Announcing our Series A and $50M in total funding to protect the world from AI agent failures
Today we're excited to announce our Series A and a brand new product called Raindrop Simulations.
We've now raised a total of $50 million led by CRV, and researchers and executives at Anthropic, OpenAI and Thinking Machines. Our goal at Raindrop is to protect the world from agent failures, big and small.
In the last year Raindrop went from a tool loved by startups to a tool that's also now trusted by some of the largest enterprises on earth across healthcare, logistics, and more.
Our focus in the last year has been unsupervised, real-time issue detection to find things going wrong in production agents. This is everything from small things like the bugs that annoy you in apps you use every day to the big things that impact the real-world health and safety of millions of people.
Simulations
Today we're taking that one step further. In addition to production monitoring, we're releasing the first way to make sure those issues don't happen in the first place: simulations. Simulations run on every PR, and show you the impact of your agent changes in the real world.
Traditional evals depend on pre-defined test cases and primarily measure failures a team has anticipated. Simulations replay real production traffic and existing test cases against a proposed change to an agent harness, then apply Raindrop's anomaly detection to the results. Teams can measure performance against their own test cases while also detecting unexpected behavior changes before they reach production. In other words, AI engineers can now see "what their change will change."
The really hard part about simulations is doing it in a harness-agnostic way. You can't just replay network calls or the old cached tool responses from the trace. You have to actually simulate the entire world around the agent. Imagine you add a brand new tool. Nothing exists for that tool in any trace. Imagine you move your entire harness to Pi. What now?
We think about it a little bit like Swiss cheese. Every tool call is a hole showing you the actual state of the underlying world. That's what we use to create simulations.
Simulations mark the first time companies have access to the same training and testing process used by frontier labs. OpenAI recently published research on deployment simulation, which regenerates responses to de-identified production conversations with a candidate model to predict misbehavior rates before release. Anthropic builds synthetic universes to train and stress-test its agents.
To date we've been building Raindrop Simulations hand in hand with our Fortune 100 partners. Today we're broadening early access and will be making it generally available over the coming month. Join the waitlist.
We're hiring
If you've read the news recently, you know our work is far, far from over. We're hiring across the board, with a focus on GTM and ML engineering. Our team has invented fraud transformer models at Robinhood, pioneered Pinterest's recommender systems, built agents for the federal government, and a lot more.
If this is an area you care about as much as we do, please email us.