years. „Nova is designed precisely for this life cycle: as a permanent platform for analysis, documentation, fit-to-standard assessment, clean core governance, new development and continuous optimization,“ emphasizes Zeier and says: „If you only use Nova for migration, you only use a fraction of it - the platform accompanies custom code from analysis to modernization to new development. This is also reflected in our PAC ROI calculations. In the pure migration scenario and with the continuous further development of the S/4 environment, we can already demonstrate ROI values. If you combine both scenarios, you get ROI values of around 1400 percent in an exemplary deployment scenario with around 100,000 custom code objects. This is not a project ROI. This is a structural, long-term economic advantage.“

"SAP landscapes are constantly changing. Nova is designed precisely for this life cycle: as an agentic AI platform for analysis, clean-core governance, new development and continuous optimization. dard, clean-core governance.“

Prof. Dr. Alexander Zeier,
SAP Hana co-inventor and
Nova Chief Scientist

The concept of the clean core is tirelessly praised by SAP marketing strategists as the ultimate promise of salvation for the future of corporate IT, but for many existing SAP customers it represents a far-reaching transformation requirement whose challenge lies less in the target image than in its practical feasibility. Essentially, the clean core strategy means that the digital ERP core of S/4 must remain absolutely pure and untouched; any direct modifications to the standard code, unauthorized database access or in-depth interventions in the old Z namespace are strictly prohibited. In-depth interventions in the classic Z namespace are replaced by the modern Abap Cloud model.

The philosophy behind Clean Core aims to make the software low-maintenance through standardization, ensure seamless and automated upgrades in the cloud and keep the system agile enough to continuously absorb new innovations, especially in the field of artificial intelligence, smoothly. In-depth interventions in the classic Z namespace are being replaced by the modern Abap cloud model. Alternatively, so-called on-stack extensions are only permitted via the highly regulated Abap Cloud model - known in the community as steampunk - and via officially released, upgrade-stable APIs (Application Programming Interfaces). For existing SAP customers, this requires a fundamental realignment of existing business processes and system architectures.

The harsh reality of clean-core transformation in connection with brownfield, greenfield or a selective S/4 conversion approach forces a gigantic refactoring project in which the entire historical code base has to be painstakingly analyzed, cleaned up and in many cases completely rewritten, which puts a massive strain on chronically overburdened IT departments in terms of personnel and finances. Existing SAP customers must also bear in mind that refusing to follow these clean-core principles will inevitably lead to a technical dead end.

The roadmap for an S/4 conversion has so far included brownfield, greenfield and a selective data approach. Is Cleanfield - derived from Clean Core - an adequate term for another approach? „Yes,“ answers Professor Zeier in the E3 interview. „And it precisely describes what is practically possible for the first time with Nova.“ Brownfield is fast and attractive in the short term, but transfers technical legacy into the new world. Greenfield promises the architecturally cleaner target environment - but often fails in practice due to costs, time pressure and the fact that business-critical differentiation logic must first be fully understood and then rebuilt manually. „Cleanfield describes the operational implementation approach for realizing the clean core target image in a structured and scalable way,“ defines Emma Qian, Nova Chief Executive Officer (CEO), and she explains: „The existing landscape is neither simply adopted nor discarded across the board. Instead, it is systematically analyzed: What is obsolete? What does the SAP standard already cover? What is real, differentiating business logic that needs to be redeveloped in line with Clean Core?“

„The result of Cleanfield is an S/4 system that is lean, standardized and upgradeable from the outset - without losing any business-relevant features,“ says Nova CEO Emma Qian and emphasizes: „Cleanfield is not a marketing term. It is a description of a method that could not be implemented in this form without Nova.“

Brownfield inevitably means that the immense technical potential of the Hana in-memory database is not nearly fully exploited. A lift-and-shift is technically feasible, says Hana co-inventor Alexander Zeier - but it would be the most costly decision a company could make in an S/4 project. „Because it would mean adopting exactly what S/4 Hana is supposed to overcome: decades of technical legacy, poorly documented in-house developments and outdated architecture patterns,“ explains Professor Zeier. He once co-developed the HANA database with Professor Hasso Plattner at the HPI Institute at the University of Potsdam.

