The round. Mintive raised a $7 million seed led by Cowboy Ventures, bringing total funding to $11 million, per TechCrunch. Coatue, Operator Collective and Hustle Fund joined. So did angels including Okta co-founder Frederic Kerrest, Jeff Dean and Andrej Karpathy. The valuation is undisclosed. The founders are Jiquan Ngiam, Coursera’s first engineer, and Vijay Vasudevan, a TensorFlow co-creator.
What it does. Mintive gives every AI agent its own identity and its own scoped permissions, the way a company gives each employee a login. Agent traffic runs through Mintive’s gateways for model calls and for MCP, the protocol agents use to reach tools. The company says 50-plus paying customers, including Ro, Coursera and the University of Michigan, push about 2.5 million agent calls a day through them.
The problem in plain words. Most agents today borrow someone’s keys. A developer’s token. A shared service account. It works until something goes wrong. Then the log says “service account did it,” and nobody can tell you which person asked, which agent acted, or why it was allowed. That’s not a security gap you can explain to an auditor.
The login is table stakes. Issuing an agent a credential is the easy part. Every identity vendor and every cloud can do some version of it. The valuable part is the chain: this tool call came from this agent, acting for this person, under this rule. Mintive says it traces each action back to the human who started it. That trace is the product. It’s what lets a company say yes to agents at all.
Why the tape this week matters. CrowdStrike just agreed to buy Fiddler, which records what AI agents did and explains why, and Dreadnode, which attacks AI systems before launch. CrowdStrike didn’t disclose prices. The Wall Street Journal, via SecurityWeek, reported about $400 million and $95 million. Whatever the exact number, a platform is paying real money to own the agent lifecycle: test it, watch it, stop it. Identity is the piece that decides what an agent is allowed to do in the first place.
Where I’m skeptical. The space is crowded, and the crowd includes companies with distribution Mintive can’t match. Microsoft, Amazon and OpenAI all pitch agent identity in some form, and dozens of startups do too. A gateway also has to sit in the path of every call. That’s a latency, reliability and trust ask. If the gateway goes down, so do the agents. Seed-stage customer counts are encouraging, but 50 customers is a signal, not a market.
The only defensible position. Neutrality. Big platforms will govern agents well inside their own walls. Enterprises won’t live inside one wall. They’ll run several models, several agent frameworks and hundreds of tools. One rule set across all of them is something a platform has little reason to build. That’s the opening. It’s also, frankly, the profile of an acquisition target for an identity or endpoint platform that wants to own the agent layer later.
For buyers. Before you pilot anything here, ask three questions. Can it show, for one agent action, the human, the agent and the rule in a single record? What happens to your agents when the gateway is down? And does it work the same across every model you run, not just one?
Where I land. Agent identity is becoming a real category, and fast. But the category isn’t “logins for bots.” It’s accountability for bots. The company that makes the audit trail boring and complete wins. Everyone else is selling a credential.