“Build AI that knows your business” is a tag line SAP uses in its online collateral, and AI tooling with business context sits at the core of several announcements SAP made at TechEd 2026 (Berlin, Oct 27-29). We came away with the impression that SAP’s most powerful differentiators are at risk of being obscured by a wall of AI capability announcements that – while impressive – promise more than they can deliver today.

Here’s what impressed us most, and why:

  • A strong focus on process understanding. SAP repeatedly emphasized the need for process understanding and a process mindset to achieve optimal AI results. The SAP Signavio Process Consulting Agent and SAP Knowledge Graph reinforce that emphasis.
  • Updates to the tabular AI model, SAP-RPT. The existing SAP-RPT-1.5 is a relational pretrained foundation model specifically developed for prediction tasks on structured and relational data; it supports classification and regression. Announcements included SAP-RPT-2.0, which will support anomaly and outlier detection; SAP-RPT-TS will add time series. Providing a tabular model besides standard LLMs is extremely valuable for client use cases that involve number-crunching tasks while leaving their annual token budget almost untouched. We regard this as a differentiator for SAP.
  • Guided agent-building to ensure grounding in enterprise context. Joule Studio provides a guided experience for building agents. The generated agents are grounded in companies’ operational knowledge by seamlessly harnessing information from the SAP Knowledge Graph, LeanIX data, and contextual ERP data. Such agents can provide more accurate results than those built on a generic agentic platform using only intent, constraints, and the usual dimensions.
  • An expanded SAP Knowledge Graph. The first iteration of the SAP Knowledge Graph essentially represents SAP’s domain model, managed by SAP; it captures 50 years of business knowledge. This in itself already helps ground AI in the right data, but isn’t quite enough. Hence, SAP is adding the ability to bring a customer’s own knowledge graph into the SAP-managed environment. For customers wanting more control, SAP is also adding a customer-managed customer knowledge graph. Note: SAP also offers SAP Domain Models, which are foundation models enhanced with SAP’s proprietary and business logic.
  • Turning headless isn’t part of the plan. While some software platform vendors promote their suites becoming headless as a way to streamline AI access, the SAP team intentionally decided against this trend, arguing that they want to provide the best experience for any user type, from business user and business developer (where WalkMe is a key enabler) to professional developer and AI agent. SAP offers headless as an option, but doesn’t intend to replace other experiences.

TechEd is a developer conference, so expecting too much business-focused content would be unfair. That said, we see a few areas that deserve more attention:

  • Customers need more clarity on how everything fits together. SAP itself is on a major technology transformation journey and offers a rich portfolio of existing and new solutions. Not all customers are ready to embrace a “Business AI Platform”, particularly when they can’t immediately tell what is is evolutionary and what is revolutionary.
  • Joule Studio and Joule Work could benefit from some design improvements. This isn’t to distract from the tremendous amount of work that’s gone into these tools, and how powerful they are. But if some of these capabilities are to be put into the hands of business people, a few tweaks would improve usability for that target audience. The current design can make it challenging for the human approving an agent or agentic workflow to know what exactly they are approving.
  • The positioning of AI risk needs more nuance. SAP has a strong focus on AI safety and security before releasing software, whether it’s a new version of a product or a single AI agent. Its focus on governance capabilities is also laudable. What’s missing is a more discerning approach to AI risk. “Human-in-the-loop” cannot be the stock response to the perennial disclaimer “check for accuracy”; human-in-the-loop can neither scale nor guarantee correctness. What’s being done technically to minimize error and to make it easier for the human to ensure accuracy? In our view, neither aspect received the attention it deserves.

“The Autonomous Enterprise Becomes a Reality” is an aspirational statement that matches SAP’s ambitions, and the product announcements at TechEd, both GA and roadmap, reflect those ambitions. Companies ready to transform their business, and whose center of gravity lies with SAP, should ask hard questions about product capabilities and cost. But they should also remember one important point: Nobody knows SAP better than SAP.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Share