Ukraine to AI voice agents: Eva Karnaukh’s founder journey

Eva Karnaukh grew up in Kyiv, Ukraine, where her father - an entrepreneur himself - shaped the way she thinks about building things. He has since passed, but the entrepreneurial instinct he modeled carried Eva from Ukraine to the United States, through an M&A consulting career advising SaaS companies on deals and growth, and eventually into founding her own AI voice agent company. She runs a team of 20 and is heading into a seed round.

The pivot from M&A consulting to voice AI was not random. Advising SaaS companies on acquisitions meant spending years watching where the leverage actually lives in a software business - and increasingly, the answer pointed toward the interface layer. Voice and conversation, Eva argues, are the most fundamental communication technology humans have ever developed. Building AI agents that can participate in that channel fluently is not a feature; it is a platform. Her company builds AI voice agents for enterprise clients, with a philosophy centered on low latency, honest AI disclosure, and what she calls “outcome as a service.”

From M&A Consulting to Voice AI: The Pivot

The M&A consulting background is more relevant to what Eva built than it might appear. Working on SaaS acquisitions puts you inside dozens of companies' operational structures - you see where processes break down, where human labor is doing work that should not require a human, and where the cost-to-outcome ratio in customer-facing roles is wildly unfavorable. Voice and conversation kept surfacing as the friction point that no one had truly solved at scale.

The insight that pushed Eva to build rather than advise: voice is not just a communication channel, it is the oldest and most deeply human one. Hundreds of thousands of years of scraping berries off hillsides gave way to language, and language built civilization. AI voice agents that can participate in that channel with low latency, natural inflection, and contextual intelligence are not incremental - they are a category shift in how businesses interact with customers.

The Technical Foundation: Latency as the Make-or-Break Variable

Eva's company builds on top of existing voice AI infrastructure - platforms like Vapi - rather than developing the underlying synthesis layer from scratch. The engineering focus is on minimizing latency: the gap between when a user stops speaking and when the agent responds. In human conversation, silence is uncomfortable. Even a pause of a second or two creates tension that the listener will rush to fill. A voice agent with noticeable lag does not just feel slow - it breaks the conversational rhythm entirely, triggering the uncanny valley and destroying trust in a single interaction.

The other dimension Eva's team works on is the musicality of language - the inflection, tone, and filler language (what she calls “ad libs”) that carry as much meaning as the words themselves. A voice agent that responds with technically correct language but robotic cadence reads as inhuman immediately. Engineering the small non-verbal cues - the “hmm,” the slight hesitation before a complex answer, the warmth in a greeting - is where the real differentiation lives.

Disclosing the AI Upfront: The Uncanny Valley Decision

One of the most consequential product decisions for any AI voice company is the uncanny valley question: do you disclose that the caller is speaking to an AI, or do you engineer the agent to pass as human for as long as possible? Eva's position is clear - disclose upfront, every time. The alternative is to gamble with the caller's trust. If they realize they have been speaking to an AI for 60 or 90 seconds without being told, the primary emotion is not delight - it is embarrassment. They feel fooled. And that feeling transfers directly onto the brand whose agent deceived them.

There is a second benefit that comes with upfront disclosure: it resets the caller's expectations in a productive direction. When you know you are speaking to an AI, you focus on getting the outcome you need rather than trying to read social signals. The interaction becomes transactional in the best sense. And the AI can deliver: consistent language, no hold music, no accent barriers, no bad days. Eva frames the multilingual capability specifically as a major advantage - a caller navigating customer support in Spanish, French, or any other language gets the same quality of experience as an English-only call, without the friction of a heavily accented human agent struggling to communicate.

Landing Enterprise Clients: The Design Partner Model

Enterprise clients are notoriously difficult to acquire, particularly for a young company without a long reference list. Eva's approach was to come in as a design partner rather than a finished product vendor - transparent about where the technology was and where it was going, and asking for collaboration rather than pure purchase. This framing changes the dynamic entirely. Instead of a sales pitch, it is a co-development conversation: here is what we are building, here is the outcome we believe it can deliver for you, help us get it right for your context.

One of the primary technical challenges that emerged from those early partnerships was knowledge base completeness. An AI voice agent is only as good as the information it can draw on when a caller asks a question. Enterprise clients often discover - through the design partner process - that their own internal knowledge bases are incomplete, outdated, or inconsistently structured. Surfacing and solving that problem for clients became part of Eva's value proposition.

“Outcome as a Service”: The Positioning Shift

Eva uses a framing that cuts through the noise around AI terminology: outcome as a service. Rather than selling AI voice agents as a technology product, the pitch is about specific, measurable outcomes - calls handled, resolutions per hour, conversion rates, customer satisfaction scores. The enterprise buyer does not need to understand how the model works; they need to know that their support queue gets shorter, their customers get answers faster, and their team focuses on work that actually requires human judgment.

The “services as software” model - where AI replaces a headcount line rather than a software subscription line - is where Eva sees the real economic upside. The pricing logic shifts from seat-based or usage-based to outcome-based, which aligns incentives cleanly and makes the ROI case obvious to any buyer who has ever looked at a customer service labor budget.

Tools & Resources Mentioned

• Vapi - voice AI infrastructure platform referenced as an example of the layer Eva's company builds on top of
• AI Voice Agents - the core product category; inbound and outbound conversation handling for enterprise clients
• Outcome as a Service - Eva's preferred positioning for AI-driven labor replacement; pricing tied to measurable outcomes rather than seats or usage
• Design Partner Model - Eva's early enterprise acquisition strategy: transparent co-development rather than finished product sales

Frameworks

FAQ