Custom code versus standard

„What our customers notice again and again in practice: A large part of the custom code inventory is either no longer actively used or could be covered by SAP standard functions - without any in-house development,“ says Sam Yang, Co-Founder Nova Intelligence, describing the current situation in the SAP community. With a lift-and-shift, existing SAP customers transfer all the ballast to the new world and continue to pay for it for years to come in the form of unnecessarily increased maintenance effort, increased testing effort with every release change and significantly slower implementation of innovations. „The S/4 Hana migration is perhaps a unique opportunity to clean up this inventory intelligently. Anyone who misses it with a technical lift-and-shift will regret it in the medium term,“ emphasizes Sam Yang emphatically in the E3 discussion.

On the diametrically opposed side is the greenfield approach, which advocates an uncompromising new implementation of SAP software on a greenfield site. Nova Intelligence enters this field of tension between brownfield legacy and the often cost-intensive greenfield restart with its cleanfield approach - a superior migration method that is only made possible by Nova's agentic AI platform and also supports custom code throughout its entire lifecycle. Nova Intelligence is backed by outstanding minds in the IT industry, above all Alexander Zeier, who invented the disruptive in-memory database SAP Hana together with Hasso Plattner and thus has an unparalleled in-depth understanding of SAP's architectural DNA, and Emma Qian, a former AI researcher at Google DeepMind, who brings cutting-edge expertise in the field of artificial intelligence to the company.

„Nova is not positioned as a competitor to SAP's own technology platform, but as a complement.”

Emma Qian,
Founder,
Nova Intelligence

„As a co-inventor of SAP Hana and as someone who worked for many years as CTO of the SAP unit at Accenture, I know the problem from both sides: I helped lay the technological foundation of S/4 Hana - and I experienced first-hand how deeply and stubbornly entrenched the problem of grown custom code is in large SAP organizations,“ says Professor Alexander Zeier, describing his motivation. According to Alexander Zeier, Chief Scientist at Nova Intelligence, the decisive insight was: „This problem cannot be solved with traditional tools because it is not a purely technical problem, but a semantic one. Custom code is not just code - it is solidified corporate knowledge that has been stored in lines of program code for decades and that hardly anyone understands in its entirety.“

„The moment Nova became feasible for me was the observation that large language models have reached a level of maturity in the past two years that allows precisely these semantics to be tapped into by machines for the first time,“ says Alexander Zeier, explaining the history of Nova Intelligence. „My co-founder Emma Qian brings this AI excellence directly from Google DeepMind, Sam Yang the engineering depth. The combination of the deepest SAP architecture knowledge available and state-of-the-art agentic AI technology - that's Nova. And there is no other combination like it.“

Cleanfield with Agentic AI

Nova Intelligence is an agentic AI platform for SAP that covers a wide range of application areas across the entire custom code lifecycle. In view of the 2030 migration deadline, the S/4HANA transformation is currently the most urgent field of application - and this is precisely where the Cleanfield approach developed by Nova comes in as a superior migration method that can only be realized with Nova's agentic AI platform. Cleanfield combines the pragmatic advantages of brownfield with the architectural quality of greenfield: the existing, historically evolved SAP landscape is neither blindly copied into the new world nor broadly discarded. Instead, autonomous AI agents analyze the entire custom code inventory systematically, semantically and structurally, eliminate superfluous or obsolete code, maximize the use of the SAP standard and recreate only those software artifacts that represent real value-adding business logic - in a Clean Core-compliant manner in the new target architecture.

No abap code in ChatGPT

„The difference lies in how deep the AI can actually go,“ says Zeier, defining the approach. Today, basically anyone can already insert Abap code into ChatGPT or Claude and get a useful suggestion back. This is exactly what AI-assisted conversion is at its core: a slightly better packaged version of this interaction pattern. Classic AI co-pilots provide selective support with analysis and suggestions - Nova, on the other hand, carries out the entire process from analysis to decision-making to implementation end-to-end. „But everything that really counts - understanding the business logic behind the code, determining whether the SAP standard already covers it, tracing dependencies system-wide, rebuilding it cleanly, testing it and putting it into a transport - remains manual work,“ says Alexander Zeier from his observations in the SAP community.

