SAP unveiled GenAI features at TechEd, featuring SAP Build Code for pro-code development with Joule, its GenAI assistant. Also, SAP HANA Cloud now features vector engine capabilities, expanding GenAI possibilities for SAP and its customers.
Omdia view
Summary
At its annual TechEd developer conference, SAP rolled out several new generative artificial intelligence (GAI) features and supportive capabilities designed to accelerate customer innovation. First, following up on the 2022 TechEd release of SAP Build as a low/no-code development experience, the company introduced SAP Build Code, a pro-code experience powered by the company’s in-house GAI assistant, Joule. Second, the company added vector engine capabilities to SAP HANA Cloud, opening up a host of GAI opportunities for both SAP and its customers.
GAI app innovation
SAP is well-renowned for not just publishing but also adhering to a long-term software roadmap that brings new technologies forward in a very evenly-paced manner and at a predictable cadence—something crucial for a company with more than 22,000 global enterprise customers that use its cloud native and cloud-agnostic business technology platform (BTP).
This carefully managed approach to software delivery was very much on display at TechEd 2023 with the introduction of SAP Build Code, an artificial intelligence (AI)-powered productivity solution tailored to Java and JavaScript development of SAP data models, application logic, application test scripts, sample data, and UI annotations, all of which are generated with SAP’s new GAI copilot, Joule. SAP Build Code integrates with SAP Build, a low-code tool introduced last year at TechEd with the aim of helping non-developers build apps, design business sites, and automate business processes. SAP Build Code can work with pre-built integrations, APIs, connectors to both SAP and non-SAP resources, and templates, as well as productized SAP best practices.
Together, these two unified development experiences work in concert to accelerate innovation with a consistent set of tools that support SAP business users, IT shops, and independent software vendor (ISV) partners. For example, using a custom data integration pipeline created with SAP Build Code, SAP Build business users can quickly connect that pipeline with their own analytics reporting dashboards, all using the same set of underlying infrastructure controls (security, privacy, provisioning, etc.) found within BTP.
It is worth noting that SAP’s Joule is not just a basic code generation model tied to SAP Build Code. It has been built specifically to align with SAP’s view of corporate data, a view garnered from supporting more than 300 million business users worldwide. With this foundational understanding of business practices, data models, and other enterprise assets, it generates code and application logic based on SAP-centric programming models. Joule can return insights based on natural language interactions. It can also create data models and sample data that conforms to specific applications. Moreover, it can generate synthetic data, which is fast becoming a critical capability for companies seeking to innovate without exposing private or sensitive data.
Therefore, SAP’s Joule can serve as both a declarative code generation model and a generalized chatbot assistant for business users. It is not surprising then to note that SAP intends to make Joule available as an assistant with SAP SuccessFactors solutions and the SAP Start site later this year. Early next year, Joule will be available with SAP S/4HANA Cloud, public edition, as well, which will open up several interesting opportunities for SAP and the BTP ecosystem.
Do-it-all databases
SAP’s AI-fueled Build and Build Code tools together certainly demonstrate its desire to unify professional and citizen developers. However, that is only part of the tooling necessary to accelerate innovation in today’s enterprise. Enterprise practitioners seeking to put GAI into production have discovered an important truth over 2023.
GAI doesn’t run on massive amounts of just any old data; it runs on copious amounts of very carefully curated data. Whether in support of pre-training, fine-tuning, or in-context learning, as with retrieval augmented generation (RAG), companies have learned to prioritize access to high-quality data, particularly unstructured data such as text, audio, and images.
In the realm of large language models (LLMs), data is typically represented not as text but as a set of numbers, a vectorized representation of the original data, which can be readily consumed and understood by LLMs. More than that, these vectorized representations (also known as vector embeddings) can be used to find semantic similarities within huge amounts of unstructured text, which can then be fed to an LLM to equip the LLM with timely and contextual information that may not be a part of the model’s original training data (e.g., in context learning via RAG).
To support crucial GAI use cases, such as RAG, SAP announced at TechEd 2023 the addition of a vector engine within SAP HANA Cloud. This has become quite common across the technology marketplace, with most database providers adding in some degree of support for vector embeddings. With this announcement, SAP is taking an approach similar to that of rival Oracle, emphasizing the use of a single, multi-model database capable of supporting vectors alongside relational geospatial, graph, JSON document, and other data types.
Available as a native feature in SAP HANA Cloud, a database service in SAP BTP, this new vector engine will enable customers to use their own customer data within a secure, private framework to improve model output and cut down on quality issues such as model hallucination. More importantly, by taking a multi-model database approach to vector embeddings, SAP allows customers to combine both structured and unstructured data in support of both GAI and traditional (e.g., predictive) AI.
Why this matters
As evidenced by the announcements made by SAP at its annual TechEd developer conference, which was hosted live in Bangalore, India, the topic of GAI remains alive and well, steering the enterprise technology marketplace. And understandably so. With GAI expected to inject more than $3bn into the AI software marketplace before the close of 2023 (see Generative AI Software Market Forecast – 2H23 Analysis), technology providers have been eager to fill the market with new GAI models, development tools, and supportive capabilities.
Typically, such efforts have revolved around one of two approaches. Often in the form of a chatbot, companies build GAI into existing business software as a means of augmenting and often automating complex, language-based user interactions. Alternatively, companies will roll out development tools and supportive technologies meant to help customers and ecosystem partners build GAI-fueled capabilities—and do so rapidly.
Line-of-business vendors have focused on the former, while platform providers have emphasized the latter. SAP has chosen to take on both approaches simultaneously. The company’s introduction of SAP Build Code with a built-in code generation (Joule), considered together with the addition of vector support within SAP HANA Cloud, speaks to SAP’s genuine interest in helping companies harness the power of GAI to empower both developers and data professionals alike, using the same underlying capabilities built into BTP.
Why is this so important? According to a recent Omdia study of more than 6,000 enterprise practitioners (2024 IT Enterprise Insights Survey, Executive Summary), nearly half of all enterprises surveyed (over 47%) are already putting GAI into production or have already done so. If such early indicators are to be trusted, then it won’t be long before every developer becomes an AI developer and every business becomes an AI-assisted business user.
Appendix
Further reading
Generative AI Software Market Forecast – 2H23 Analysis (August 2023)
2024 IT Enterprise Insights Survey, Executive Summary (October 2023)
Author
Bradley Shimmin, AI platforms, Analytics and Data Management