SAP Joule Compliance

SAP Joule, my commands shall soon obey

Johann Wolfgang von Goethe’s ballad goes, "That old sorcerer has vanished, / And for once has gone away! / Spirits called by him, now banished, / My commands shall soon obey." According to SAP's vision, ERP users should control the Joule AI agent so it can perform tasks. The technology is well advanced. However, a framework for organization and control is missing. An AI framework that can control and orchestrate AI agents is needed.

Various providers are currently experimenting with AI platforms for intelligent agents. Some proposals suggest controlling and managing future AI agents similarly to employees, with a system called AI-ACM (AI agent capital management) as opposed to HCM (human capital management). However, this approach falls short when it comes to liability and compliance issues.

ERP compliance is not child's play

One of the major problems in computer science is providing mechanical and mathematical proof of the correctness of algorithms. Although a financial process can be tested millions of times with millions of pieces of data, it is still impossible to say with absolute certainty whether the ERP algorithm is correct. In this context, reference should be made to Sir Karl Popper and his falsificationism. Falsifiable statements are claims about empirical facts, such as an ERP algorithm. Therefore, like theories, they cannot be definitively proven. According to Popper, a finite number of observations (iterative tests in an ERP system) cannot logically validate universal statements, the core of scientific theories. His alternative proposal is falsificationism: theories can be proven or falsified through empirical testing but never verified.

This means logic and mathematics cannot be used to determine the correct, rule-based behavior of an accounting system. A human must issue a system a certificate in accordance with the general rules to certify compliance.

Where does the AI agent stand in this compliance cosmos? Is it a passive component that must undergo a compliance check, or will there be AI agents that actively assist artificial colleagues with their work? This is probably similar to autonomous driving. Who has the ultimate responsibility?

Competition between AI agents

Regulating and orchestrating the compliance of an AI system, such as SAP Joule, is not the end of the discussion. It is likely that numerous AI agents from different providers will be present in composable enterprise systems.

The concept of a composable enterprise has been discussed for many years. This ERP discourse is similar to the decades-long discussion about artificial intelligence. It was only with the widespread availability of powerful hardware that the leap from theory to practice was made. Neural network models were brought to life through the widespread use of graphics cards. Through back-referencing, AI systems improved and evolved into large language models (LLMs).

Similarly, the composable enterprise could become part of everyday ERP operations through the widespread use of AI agents. However, an ERP system is more than just a collection of algorithms and data; it is also an entity that is evaluated for consistency and traceability. Having conflicting AI agents from different providers in a composable ERP would be a major disaster.

In practical terms, the following question arises: can the Salesforce AI agent perform ERP tasks, or must it remain in the CRM area? Which system is leading? People have been asking this question for decades, and there is still no answer. AI agents probably won't find a solution either. Compliance with AI agents will be necessary!