Sofia becomes an incubator for SAP's autonomous AI agents
The company is stepping up its efforts in the field of artificial intelligence with a new AI Incubation team
While the mass IT sector is going through a phase of cooling off and rethinking its hiring strategies, SAP Labs Bulgaria is signaling the opposite. Unsurprisingly, the company's focus is on artificial intelligence. The latest development in this direction is the expansion of the capacity of its development center in Bulgaria with a new AI Incubation team.
Although small in size, it places an important emphasis on the next generation of enterprise software. It also highlights the role of the Sofia office as a key location for SAP.
We recently launched an AI Incubation team focused on developing AI agents to optimize SAP's cloud operations," Radoslav Nikolov, CEO of SAP Labs Bulgaria, told Economic.bg.
The newly formed AI Incubation team in Sofia is part of the global GCID (Global Cloud Infrastructure and Delivery) division and is the second largest after the one in Germany. Its task is key to the company's cloud infrastructure – developing autonomous AI agents to optimize SAP's complex cloud operations.
The company has not specified the size of the team, but its creation is a sign of intensified AI efforts.
From automation to autonomous "thinking": What is Sofia developing?
Several different AI-focused teams are already operating in Sofia. One is part of the internal AI Center of Excellence, focused on automating internal processes – from ticket processing to contract preparation. The other is building a platform for extracting and preparing business data used to train AI scenarios in various product lines, including SAP's own foundation model – RPT-1.
Radoslav Nikolov outlines the conceptual boundary that is key to understanding the company's market position. The traditional automation that SAP has been using for decades is based on predefined rules – each task requires explicit programming and the system does not adapt to new conditions.
AI agents work according to a different logic. They are designed to make decisions autonomously, choose tools and approaches depending on the context, communicate with each other, and learn from the environment in which they operate. The practical difference can be illustrated with a specific example from SAP's own operations: when a cloud service is interrupted – an event that can mean millions in losses for customers in a matter of minutes – a co-pilot developed by Bulgarian teams is now automatically activated, identifies the cause, and proposes solutions, involving specialists from neighboring teams if necessary. Response time is drastically reduced.
These agents are capable of making decisions, such as choosing the right tools or actions, solving much more complex tasks, learning from the environment in which they operate, and communicating with each other."
The long-term vision is more radical: SAP customers will interact only with the Joule co-pilot, while AI agents manage every business process and every transaction in the background.
As for the programmers themselves, Radoslav Nikolov believes that thanks to all the tools available, senior developers can free up a large part of their time spent writing code and redirect it to more creative tasks or to developing innovations.
Search for talent
Against the backdrop of a general slowdown in the IT sector in Bulgaria, SAP is advertising mainly senior-level positions – specifically seeking engineers with experience in Python and Large Language Models (LLM). It is telling that candidates are required to have developed at least one AI agent, even if only as part of a personal project.
Although the current vacancies are mainly for senior positions, in recent years we have continued to hire junior specialists, including through our SAP Next Talent program, which is aimed at graduating students. I expect new opportunities to open up this year as well, as our goal is to have a balanced mix of newcomers and more experienced colleagues," commented Radoslav Nikolov.
Translated with DeepL.