The most popular topics is about AI. We want a simple way to create the things we find hard to take a lot of time. Does it even make sense to learn SAP Integration or is it better to find something else.
Is AI going to disrupt SAP Integration work?
I got a question on if SAP Integration work is going to be replaced and if it still makes sense to learn SAP Integration. I am biased on the topic but let us try to explore it.
What is the job of SAP Integration developers?
An SAP Integration developers job is to create a way to understand business requirement around data processing and turn it into an interface that automatically can exchange data between the two systems. Interview the business experts to work on the integration.
Then it is about managing the integration and fixing problems that arise.
Never give up to resolve a problem that you experience and then find the skills to resolve it.
What is required for AI to work?
For the role of Integration Developer to be extinct, we would require that the AI become so good that an business expert can prompt the way to build good integrations that satisfies the requirement for the integration.
That business can monitor and report the bugs that arise when you explore more complicated test scenarios.
Current AI integration tools
There are current flows that people see probably more the consumer versions of integration like Zappier or n8n where you have a better integration with AI. In Zappier you can now create some integrations via their AI model, some parts does simplify the building of integration but you still get to the hard part around mapping. Zappier does not support the complex mappings(to my knowledge) that often exist in enterprise landscapes.
What is the hard part of the job of Integration Development?
- Understanding the business context
- All the missing requirements (it is easy to map Purchase Order Number, but the field that should be filed in if condition is S is something that nobody thing on)
- Find the missing rules that make it possible to have the mapping
- Building solid integration flows and mappings
- Support an integration where you will continues be getting new error because something else happens.
- Understanding API, authentication and fault processes.
- Optimize for performance, scaling and license cost
- Communication with stakeholders to ensure the are okay
- Exchange security material
- Understanding and fixing errors that are not well described just a 500 return code and you will need to resolve.
Where can AI help
iFlow generation or adding parts.
SAP have in the latest release added and option to generate iFlow content based on a description. It is a good showcase, but it still requires the developer to fill in all the challenging parts of the iFlow.
This flow does not match my iterative way of building iFlows, were I add some steps, test them and then validate the result. Then continue to add more components.
It will likely get a Joule component that you will be able to chat with that can add 2-3 filled in flow steps as you build the integration. But it is likely more complex to build from the SAP side.
Documentation
I think the best place would be to train AI in documentation interfaces, because then it would learn the different good patterns to use in certain scenarios.
The learning is then to feed it good examples of integration that support different use case patterns.
“Vibe” Groovy coding
This is like the easiest part because there are already some good models for building Groovy. We already have some of this in Figaf with our Groovy editor. The important part is to be able to check the result fast and call for a way to optimize it.
Moreover, we have some ideas on how it can be improved to make the process even faster of writing code.
Read the API
Reading and interpreting an API specification can be challenging. It is not always update or easy to understand. AI may be able to give some simple way to simplify reading it.
Processing errors
AI will be able to give advises on how to resolve some errors. Though the content may not be enough to let ti handle the errors self.
What is blocking easy AI building of integration?
There are some big challenges with terms of AI at the moment for it to talk the full job form SAP Integration consultants.
Mapping
I think mapping is going to be one of the challenging parts. Message Mapping is quite complex but I would not allow an AI loose and create a mapping for me. Building them correctly requires you to understand context and queues, and I think it may be one of the more complex criteria to create.
The MAG did show some promise but it is still an expert tool that require you know how to use the context of it well. It may be just as challenging as the message mapping to learn. It is building with classical statistic models which works well in B2B scenarios with EDI documents with a unified coding but not as much in other areas.
Business context
The business context is often spread between different partners that are required to work together and the integration developer have normally been the interpretation of this context.
Marketing vs Reality
As a person that both works with marketing and development is see a big difference in expectation vs reality. We see the post I build a full SaaS in 3 hours that I should for 10.000 USD. And think wow, is it that easy.
Then when I try to build an application it gives me a good start on building some application. Then at some point I hit a ceiling on what can be made and then you are stuck.
There is no real help. You need to find a real person that can help resolve the issue. This is also the challenge for integration.
Architecture
Good integration developers spend a lot of time on understanding the patterns they are supporting trying to build something that simplifies process over all. An example of this would be the Pipeline Concept that allows users to separate routing from retry logic.
Error-handling
When working on integration, the error is less than ideal it should be possible to understand the errors.
Data privacy
Data privacy is a hot topic for SAP customers they dont want their data in an AI. There can be some difference between data used in development vs real productive data. But then how is the monitoring done if it is done via an AI interface.
Normal 3-tier environment
It should be possible to handle the governance around the integration in a normal way. And control the integration. This is possible with SAP and better if combined with Figaf
Work with internal data sources
Many integrations are using internal systems that have been built using some database. It is a little challenging to figure out how this integration works in the real world.
Conclusion
There is a some way with many hurdles to complete before we get to a point where we there will not be a need for SAP Integration Developers. That is both if you use SAP Integration Suite but also other platforms that may have faster innovation cycles.
At the he moment I only see the mapping with XSLT/Groovy can be assisted in a good way with AI. And also with resolving errors.
The AI will be able to simplify some of the more labor intensive work. This will free up resources to work on more integration and work on higher level tasks. As with everything with AI, it may be difficult to see when something like this will change. And if you are good, you may be able to keep working in the field but with much better tools to automate the different tasks.