Junior employees tended to describe checking AI output less critically than their senior colleagues. That was one finding from Sona8’s interviews with several hundred knowledge workers at a professional services firm.
The team suspected this was not entirely about AI. Senior employees are used to reviewing work from other people, including junior colleagues, and may bring the same habit to a model’s output. It was an observation from interviews at one firm, not a general rule about seniority.
Start with the working day
The conversations covered employees’ regular tasks, which processes caused problems and where AI was proving useful. The agent asked people to explain how they worked, rather than simply whether they used AI.
Looking at the successes mattered as much as looking at the problems. An approach that worked for one group could help colleagues struggling with a similar task. The analysis gave leadership a basis for developing AI use cases and sharing practices already in use within the firm.
The difference in how employees checked AI output also helped shape plans for training. It gave leadership something specific to address beyond encouraging people to use the tools.
The people missing from the interviews
Anton Hantel, Sona8’s cofounder and CEO, saw the limits of stakeholder interviews while working at BCG. His digital growth and transformation projects included programmes affecting the daily work of more than 10,000 people. Yet the teams could interview only a small part of the workforce.
Anton describes the reporting problem through the image of a watermelon: green in the status updates, red underneath. A programme can look healthy in steering committee meetings while the people doing the work experience something quite different. He wanted to hear from more of those people without losing the questions and detail that make an interview useful.
Anton and Thilo Tamme, now Sona8’s cofounder and chief product officer, worked on the same transformation programme in Hamburg. They often discussed the problem while running after work. Thilo was interested in voice technology, while Anton kept returning to how they could interview more of the workforce.
Sona8’s voice agent asks how a process works, why a step is still done manually and who else is involved. It asks further questions when an answer needs more detail. The company then analyses the conversations to identify processes, problems and differences across teams.
From research to a company
After BCG, Anton began an MBA at MIT and Thilo pursued a PhD at the Technical University of Munich, studying voice technology and its applications. They worked with TUM colleagues Michael Saatkamp and Alexander Scheuer on research into conversational agents for qualitative interviews inside organisations.
Their contribution, AI-Native Qualitative Interviews At Scale: Leveraging Conversational Agents For Organisational Research, was published in the ECIS 2026 TREO collection.
As the team looked for pilot partners, prospective customers asked whether the approach could become a product they could keep using. Anton and Thilo initially worked on Sona8 alongside their degrees, then left their programmes to focus on the company.
Anton now spends much of his time supporting pilot clients, helping them decide what to ask employees and how to use the findings. He has also led much of Sona8’s work with investors and accelerator programmes. The company joined Y Combinator’s Fall 2026 batch and MIT Sandbox, MIT’s entrepreneurship programme (which has also supported Cursor, Fireflies.ai, Lightmatter and many more).
Thilo works closely with the pilot partners on the product. He leads how the agent conducts interviews, including how it listens, when it asks for clarification and how it follows up on an answer. Feedback from those deployments helps him decide what to improve next.
Madeleine Malmsten joined as cofounder and CTO after six years at QuantumBlack, McKinsey’s AI unit. She leads engineering and works with Jakob Schepers, Sona8’s founding engineer, to build on the platform he started. Her focus is on making the system’s different parts work reliably together and improving the product as the team learns from pilots.
Keeping an important detail from disappearing
Jakob built Sona8’s platform in the first months and remains responsible for it. His work spans architecture, deployment, enterprise integrations and data protection, as well as the AI engineering around the agent’s voice models and conversation analysis. Before Sona8, he worked in Volkswagen Group IT on systems used in vehicle production.
Much of Jakob’s work now focuses on the analysis. The interviews generate more text than anyone will read, and the most important observations are not always the most frequent. A recurring complaint may be routine, while a less common observation may point to a serious problem.
That distinction can get lost as transcripts pass through several rounds of summarisation. A detail may survive the first summary and disappear from the next. Jakob built the analysis to carry potentially important observations through that process rather than let frequency alone decide what reaches leadership.
Thilo addresses the same issue during the interview, when the agent needs to recognise that an answer deserves another question. When it is unclear whether an observation is isolated or part of a wider pattern, the team can explore it in later interviews.
A conversation still needs trust
Talking to an AI about problems at work can still feel risky. Employees may be criticising a process their manager introduced or admitting uncertainty. A machine on the other end does not remove the question of who will see their answers.
Sona8 says participation is voluntary. The founders report that some employees find the format easier to speak openly in than a conventional interview.
Employees still need to know who can access their conversations, what leaders receive and how identifying details are handled. Someone in an unusual role, or describing a distinctive incident, may be recognisable without being named.
What happens after the interviews
Sona8 plans to return to employees after an initiative and ask what changed. Over time, the company wants to combine those conversations with information from existing business systems, giving teams a more current account of how work gets done.
For the firm planning its AI training, the interviews provided examples of useful practices and a clearer picture of where employees needed support. A later round could help leadership understand whether those practices had spread and whether people had changed how they checked and applied AI output.
