Raindrop announced a Series A funding round led by CRV on 17 September, bringing its total capital raised to $50m. The San Francisco-based company did not disclose the specific size of the Series A itself. The startup focuses on monitoring AI agents in live environments, scanning production traces to identify what it terms semantic anomalies—hallucinated responses, misapplied tools, and unexpected behaviour shifts that emerge following model updates. Teams using Raindrop gain visibility into what changed, when the change occurred, and which users were affected.
Agents now run for hours, call thousands of tools, and handle real money, real health data, and real customers. When an agent fails, it does the wrong thing convincingly at scale until someone happens to notice.
Zubin Koticha, chief executive of Raindrop
Testing the change before it ships
Raindrop unveiled Simulations alongside the funding announcement, currently available in research preview. The product replays actual production traffic and existing test cases against a proposed agent modification, then applies the company's anomaly detection system to the results.
The offering challenges the conventional approach to evaluation. Raindrop argues that traditional evaluation methods depend on test cases prepared beforehand, which typically only catch failures the team anticipated. Simulations operates on every pull request and aims to uncover behaviour shifts that teams did not predict.
The company references METR research demonstrating the rapid pace of agent capability expansion. Task duration that agents can complete independently roughly doubles every seven months. A single task execution can now span multiple days and involve thousands of tool interactions.
Raindrop also contends that Simulations grants companies outside the frontier labs access to the testing infrastructure those labs employ internally. The company points to OpenAI's work on deployment simulation, which regenerates model responses to de-identified production conversations. It also cites Anthropic's approach of constructing synthetic environments to evaluate agent robustness.
Who is backing it
Lightspeed Venture Partners and Y Combinator, both existing investors, participated in the round. Lead researchers from OpenAI, Anthropic, and Thinking Machines made personal investments. The company also counts Figma Ventures and Vercel Ventures among its backers.
Raindrop lists Vercel, Framer, and Clay as customers, along with undisclosed Fortune 100 companies operating in healthcare and logistics sectors.
Agents are fundamentally different from traditional software. They are highly capable, autonomous, and non-deterministic. Raindrop treats agent failure as a detection problem, the way a security company would.
Reid Christian, general partner at CRV
If we're having an issue like a build failure or agents stuck in a loop, we see that issue in Slack.
Bani Singh, AI engineer at Vercel
Bucky Moore, a partner at Lightspeed, warned that problematic agent behaviour will escalate to catastrophic consequences in high-stakes domains such as defence.
Zubin Koticha founded Raindrop alongside Ben Hylak and Alexis Gauba. The team brings together engineers who previously developed fraud detection systems at Robinhood and anomaly detection capabilities at Square. Security specialists from Segment, Semgrep, and Socket.dev have also joined the company.
A category getting crowded
Capital continues flowing into this space. groundcover raised $100m in July for observability solutions tailored to the AI era. Scaled Cognition secured $100m from Khosla in June to build reliable agents. Harvey acquired Guardrails AI this month.
The distinguishing factor among these companies lies in their point of intervention. Raindrop's thesis positions agent reliability as fundamentally a detection challenge rather than a design challenge. The company also bets that the critical moment for intervention is the pull request stage. Since Simulations remains in preview, this strategic positioning has not yet been validated by the market.
Source: The Next Web