"Our platform not only understands Abap, but also the architecture of S/4 and the logic of the Business Technology Platform.“

Sam Yang,
Co-Founder,
Nova Intelligence

„Nova has a different architectural structure,“ emphasizes CEO Emma Qian and her colleague Sam Yang adds in the E3 interview: „We have created the infrastructure that enables AI agents to work effectively, securely and reliably throughout the SAP system. For us, these are the basic building blocks. They make the difference between an AI that can talk about SAP and an AI that can actually work in SAP.“

Nova Chief Scientist Zeier explains what this means in practical terms: „You can give Nova a broad task - such as ‚help me understand how our order processing works and whether we still need these customer-specific enhancements‘ - and Nova can actually perform this task.“ The platform examines the system, tracks the relevant programs and dependencies, evaluates them against the SAP standard and comes back with a substantive answer. „With a code assistant, you would first have to know yourself which programs to look at, which lines to question and which API to check. Nova can do this and many other tasks end-to-end,“ emphasizes Sam Yang emphatically in the E3 interview. The IT market is flooded with generative AI models such as ChatGPT, Claude or Gemini, and developers can copy a snippet of Abap code into such a chatbot to generate a suggestion for a syntactic correction or modernization. However, this form of AI-assisted code conversion only scratches the absolute surface of the problem and remains a manual, tedious and error-prone process at its core. A generic language model, as powerful as it may be in the general understanding of text, must inevitably fail in the highly complex, proprietary world of SAP, as SAP is not a generic IT system. The SAP ecosystem is characterized by specific transport mechanisms that have evolved over decades, a unique authorization logic, proprietary development paradigms and a deep, confusing context dependency between thousands and thousands of tables, interfaces and business objects. If you do not understand these interwoven structures from the ground up, it is impossible to generate software code that is executable, secure, up-grade stable and, above all, clean-core compliant in an S/4 or BTP environment. Nova Intelligence differs radically from this, as it was architecturally designed from the outset exclusively for this specific SAP context.

S/4 architecture principles

Nova does not act as a passive chatbot waiting for isolated prompts, but as an autonomous architectural partner. Sam Yang: „Our platform not only understands Abap syntax, but also the architectural principles of S/4 Hana, the extension models of Abap Cloud and Rap (Abap RESTful Application Programming Model), the integration logic of the Business Technology Platform. And it knows when which technology is the right answer. This enables decisions that a generic coding assistant simply cannot make: Is this extension true differentiation - or has it long been included in the standard? Can it be rebuilt to conform to Clean Core - and if so, how? It is precisely this decision-making intelligence that is the economic lever.“

The operational excellence and invaluable value of Nova Intelligence unfolds in a three-step process that gets to the root of the problem of technical debt, starting with the AI Code Intelligence module component. In most companies, there is a frightening lack of transparency about the actual state and use of custom code. Nova not only analyzes this uncontrolled growth statically, but also captures the semantic execution logic and the hidden business intent behind the code and translates these findings into structured, detailed documentation that corresponds to the level of a professional specification.

The new foundation of knowledge

The second building block, „AI Fit-to-Standard“, which represents the greatest economic lever in the entire SAP transformation, operates on „AI Code Intelligence“ - the newly created knowledge foundation. The AI agents systematically and unerringly compare the reconstructed business logic with the current capabilities of the S/4 standard. Agentic AI identifies with analytical precision those code fragments that are obsolete or whose function can now be covered by the SAP standard out of the box, and makes well-founded recommendations for the rigorous elimination of this ballast.

Finally, the „AI Build“ component comes into action for the indispensable extensions that ensure competitive differentiation. Nova does not simply generate abstract code snippets, but develops completely new, clean-core-compliant code fully automatically, which respects the clear separation between the untouchable SAP core and the flexible extension layer. The platform is a master of the entire range of modern SAP architecture: it generates on-stack extensions in Abap Cloud (embedded steampunk) or builds decoupled side-by-side applications for the Business Technology Platform (SAP BTP) using the Cloud Application Programming Model (Cap) in Node.js or Java.

„Nova is constantly learning,“ says CEO Emma Qian, describing the company's own approach. „But not in the way you would typically imagine. We don't train or fine-tune models on customer data. Instead, we build a structured, customer-specific knowledge core that grows with every interaction.“ Agent skills at Nova are modular knowledge units - i.e. architecture patterns, naming conventions, documentation standards, validated APIs as well as decisions made and their rationale. „This knowledge is persistent, can be shared within the team and is continuously refined,“ adds Sam Yang.

