Your Showcase Primer: Armadin, Ricursive, and Raindrop
A new world where AI builds AI, and AI agents run amuck
🌟 Engineers — want to meet these three founders and seven others? Apply to attend our March 25th SF Startup Showcase.
Something fundamental is shifting in software. Software systems aren’t just tools we write and run, they’re autonomous agents dynamically operating out in the real world. Ricursive is rethinking the physical layer, creating a self-improving feedback loop where AI models design faster, more efficient, next-gen chips, which in turn enable the training of even more powerful AI systems. Raindrop is making live AI systems observable, so engineers can understand how agents actually behave once they’re deployed. Armadin is preparing for a world where AI will be weaponized, building AI adversaries to test our defenses before attackers do. Different parts of the stack, same underlying change: once software starts making decisions on its own, the way we’ve traditionally built and managed systems need to be updated.
If you’re an engineer who wants to build the infrastructure of this new world, come meet these 3 founders at our next showcase.
Below, learn more about:
Travis Landham’s AI pen testing company, Armadin
Anna Goldie’s frontier lab, Ricursive Intelligence
Ben Hylak’s observability platform for AI agents, Raindrop
Travis Lanham, Founder of Armadin
Kevin Mandia could have ended his founder story when Google bought his startup Mandiant for roughly $5.4 billion in 2022. Instead, he is back at the starting line with Armadin because he’s convinced the ground is about to move under everyone’s feet: “Offense is going to be all‑AI in under two years,” he says, and if that’s true, most of today’s defensive stacks are structurally too slow.
At the center of that vision is Mandia’s cofounder, Travis Lanham, a former Google Cloud security leader who helped drive technical strategy and operations for exabyte-scale security systems inside Google after its acquisition of Mandiant. If Mandia brings the strategic vision of how cyber conflict is evolving, Lanham brings the real-world engineering experience of securing the very infrastructure that defends against today’s most sophisticated attackers.
Armadin’s core premise is that if attackers are going to leverage AI agents to lead offensive attacks, your defense can’t rely on human-driven, periodic red-team exercises. Businesses need an AI-driven adversary supercharging the business of finding vulnerabilities before someone else’s agent does.
The capital structure underscores how seriously the market takes that thesis. Armadin has already raised $24M in seed funding from Ballistic Ventures, the cyber‑focused firm Mandia co‑founded. It is in discussions to raise over $100M more from firms like Accel, GV, and Kleiner Perkins at a valuation north of $600M, even before the full story of the product is public. That kind of early valuation only makes sense if investors buy two things at once: that AI‑accelerated offense is not hype, and that Mandia and Lanham are uniquely positioned to build the defensive counterpart.
Which tech savvy pup are you? Find out at our next show!
Anna Goldie, Founder of Ricursive
Anna Goldie has spent her career in places where human intuition breaks down. At Google Brain, she co-led AlphaChip, the system that compressed chip floor planning from months into hours. The RL agents she trained to design chips looked wrong, with strange geometries and unconventional clustering, but the results came back working exactly as predicted, often better than human alternatives.
AlphaChip confirmed a constraint within the industry: AI could move faster if it didn’t take so long and cost so much to translate ideas into hardware. While models improve every quarter, silicon innovates every decade, and the industry quietly agrees to treat graphics GPUs as the long-term foundation for frontier AI.
Ricursive is Goldie’s answer to that mismatch — it’s her new frontier lab built around a simple, aggressive premise: AI should be designing the chips that power AI. Goldie believes teams should be able to describe a workload, specify performance and power constraints, and receive manufacturable silicon without needing to build a massive in-house design organization. The goal is to collapse the iteration loop for custom chips until it starts to feel less like hardware development and more like deploying software.
Ricursive’s team is dense with engineers who have already shipped frontier systems at scale. Beyond Goldie’s co-founder and AlphaChip collaborator Azalia Mirhoseini, the early group includes several engineers who worked directly on AI-for-chip and physical design pipelines inside large labs and tech companies. The team raised a $300M Series A led by Lightspeed — only a couple of months after raising a $35 million seed round led by Sequoia — and are hiring engineers.
Ben Hylak, Founder of Raindrop
Ben Hylak spent years at Apple helping build visionOS, working on systems where interaction and behavior had to line up perfectly. Spatial computing forces you to think beyond the demo. Once millions of people are using something, all the strange edge cases surface and what looked solid in testing suddenly behaves very differently in the real world.
When generative AI started showing up inside real products, Ben saw that same dynamic. The demos were impressive. But as companies move agents into production, silent failures, hallucinated actions, brittle tool calls, and unpredictable edge cases are becoming real operational risks. Traditional logging and APM tools weren’t designed for probabilistic systems, so they miss the kinds of issues that matter most in AI workflows.
Raindrop is building “Sentry for AI agents,” the first true monitoring and observability layer purpose-built for agentic systems. It provides real-time alerts, semantic issue detection, deep search, and root-cause insights that allow engineering teams to discover, track, and fix problems that would otherwise go unnoticed.
Since launching last year, they’ve grown 30% month-over-month. Clay, Tolan, Framer, AngelList, Type, and others already rely on it as they scale. Tolan reduced memory issues by nearly a third and almost eliminated “lore inconsistencies.” Founders call it “irreplaceable,” “like iOS crash reports but for our AI,” and “the thing that tells me what to prioritize when I walk into the office.”
Observability for AI agents is foundational infrastructure. Every company running agents in production will eventually need it. That makes Raindrop the kind of company where early engineers will define a category, not just contribute to one.
Want to meet these founders and ones of other breakout startups? Apply to attend our March 25th SF Startup Showcase.