The twelve startups gathered at Columbus's Vitria on the Square that late May afternoon in 2024 had the typical accelerator polish—slick decks, practiced pitches, nervous energy. But one team stood out for reasons that had nothing to do with presentation skills.
Daniel Kornev and Dr. Mikhail Burtsev weren't first-time founders piecing together a minimum viable product. They'd already shipped conversational AI to tens of millions of users at DeepPavlov, where Kornev ran product and Burtsev founded the company itself. Now, through their new venture Sentius, they were making a second bet on AI—this time building autonomous agents to automate enterprise tasks.
When Techstars Columbus announced its inaugural cohort on March 11, 2024, Sentius brought something the accelerator's emerging tech mandate practically demanded: battle-tested expertise in making AI systems actually work at scale. Whether that would translate to the messy world of enterprise software remained an open question.
Second Acts in Conversational AI
The DeepPavlov credentials were considerable. Burtsev had built a research profile spanning more than 100 papers in AI and machine learning. Kornev managed teams at DeepPavlov that developed AI products reaching significant user bases—experience that positioned them well for their next venture.
The third founder, Diliara Zharikova, led DeepPavlov's team competing in Amazon's Alexa Prize from 2019 to 2021. That particular crucible—building conversational systems capable of holding coherent, multi-turn dialogues—turned out to be formative. The lessons learned there, particularly around handling context and understanding user intent, would shape Sentius's technical approach.
DeepPavlov wasn't just an academic exercise. The team understood production systems, user behavior, failure modes. That institutional knowledge became the foundation for what they pitched as an "enterprise AI agent platform," though the gap between consumer chat and enterprise automation is wider than it might appear.
The Columbus Bet
Techstars Columbus, powered by The Ohio State University and led by Managing Director Tim Grace, made an explicit bet on deep tech for its first cohort. AI, quantum computing, technologies applied to foundational industries. Sentius fit neatly: autonomous agents designed to automate enterprise knowledge work.
The cohort itself reflected the scatter-shot nature of early-stage AI applications. Aspect Health tracked hormonal imbalances using continuous glucose monitoring and AI coaching. Pedal Data monitored gait patterns to catch early signs of neurodegenerative disease. Rubato brought AI-powered music therapy to dementia patients. Blended Edge built integration tools for software teams working with AI-native applications.
All twelve companies pitched at Demo Day on May 29, 2024. Sentius framed its vision around what it called "Teach & Repeat"—a platform where companies could build autonomous agents that learn workflows rather than requiring hard-coded scripts. Each startup received Techstars funding and equity designed to propel them toward Series A, though the accelerator would later update its standard deal to $220,000 in April 2025.
The Workflow Automation Problem
Here's what Sentius was actually proposing: instead of traditional robotic process automation—rigid, brittle, breaking whenever a website changes its layout—they wanted to build agents that could understand context. Observe a task once, understand the underlying logic, execute it autonomously going forward.
The technical architecture, detailed in a white paper the company published in December 2024, included several components. A Task Recorder to capture workflows. A Multi-Agent System with a Workflow Engine to orchestrate them. Specialized agents for different domains—a Browser Agent for web tasks, an OpenAPI Agent for service integration, a Prompt Agent for language model interactions. Teams could build and test agents in Sentius Studio, a Windows application, before pushing them to production.
In a December 16, 2024 press release, Sentius claimed a 94% success rate on the WebVoyager benchmark as of that date. Web-agent benchmarks evolve quickly and comparison gets tricky, but the figure signaled technical ambition in a field where even minor improvements matter.
The real question, though, wasn't technical capability. It was whether enterprises—cautious, risk-averse, committee-driven—would actually deploy autonomous agents for mission-critical workflows. Plenty of companies experiment with AI. Far fewer bet their operations on it.
What Happened After Demo Day
Several companies from the inaugural Techstars Columbus class found traction. Cariqa, building EV charging infrastructure with integrated financial services, secured €1 million by September 2024. Other cohort members continued iterating across healthcare, climate, and infrastructure applications, the typical post-accelerator grind.
Sentius announced a collaboration with Koop Technologies in March 2025 to launch "Housekeeper," an AI agent targeting compliance and insurance automation—sectors where workflow repetition meets regulatory precision. The company released platform documentation and continued developing its agent architecture. No additional funding rounds beyond the Techstars investment have been publicly disclosed, though that's not unusual for startups still in product-market fit mode.
The broader industry momentum seemed to be moving Sentius's direction, perhaps more than the founders expected when they started. Enterprise interest in agentic AI systems has grown, even as questions around transparency, reliability, and safety remain very much open.
The Second-Timer Advantage
For a team that already understood conversational AI at production scale, the shift to enterprise agents represented evolution rather than wholesale reinvention. They'd seen what happens when AI systems meet millions of real users—the edge cases, the failures, the unexpected ways people actually interact with technology.
Whether the "Teach & Repeat" model gains real traction will depend on timing as much as technology. Enterprises are experimenting with AI agents now, but the distance between pilot project and production deployment remains considerable. Security teams want assurances. Procurement wants pricing models. IT wants integration guarantees.
But Sentius brought something most accelerator startups lack on day one: they'd already shipped AI to millions of users, watched it succeed and fail, and lived to build again. In a field crowded with first-time founders pitching speculative visions, that experience counts for something. The question is whether it counts for enough.
