How Hostinger Went From Weekend Chatbot Experiment to 1 Million AI Conversations a Month

When Hostinger's AI Tech Lead Mantas Lukauskas first experimented with neural networks over a weekend to see if they could answer customer questions, he had no product roadmap in mind—just curiosity. Five years later, that experiment has evolved into Kodee, an agentic AI assistant handling over one million customer conversations per month, with an automation rate that climbed from 50% in January 2025 to 85% today.

Lukauskas shared the full journey at DevSparks 2026 Bengaluru, speaking with YourStory in a session titled "From Chatbot to Agent: Building AI That Can Run Infrastructure." The story is a practical blueprint for any business trying to bridge the gap between AI experimentation and production-grade deployment.

The Three Lessons That Shaped Kodee

Hostinger's path from chatbot to agentic system surfaced hard-won lessons that apply broadly:

• Never rely on a single model provider. AI providers experience outages. Hostinger—handling around 200,000 customer chats per week—built fallback mechanisms using open-source models and cloud-based routing. A single-provider dependency is a single point of failure.
• Cost discipline matters as much as capability. Lukauskas was direct: "If you want to kill a fly, you don't need a bazooka." Not every interaction needs generative AI. Identifying which tasks don't require it is as important as identifying which ones do.
• Keep humans in the loop intentionally. Hostinger gave customers the option to escalate to a human agent from day one. Some companies rushed to remove human support entirely and had to reverse course. Hostinger framed AI as expanding customer options, not replacing them.

Key Takeaways

• Agentic systems can remember context, access external tools, and complete multi-step tasks—but successful deployment requires infrastructure, latency management, security governance, and benchmark evaluation, not just model selection.
• Hostinger uses large language models as automated evaluators, scoring Kodee's outputs on accuracy, tone, and brand alignment.
• The biggest barrier to AI adoption, Lukauskas argued, is not technical—it's hesitation. Modern tooling has lowered the barrier to building and launching a prototype to the point where getting started is the most important step.

Read the full article on YourStory