Agent identity startup Mintive lands Cowboy Ventures, Coatue, angels Jeff Dean, Andrej Karpathy

We are rapidly moving from a world where we use software to a world where software uses software on our behalf. In the tech industry, we like to call these agents. But for the person running a corporate IT department or building a secure B2B startup, these agents are currently a nightmare. They are ghosts in the machine with high-level permissions and zero accountability.

Mintive, a new startup founded by veteran engineers from Google Brain and Coursera, just stepped out of stealth to fix this. They secured a seed round from Cowboy Ventures and Coatue, along with a list of angel investors that reads like a who-is-who of modern AI, including Jeff Dean and Andrej Karpathy. When names like that put money into a specific corner of the infrastructure, it is usually because they have seen the friction firsthand.

The Permissions Problem Nobody Wants to Talk About

For the last decade, we have perfected the art of Identity and Access Management (IAM) for humans. We have multi-factor authentication, single sign-on, and biometrics. We know exactly who logged in, where they were, and what they touched. But when an autonomous agent is given a task—say, to pull data from a CRM, analyze it, and then update a financial projection in a spreadsheet—it usually operates under a broad, static API key or a service account.

This is a massive security hole. If that agent gets compromised, or if the underlying model hallucinates a command it shouldn't execute, the system has no way to say no. Current enterprise systems are binary: you are either in or you are out. They aren't designed for entities that act autonomously and evolve their behavior based on the prompts they receive.

Mintive is building the layer that treats agents as distinct identities. Think of it as a passport and a background check for every autonomous bot running in your stack. It allows IT teams to set guardrails on what an agent can do, which data it can see, and when its permissions expire.

Why Builders Should Care

If you are building in the AI space right now, you are probably focused on latency, accuracy, and token costs. That is fine for a demo, but you will hit a brick wall the moment you try to sell to a Fortune 500 company. Their first question will not be about your LLM's benchmarks; it will be about how they can audit your agent.

The founders of Mintive have been in the trenches at Google Brain and Coursera long enough to know that enterprise adoption always lags behind technical capability because of risk. By providing a standardized way to manage agent identity, they are essentially building the plumbing that makes the whole agentic economy viable for big business.

• Auditability: You need to be able to show a trail of why an agent made a specific decision.
• Granular Control: Giving an agent 'Admin' access is a recipe for disaster; Mintive enables just-in-time permissions.
• Interoperability: Agents need to talk to other agents across different platforms without sharing raw credentials.

The Founder Perspective: Infrastructure Over Hype

What I find interesting about this round is the lack of noise. It is a fundamental infrastructure play. We have seen a lot of money thrown at wrappers and niche creative tools that will likely be sherlocked by OpenAI or Google within six months. But building the identity layer for the future of work is a much harder, much stickier problem to solve.

When I look at the involvement of people like Andrej Karpathy, I see a signal that we are moving past the 'wow' phase of LLMs and into the 'how do we actually run this' phase. Karpathy has been vocal about the concept of the Large Language Model as an Operating System (LLM OS). If the LLM is the OS, then Mintive is trying to build the security kernel.

The biggest bottleneck to agentic workflows isn't the intelligence of the model; it's the trust we have in the system's ability to stay within the lines.

The Skeptical Take

As much as the pedigree of the founders and investors is impressive, the challenge here is adoption. Every major cloud provider—AWS, Azure, Google Cloud—is going to try to build their own version of this. Microsoft, in particular, has a stranglehold on enterprise identity through Active Directory. For Mintive to win, they have to remain fiercely platform-agnostic.

They also have to solve the latency problem. If every action an agent takes requires a round-trip to an identity provider to verify permissions, the speed advantage of using agents starts to disappear. Builders will have to weigh the security benefits against the performance overhead.

What Happens Next

Expect to see more startups focusing on this 'middle layer.' We have the models (the brains) and we have the applications (the hands), but we are missing the nervous system that regulates how they interact. Mintive is the first major move in that direction, but it won't be the last.

For founders, the takeaway is clear: stop thinking only about what your AI can do, and start thinking about how your AI can be managed. If you can't prove that your agent is safe and controllable, it doesn't matter how smart it is. The enterprise won't buy it.

Mintive is betting that the future of the internet is populated by billions of autonomous agents. If they are right, the identity layer is the most valuable real estate in the stack. We are watching a fundamental shift from human-centric security to machine-centric security, and the stakes couldn't be higher.

Read the original at TechCrunch Startups →