In 2023, chatbots answered questions. By 2025, AI agents can code and design entire applications and services from scratch, as well as do deep, nearly scientific-grade research on any topic. Now, as enterprises deploy armies of autonomous agents, a critical question emerges: How do we prevent these powerful tools from descending into chaos in the coming years? We at Trevolution chose not to restrain our ambition but redesign it instead.
Our own journey in developing AI in 2023 had a rocky start: We were building and testing a chatbot, Olivia, for customer support. It could answer simple questions — think along the lines of early ChatGPT functionality; nothing but a chatbot. It sounded good in theory; however, our market analysis indicated that the real-world application would have limited utility. Our analysis revealed that customers in travel don’t contact support to chat — they require specific actions to be performed. Industry experience had shown that customers typically expect support systems to handle actionable requests: rebooking flights, fixing reservations and processing ticket refund inquiries. However, Olivia functioned solely as a conversational chatbot and lacked the capability to execute these operational tasks, which can only be performed by trained customer service agents with appropriate system access.
Following this assessment, we decided to reorient our approach, focusing on internal AI applications: testing how Olivia could assist employees rather than customers. This approach also offered reduced complexity, more structured feedback mechanisms and a controlled operational scope. By late 2023, Olivia had been developed as an AI assistant with clearly defined responsibilities and demonstrated consistent performance in controlled testing environments according to established metrics, though we knew it was capable of so much more…