Joule, the SAP agent thriller

Drawing the line between operational ERP and GenAI Agents

The Austrian philosopher Ludwig Wittgenstein (1889-1951) is credited with the following quote: The limits of my language are the limits of my world. What may sound harmless or even interesting from a philosophical point of view could be revolutionary or even dangerous in the field of GenAI, LLM and AI agents. Why?

With the current SAP vocabulary, i.e. the language of the SAP community, ERP is built and operated sequentially. S/4 is better than ECC 6.0 and this ERP is better than R/3. The SAP user orchestrates. The ERP system works. There is an extensive division of labor between man and machine, which is also defined by the work and specifications of SAP.

In an agile, open, composite ERP that is managed by AI agents, the division of work and tasks is no longer clear, and is probably no longer reversible. AI agents do not automate, but should orchestrate independently. With this instruction, AI agents will very quickly exceed the limits of our ERP language.

In a composable ERP, AI agents can develop independently and dynamically. To paraphrase Ludwig Wittgenstein: AI agents independently learn new ERP languages (consolidation, harmonization and orchestration) and build a composable ERP far beyond our own borders. Perhaps SAP will become the new DeepL of the ERP scene with Joule and GenAI Hub from BTP.

SAP playground: Composable ERP

SAP has already spoken often and intensively about the intelligent enterprise in the past. SAP provides many IT tools for the implementation of an intelligent business structure and process organization: the Hana engines Graph and Vector, BTP GenAI Hub, Joule, etc. What SAP has not yet done is define positions and assign tasks: What should and what are AI agents allowed to do? Where should and may LLM, large language models, look? Are there limits for the GenAI Hub in Composable ERP?

Many years ago, AI researchers experimented with neural networks and the well-known board game Go. Initially, the neural network (nowadays we would probably call it LLM) was familiarized with the rules of Go and trained with the existing Go literature. Soon afterwards, the Go computer was playing much better than the world's best Go player. The machine played moves that Go experts considered to be absolutely absurd and counterproductive - but in the end the machine always won.

The limits of my language are the limits of my world: in a second round, the AI researchers fed the neural network only the fundamental rules of Go and let the machine play against itself continuously. The Go computer learned and perfected its Go language. In doing so, it crossed boundaries that put the entire Go literature of the past thousand years in the shade. Through back-reference, the Go computer left human boundaries far behind.

Has SAP also considered this option when AI agents move, develop and optimize in a composable ERP? Theoretically, SAP Joule could use the building blocks of a composite ERP to develop a whole new business management system and new forms of organization.

IT in general and ERP in particular are in the midst of an agent thriller. AI agents such as SAP Joule have arrived to orchestrate ERP work and optimize the business structure and process organization. The agents create the composable ERP. Further development, research and version changes are carried out by AI agents. Humans will hopefully remain beneficiaries or just spectators.