In this episode, we talk to Seb Hapte-Selassie, co-founder and CTO of telli, a Berlin-based startup building AI agents that run a company's customer-facing operations. Seb studied computer science at Stanford and built product at N26, Pitch and BCG Digital Ventures before starting telli, which went through Y Combinator and recently raised a $15 million seed round led by Redalpine. telli's voice agents already handle millions of conversations for companies like Sky, Enpal and Vaillant. We start with the person he credits most, his mother, who fled Ethiopia, made it to Europe and went on to Harvard Business School. Seb describes her as the most high agency person he has ever met, someone who simply did not accept that things were impossible, and he still catches himself asking what she would have done in a given situation. He also reflects on what surprised him at Stanford, where professors gave students the benefit of the doubt rather than trying to make them fail, and where he learned that the people building huge companies are not from another universe. A neighbor in his freshman dorm casually mentioned he had just raised half a million dollars from Y Combinator, which was the moment startups stopped feeling abstract. On finding co-founders, Seb is clear about what actually matters. The idea matters somewhat, the market matters somewhat, complementary skills matter somewhat, but what really counts is whether you enjoy spending enormous amounts of time together, because everything else can change. He and his co-founders Finn zur Mühlen and Philipp Baumanns worked on projects for six months before committing, and used explicit time boxes rather than open-ended deliberation. They agreed to work together until October 2024 and then reevaluate. By that point they had the idea for telli, their first three customers and a place at Y Combinator. We get into how the product actually works. telli started with energy companies, an insight that came from his co-founders scaling a 150 person phone team at Enpal, and has expanded into pre-qualification, reminder calls and inbound support. Under the hood it is a harder engineering problem than it looks. telli runs seven models in parallel in a single conversation, falls back automatically when a provider degrades, and switches models mid-call for tasks like address collection, where LLMs struggle. Seb also introduces Charlie, their new agent for customer-facing teams, which lives in Slack, Teams, HubSpot and Salesforce rather than inside telli's own app. On the business side he is unusually direct about limits. telli does not do cold calling and turns down the revenue that would come with it. He tells us about the call where an agent stayed patient with a furious customer long enough to calm them down, and makes the case that agents free call center teams for the work that is actually human. He also explains why he does not think the big model labs will simply absorb the application layer: a model provider does well when you consume more tokens, an application company does well when it finds the cheapest model that solves the problem. We close on telli's principles, including hell yes or no, never blocked, and what he calls calm urgency, plus why he would advise most people not to found straight out of university. We wrap up with a rapid fire round covering everything from Berlin versus Silicon Valley to LinkedIn flexing and whether AI creates or destroys jobs.
