Capgemini exec shares lessons from SAP agentic AI projects
In this podcast, Capgemini VP Gianluca Simeone explains how it's possible to build useful, cross-platform agentic AI applications in weeks, despite a lack of industry standards.
Agentic AI is the hottest trend in ERP as vendors race to offer special-purpose AI agents, development tools for building custom agents and new data products expressly designed for AI.
SAP, along with ERP archrival Oracle, is leading this AI push, especially since the debut of generative AI (GenAI) three years ago. Building on earlier machine learning technology, it first added a GenAI copilot called Joule, later augmenting it with the decision-making and autonomy that define agentic AI. It then added Business Data Cloud for the cross-platform data integration AI needs to be effective.
The promise is AI that can anticipate users' needs, communicate with them in their native language and handle many of the tasks that otherwise require tedious data entry and switching between ERP screens.
But building agentic AI that works across SAP and non-SAP systems isn't easy. In this episode of Enterprise Apps Unpacked, Gianluca Simeone, vice president and chief technology and innovation officer at Capgemini, the French multinational IT services firm, shares lessons learned from practical experience, including the design process, development tools and challenges of integrating SAP AI agents as well as orchestrating their behavior.
Gianluca Simeone
Agents of change
Based in Milan, Italy, Simeone has more than 20 years of development experience on SAP platforms. He joined Capgemini in 2013 as an SAP enterprise architect and later worked on development teams focusing on major cloud platforms, including AWS, Google Cloud and Microsoft Azure. In recent years, his focus has been SAP AI projects.
In one case, Simeone's team reportedly took just three weeks to use Joule Studio to develop custom agents that work across SAP and non-SAP systems to automate the procure-to-pay process of an oil and gas company.
He said multi-agent orchestration, where agents from different vendors and platforms work together, is still under development as the industry works on protocols and standards. For example, Capgemini has a three-way partnership with SAP and Google to use the latter's Agent2Agent Protocol, enabling agents from both vendors to consume data and communicate with each other. They're experimental now but could be offered to Capgemini clients in a few weeks, Simeone said.
"The real value from agent-to-agent will come when we have SAP offering a list of agents to execute specific tasks and other players, like the hyperscalers or other vendors, releasing something similar, and having Capgemini as a system integrator working with our clients to connect the dots between the multiple agents to orchestrate end-to-end business processes," he said.
Other topics discussed in the podcast include the following:
- Whether nontechnical business leaders are using low-code and no-code development tools to play a meaningful role in building agents.
- The integration tools and standards used in the SAP agentic AI projects.
- What makes agentic AI projects successful.
David Essex is an industry editor who creates in-depth content on enterprise applications, emerging technology and market trends for several Informa TechTarget websites.
