SAP Joule: AI boost for BTP and Abap


I was recently able to take part in a hands-on workshop on Business AI at SAP in Zurich. A team of developers from SAP Labs India had flown in from Bangalore especially for the event, which was also attended by the CTO of SAP Suisse. Being able to receive input from top experts directly on site, discuss it with them and try out the new possibilities straight away was a special quality of this event.
AI setup via booster
The repertoire included ready-to-use services such as document information extraction, prompt engineering at the Generative AI Hub, the creation of an app in SAP Build Code with SAP Joule, SAP's in-house AI assistant, or the use of the SAP Hana Vector Engine to embed custom context for artificial intelligence.
What was impressive afterwards was how easy it was to use the services in our in-house BTP. Simply added to the BTP sub-account, they were easy to configure, in many cases even via booster, i.e. with guided set-up. A completely new quality compared to the usual configuration of services on the NetWeaver stack.
Abap gets smart, SAP Joule can do Abap
But not only the services offered in BTP are interesting, AI is also finally making its way into Abap development. Particularly exciting: with the Abap AI SDK, there is now also a native option in Abap for integrating AI models directly into any Abap application - both on S/4 Cloud Public Edition and S/4 Cloud Private Edition systems and the BTP Abap Environment. This makes it possible to create new, innovative applications as well as extend existing implementations. Last but not least, the skills of SAP Joule have also been expanded - the AI copilot from SAP can now also use Abap. Callable as a new view in the Eclipse development environment, SAP Joule is available to programmers as an Abap sparring partner.
Abap coding and CDS views can be generated, discussed and explained by SAP Joule using the chat function. SAP Joule can also be used to automatically generate the time-consuming unit tests that programmers are less fond of. This not only saves time, but also increases the robustness of the code.
Promising roadmap
A look at the development roadmap shows that the next few months will also be exciting. Extensive enhancements to the Abap functionality of SAP Joule are planned. On the agenda is the complete generation of transactional applications (data model, behavior and user interface) and analytical queries.
This also demonstrates the particular strength of SAP Business AI: it is the only AI that is trained with a focus on proprietary SAP data and the SAP architecture and therefore really "knows its way around" the SAP domain. On this basis, SAP Joule can also act as a coordinator and optimally dovetail AI agents, data and processes.
All of these features hold incredible potential for development in the Abap environment. Boring boilerplate code and test cases can be generated automatically in future and developers can concentrate on the more interesting aspects of their projects. Also with Advanced Business Application Programming Language, aka General Report Formatting Processor. Thanks to Joule & Co., Abap, which has sometimes been declared dead, is now also becoming smart.
To the partner entry:

1 comment
code quality guy
Zum Thema Joule und ABAP: Seien wir mal ehrlich: Ist denn die Testcode-Generierung wirklich das Problem gewesen bisher, warum Tests unbeliebt und selten waren? Oder ist es nicht viel mehr die Architektur eines SAP-Systems mit den ganzen historischen Altlasten und Quircks, die nie dafür gedacht und designt waren, automatisch testbar zu sein?
Wenn sich daran nichts ändert und Entwickler nicht umdenken und sich auf außerhalb der SAP Bubble längst etablierte Techniken wie bspw TDD und ernsthafte Softwarequalitätsstandards jenseits von naming conventions einlassen WOLLEN (anstatt halt in S4 so weiterzumachen wie sie es dir letzten zwanzig Jahre gewohnt waren nur mit Eclipse statt se80), wird mit historischem ABAP Coding trainierte AI auch nur das reproduzieren was schon in der Welt ist. Die Innovation und Wandel zu besserem und robusterem Code muss von Menschen ausgehen und sich in Trainingsdaten niederschlagen, bevor sie in generativer KI ankommt und das sehe ich ehrlich gesagt bei S/4 Projekten gerade nicht.
Also ich glaube, dass Boilerplate gut und schnell generiert werden kann – das allein ist ja schon ein Achievement. Aber wir werden auf absehbare Zeit keinen grundsätzlich besseren Code bekommen. Generative KI für Coding bedient sich für andere Sprachen an Open Source Projekten von denen die guten die Messlatte in puncto Coding Standards echt hochhängen und popliger Enterprise Code egal in welcher Sprache oder Plattform nie ranreicht. Wie will Joule also an gute Vorlagen kommen?