At Brightmind, we are drawn to founders tackling structural problems — the kinds of gaps adversaries exploit and operators feel acutely in the real world. When we first met Vijil, it was instantly clear they weren’t just building guardrails or another trust-and-safety wrapper. They were solving a foundational problem to enterprise adoption of Agentic AI: How do you make AI agents safe, reliable, and production-ready inside organizations where the cost of failure is enormous?

This is why we invested in vijil .

A World-Class Team for a Once-in-a-Generation Platform Shift

AI agents are poised to become the new application layer, and very few founders have the combination of AI systems engineering, enterprise software depth, and scientific horsepower required to harden this layer. Vijil’s leadership brings exactly that blend.

Across the technical team, Vijil has decades of experience in transforming research ideas into system software, building large-scale AI infrastructure, and operating deep learning services at AWS, Samsung, Microsoft, and Intel. Their scientific bench strength is exceptional — PhDs in multi-agent reinforcement learning, trustworthy AI, and responsible AI from institutions like Yale, Oxford, Caltech, Harvard, and UIUC. All with a singular focus: making AI agents reliable, secure, and safe for enterprises to use in mission-critical roles.

Vin Sharma , founder and CEO of Vijil, was former GM & director of Deep Learning at AWS. His team built the inference and training platforms of Amazon SageMaker and Bedrock, and helped scale the business to $400M ARR with 300+ employees. Working with Vin at AWS, Zdravko Pantic led all of SageMaker Training and spearheaded the development of services that made TensorFlow and PyTorch scale to AWS UltraClusters. As co-founder, he now leads all of engineering for Vijil. Radina Mihaleva, who previously scaled Lacework and Cloudflare channel strategy from $1M to $100M ARR, now leads GTM for Vijil. And Tim Rudner PhD (Oxford) joined recently as chief scientist. Tim is an assistant professor at University of Toronto, a 2024 Rising Star in Generative AI, a Rhodes Scholar, and holds joint appointments at several AI research institutions including the Vector Institute. Together, this team has the experience and expertise to build a category-defining infrastructure for trust in AI agents at scale. They are mission-driven and solving a generational challenge: the need for a trust platform that evolves just as quickly as AI agents themselves do.

A Foundational Problem That Enterprises Cannot Ignore

When we started Brightmind, we assessed the core blockers to enterprise adoption of AI. We consistently hear that enterprises want to deploy agents that can reason over ambiguous inputs, that can take actions with access to internal systems and external services, and that can operate autonomously with oversight.  We have a home lab at Brightmind, where we build and test AI systems, and are daily users of AI agents ourselves. We know as well as enterprises do that these agents introduce new classes of risk:

• AI agents hallucinate in subtle, harmful ways.
• They behave unpredictably in unseen edge cases.
• They can be manipulated through prompt injection or jailbreaks.
• They integrate with sensitive systems where mistakes create cascading failures.

Traditional cybersecurity platforms and the first wave of AI security 1.0 solutions have thrown content filters, static guardrails, or after-the-fact observability at the problem. None of these approaches are sufficient for agents making decisions in real-time in production systems.

Vijil recognized that trust cannot be tested once and then taken for granted— it’s a fabric that has to be woven into the development and operations lifecycle of every agent.

A Modular Platform Architecture for Trustworthy Agents

Where others see evaluation or guardrails as separate point tools, Vijil sees a unified trust layer running across the entire AI agent lifecycle. Their platform is built around three breakthroughs:

1. Diamond – From development to deployment, Vijil creates test harnesses bespoke to the agent in its target environment based on agent specifications, user personas, and organization policies. This isn’t benchmarking or automated red-team testing with static prompts sent from your laptop. This is continuous verification of trust tailored to each AI agent in each enterprise environment.
2. Dome – In production, Vijil provides a policy-driven runtime with observability to enforce guardrails-as-code inside or alongside the AI agent, wherever it runs. Using fine-tuned models to detect prompt injections and policy violations, Dome offers industry-leading accuracy, latency, and coverage. Not just a perimeter but a smart, “contextualized,” active defense integrated into agents built with LangGraph, Crew, Google ADK, and others.
3. Darwin – Darwin is the future of AI agent resiliency. Using multi-agent reinforcement learning over production telemetry, Darwin continuously improves trustworthiness of agents by tracing behavior back to content and context. Developers can harden the agent, its tool use, and context boundaries as well as the LLM inside, by learning from failures.

This is the most compelling part: Vijil is delivering trust that compounds. In a newly emerging category with dozens of AI security point-solutions, Vijil is executing a comprehensive platform strategy.  Vijil Diamond (customized testing) sends signals to Dome guardrails; Dome guardrails feed back telemetry to Diamond improving effectiveness of testing; Dome runtime monitors the behavior of AI agents and sends signals to Darwin, continuously improving agent resilience. Together, these modules form what enterprises have been missing: a cohesive trust infrastructure for agents that improves with every operational cycle.

Exactly What the Market Has Been Waiting For

CISOs, Heads of AI, and line-of-business owners consistently tell us the same thing: “We want agents in production but we can’t risk downtime, data leakage, or faulty decision-making.”

Vijil’s approach directly addresses these tensions.

And early traction has been strong: leading HR and productivity platforms, cloud infrastructure partners, and AI-native startups are already deploying Vijil. DigitalOcean, SmartRecruiters, and DuploCloud are happy enterprise customers. Vijil is partnering with AWS, Google, and Groq to help developers build trusted agents without compromising performance or time-to-value.

This is not a “nice-to-have.” It is table stakes for the next era of enterprise AI.

As a result, Vijil was named a Gartner Cool Vendor for AI Security, a CB Insights Top 100 Most Innovative AI Startup, and was recently featured in AWS’s 2025 Re:Invent Keynote as a “Most Innovative Startup.”

A Flywheel Only Vijil Can Unlock

Every evaluation, every guardrail enforcement, and every real-world signal feeds into a reinforcement learning loop that makes the next iteration of Vijil-built and Vijil-verified agents even stronger.  This is the most differentiating aspect of the Vijil platform:  how naturally it compounds.

That adaptive loop — the evolution of agents in enterprise environments — is where we at Brightmind believe the market ultimately converges. And Vijil is already there.

Our Conviction in Vijil’s Future

We believe Vijil is on track to become the defining trust infrastructure for the agentic era — much like identity providers shaped the SaaS era and cloud security companies shaped the cloud-native era.  It is rare in venture to find such an extraordinary technical team purpose-built to solve such a specific and timely mission.

We couldn’t be more proud to support the Vijil team on their journey.

‍