Wonderful, the Amsterdam-based platform for autonomous enterprise AI agents, which secured $134 million in total funding, launched its Central and Eastern European hub in Zagreb in September 2025. The move resonated through the region, puzzling as to why they picked Croatia, Serbia, and Slovenia as their next hub.
Solving the last mile of AI — the cultural nuances
Most startups scale in English-speaking markets first, while Wonderful has decided to do the opposite. Is the Adria hub a “stress test”, we wondered. If they can solve the linguistic and cultural complexity of Serbian, Croatian, and Slovenian, is every other market easy by comparison? “We don’t think in terms of “easy” markets or stress tests. We focus on regions with real enterprise scale and operating complexity, and this is true of many countries around the world
outside of the US — including of course Serbia, Croatia, and Slovenia“, shared Bar Winkler, CEO of Wonderful.
That said, they announced plans to position the region as a proving ground for technology that captures local intonation, sentiment, and business context — innovations the company now exports to its global operations.
Vedran Bajer, who joined as General Manager for the Adriatic after a decade at Google and a tenure leading Microsoft Croatia, frames the move plainly. “It is more of an innovation engine, than a translation desk.”
“While many global providers treat the region as a secondary market, delivering models that might be technically accurate but lack natural rhythm, we use this region to solve the ‘last mile’ of AI.”
Engineers in Zagreb will develop technology that fixes what Bajer calls the “robotic” feel of AI voice agents. That work — tuning for intonation, cultural grounding, and informal nuance in languages like Serbian, Slovenian, and Croatian — now flows back to Wonderful’s global product to address similar challenges in other linguistically complex markets.
Meanwhile, Bar Winkler also points out that they have been very impressed by the regional innovation scene.“From my perspective, this is a region with strong operational builders and deep engineering talent, the type of people who aren’t just looking to start companies, but to scale them.”
You will have the opportunity to hear more about their approach and technology in Beograd, Serbia at unlockit conference, on 19th and 20th of February. But before that, we dive into more details on Wonderful’s path: investments, product, operations and future plans.
The key is in the grounding
Wonderful secured $134 million in total funding, a $34 million seed round followed by a $100 million Series A, within less than a year of its January 2025 founding date. Bajer called the capital “one of the largest early-stage investments in the European AI sector” and said it signals high conviction that the Adriatic region can become a global proof-of-concept for high-performance, localized AI.
That confidence is well-founded, given the investors supporting them: Index Ventures, Insight Partners, IVP, Bessemer Venture Partners, and Vine Ventures, funds that previously backed the likes of Revolut, Slack, Wiz, and Notion.
What are those investors betting on? Wonderful’s platform deploys enterprise-grade AI agents that handle customer service and back-office workflows across more than 60 clients, including banks, telecoms, and postal services.
They combine the AI platform with local deployment to deliver enterprise-grade AI agents to every market and every language. They do not build foundation models from scratch. “We don’t believe in building one massive model for everything,” Bajer said. “Logic is universal, and a model like Claude reasons just as well in Serbian as in English. The real gap is in the orchestration.”
Cultural Architects
Wonderful bridges reasoning with localized voice synthesis and cultural grounding, using smaller specialized models that reduce hallucinations by limiting agents strictly to verified client data. But how do they take into account that languages such as Slovenian and Serbian often suffer from data poverty, sparse training corpora and limited commercial datasets?
Bajer reframes: “The issue isn’t a lack of words, but a lack of grounding.” Instead of training models on generic internet data, Wonderful uses Retrieval-Augmented Generation (RAG) to feed agents the client’s specific rulebooks and databases in real time. For that reason, the company employs what it calls “Cultural Architects“, AI engineers who configure and update agents to understand local business ethics and informal nuances absent from public training datasets.
Messy reality of legacy systems is not a problem
This approach proves especially valuable when clients operate legacy infrastructure. “We embrace the mess because it represents the highest potential for ROI,” Bajer said. “Most AI companies walk away when they see 20-year-old systems, whereas we can have a fully functional, transactional agent live in as little as 72 hours.” Wonderful’s agents integrate directly into CRM, ERP, and core banking systems, executing tasks such as changing service packages or processing refunds within the client’s database rather than merely providing instructions.
How do they guarantee to a bank that their proprietary data isn’t leaking to competitors? Bajer points that data sovereignty is their absolute priority, “and we meet the rigorous security criteria required by global financial institutions, including ISO, GDPR, and SOC 2 Type 2 compliance.”
“Technically, every client operates in an isolated environment. We never train a “master model” on one bank’s data to benefit another. Your data stays within your perimeter to make your agent smarter and nobody else’s.”
Agent Builder powered by Claude
In January 2026, Wonderful also hit another milestone. They introduced Agent Builder, an autonomous AI agent that independently builds, tests, and refines other enterprise agents.
Powered by Anthropic’s Claude model, the system ingests enterprise documents (policy manuals, knowledge bases, call recordings) and iterates on agent design without manual prompt engineering. Winkler described the rationale in the official release: “For AI agents to deliver real impact, enterprises need to be able to build, evolve, and manage them continuously over time. The Agent Builder is a critical step toward making that possible — not just for us, but ultimately for the enterprise teams who will own and operate these systems day to day.”
The platform supports both technical and non-technical users through chat-based interaction, allowing teams to refine agents iteratively. As enterprises scale from piloting a few agents to managing networks, Wonderful aims to provide full lifecycle controls for tuning, testing, and expanding deployments.
Wonderful is moving fast to drive this change and adoption with enterprises, which reminded us it achieved a ~$700M valuation in less than a year. As AI bubble keeps forming, we wanted to hear their conviction: what do they think makes Wonderful AI a structural pillar of the new economy rather than a feature that Big Tech could replicate in a single update?
Winkler starts by noting “There’s a huge gap between what AI models can do and what actually works inside large enterprises“.
“We’ve seen most failures happen after the demo or pilot stage, when AI hits real workflows, legacy systems, regulations, languages, and human behavior, and begins to break. That is Wonderful’s specialty.
“We excel at getting agents running successfully, safely and at scale in production in some of the most complex, highly-regulated enterprises out there. It requires investing in deep local presence, and that’s something we’ve done in the Adriatic region and across every market we operate in.”
New vision for the service economy
Before the end of our conversation, we had to touch upon the future. Bajer outlined a five-year ambition: “We aim to be the strategic AI partner for all our clients, and an operating system for the entire service economy.”
The recently launched Agent Builder allows clients to use Claude to build entire networks of agents without coding. “In five years, the role of local workers will have transformed from operators to ‘Architects of Interaction,'” Bajer said. “Humans will design the strategy and empathy, while Wonderful AI executes the scale.”
Time will tell, but for now, one thing is sure: Wonderful’s CEE hub strategy demonstrates how underserved linguistic markets can drive technical innovation with global applicability. By solving orchestration, cultural grounding, and legacy integration in Zagreb, the company positions itself to scale localized AI across diverse, multilingual, and regulated environments, turning regional complexity into a competitive advantage.