„The cumulative effect is considerable,“ Alexander Zeier knows from successful pilot projects. „On day one, Nova already knows the SAP standard in depth. After three months, Nova also knows your landscape - your business processes, best practices and architecture decisions. After twelve months, Nova has gained institutional knowledge that was previously held by a few key people.“

The long-term significance of Nova's Agentic AI system for existing SAP customers cannot be overestimated, as it goes far beyond the management of a one-off S/4 migration. Anyone who makes the mistake of viewing Nova merely as a temporary conversion project completely fails to recognize the profound, sustainable value of this technology. SAP landscapes are never a static end product; they are living, dynamic systems that are subject to constant regulatory adjustments, new business requirements, technical innovations and regular release changes.

For CIOs and CCoE managers

Nova enables CIOs and heads of Customer Centers of Expertise (CCoE) to no longer passively endure their custom code inventory as an incalculable risk, but to manage it actively, transparently and economically. The permanent guardian function of the AI agents ensures that the SAP system remains permanently lean, standardized and fully upgradeable even after the successful introduction of S/4 Hana, making the notorious phenomenon of creeping system contamination through uncontrolled custom developments a thing of the past.

The question „How many platforms does an existing SAP customer need?“ is justified because many CIOs have the feeling that every new generation of technology means another platform, says Alexander Zeier in the E3 interview. „Nova is deliberately not designed as another platform in the traditional sense, but as an intelligence layer on top of what already exists - for analysis, fit-to-standard, clean-core governance and also for the clean-core-compliant new development of future extensions.“

Clean Core-compliant extensions can be developed directly in the SAP system via Abap Cloud and Rap - or side-by-side on the SAP Business Technology Platform. Emma Qian: „Nova decides which path is the architecturally correct one, generates the corresponding code, tests it and prepares the transport requests. The customer is always responsible for triggering the transports and releasing them into the production systems. Nova fits into the existing governance structure instead of creating its own parallel world.“

A key unique selling point that will determine the future of Cleanfield and Nova Intelligence is the systematic development of the aforementioned company-specific „AI Knowledge Core“. The Nova agents are not static algorithms, but learning units; although they are explicitly not fine-tuned with sensitive customer data in order to guarantee maximum data sovereignty and compliance, they build up a persistent, structured understanding of the specific architecture, naming conventions, validated APIs and business logic of the customer's respective SAP landscape with every interaction and every problem solved. „This continuous increase in knowledge leads to an exponential increase in developer productivity: tasks in requirements clarification, manual code analysis and routine development that previously took months are now compressed into a few days or weeks,“ explains Alexander Zeier.

In summary, it is clear that the future of SAP landscapes, the innovative strength of companies and the successful adaptation of artificial intelligence are inextricably linked to the clean core paradigm and the compelling necessity of agentic AI. SAP's clean core dogma is architecturally correct and there is absolutely no alternative for the path to the cloud or for the use of generative AI, but it is simply not economically feasible for existing customers, who are crushed by technical debt, using the traditional methods of IT consulting and manual programming.

Nova Intelligence's Cleanfield approach brilliantly solves this seemingly insurmountable complexity problem by using artificial intelligence not as a cosmetic chatbot sugar-coating, but as a deeply integrated, autonomous software engineering tool. AI agents become the tireless architects and guardians of the clean core, doing the archaeological heavy lifting of code cleanup, salvaging valuable company knowledge into structured documentation and orchestrating the seamless, secure connection between the untouchable SAP transactional core and the agile innovation layer on top of the BTP.

„The development in agentic AI is rapid,“ Emma Qian, Sam Yang and Alexander Zeier also know, and Nova is built in such a way that each new generation can directly benefit this technology. „The platform is deliberately designed to be model agnostic,“ explains CEO Emma Qian. „New generations of models can be integrated without affecting governance, operating model or customer data sovereignty. This is a strategic architectural decision - because no company should be tied to a particular model today if the performance of the models improves fundamentally every year.“

Semantic code analysis

„What exactly do I expect?“ says Alexander Zeier. „The precision of semantic code analysis will continue to increase. The ability to penetrate complex dependencies between thousands of SAP objects will improve significantly. And the interface between business requirements and technical specifications - currently still a major bottleneck in almost every SAP development project - will be largely automated by Nova over the next few years. Companies that invest in an AI-ready S/4 landscape with Nova now are laying the infrastructure for an SAP world in which innovation cycles that take months today can be implemented in weeks.“

For existing SAP customers, the use of an agent-based AI system such as Nova is therefore not a technical luxury, but the operational basis on which they can set up their system landscape to be future-proof and AI-ready.