By Harsha Kapre
For much of the past year, the conversation around agentic AI was dominated by ambition. Founders and investors alike talked about autonomous systems that could reason, act and operate with minimal human involvement. As we move into ...
By Harsha Kapre
For much of the past year, the conversation around agentic AI was dominated by ambition. Founders and investors alike talked about autonomous systems that could reason, act and operate with minimal human involvement. As we move into 2026, that narrative is shifting away from what agents might do someday toward what they can reliably deliver today.
Harsha Kapre, head of Snowflake Ventures
This shift is evident in findings from Snowflake’s Startup 2026: AI Agents Mean Business report, which quotes from conversations with eight AI-focused VC investors who discuss what they see in the market today and in the year ahead. Their perspectives reflect a broader recalibration underway across the venture ecosystem. The experimentation era is giving way to one of more intentional adoption. AI is increasingly treated not as a standalone feature, but as an operating layer — embedded in workflows, governed by policy and evaluated on outcomes rather than ambition.
In practice, that means agents are finding traction in narrow, well-defined use cases. Fully autonomous agents remain rare in production, particularly for complex or high-risk workflows. What is working are agents deployed in data-rich domains like software development, customer support, sales operations and internal analytics. In these environments, human-in-the-loop designs are not a compromise; they are often the reason agents can be trusted and adopted at scale.
What investors look for
This shift has changed how startups are evaluated. As agentic tooling becomes easier to build, impressive demos have lost much of their signaling power. What matters now is evidence of usage: customers running agents in production, measurable productivity gains and early revenue momentum.
Founders need to clearly articulate how their agents improve existing workflows and why that value persists over time. Without that clarity, even technically strong products struggle to stand out.
Capital dynamics are also shaping the market. Investment continues to concentrate around a small group of foundational models and infrastructure providers. Rather than crowding out startups, many investors see this as an enabling layer. Well-capitalized platforms absorb the cost of training and inference, allowing startups to focus on application-level value.
Looking ahead, 2026 is shaping up to be less about sweeping claims of autonomy and more about execution. Enterprises want agentic solutions that fit into existing operating models, meet governance requirements and deliver quantifiable business impact.
For VCs, the hype cycle has done its job. The next phase will reward startups that turn agentic AI into focused, outcome-driven businesses and can prove it.
Harsha Kapre is the head of Snowflake Ventures where he focuses on investments to drive innovation and unlock new value on top of the Snowflake platform. A seasoned product management leader, he originally joined Snowflake in 2017 as a senior product manager, a role in which he played a pivotal role in the company’s partner ecosystem expansion. Prior to Snowflake, he spent 18 years at IBM with various roles across master data management and data platforms. Kapre earned his bachelor’s degree in electrical engineering and computer science from the University of California, Berkeley.
Illustration: Dom Guzman
