Shahar Tzafrir at TLV Partners has been investing in AI companies that solve specific business problems like drug discovery and developer tools since years before AI became mainstream, positioning him as a go-to voice on how AI can redesign workflows in biotech and other industries. He has developed a recognizable intellectual framework around what he calls the legitimacy paradox—how narratives from customers, regulators, and talent either help or hurt the perceived legitimacy of emerging tech sectors.

• A strong conviction that artificial intelligence agents operating user interfaces directly will replace traditional backend integrations. The core idea is that software capable of viewing and clicking through screens adapts to changes much better than fragile manual programming or older robotic process automation. A specific example shared compares this transition to building humanoid robots that easily fit into a physical world already built for humans, noting that models from Anthropic are already acting like trained employees to navigate other software and explicitly mentioning an active investment in a startup facilitating this capability.
• High prioritization of unit economics and smart infrastructure choices for early stage companies. There is a clear warning that relying exclusively on premium artificial intelligence models causes startups to bleed their profit margins because computing costs are not dropping fast enough. To illustrate this, attention is brought to how pragmatic founders are beginning to route different tasks dynamically, reserving expensive models for premium users or complex needs, while downgrading simpler tasks to cheaper open source alternatives like Qwen or DeepSeek, or even reverting to traditional code to drastically reduce overhead.
• Searching for foundational technological breakthroughs that serve as enablers for entirely new business categories. The underlying investment thesis focuses on catching massive technological shifts that suddenly make previous impossibilities viable. This is highlighted by referencing a past successful investment in Aidoc, which capitalized on early deep learning advancements in image processing, and comparing that historic leap directly to the current wave of agentic software which is expected to unlock a completely new generation of automation startups.