Mintive raised a $7 million seed round led by Cowboy Ventures, according to TechCrunch's October 29, 2026 report. Coatue, Operator Collective and Hustle Fund also participated, bringing total funding to $11 million. The company develops a platform for IT teams to manage AI agents' identities, permissions and activity inside organizations.
The problem begins when an agent gains access to business tools. Being able to perform a task does not, by itself, answer who the agent is, whom it represents or which actions it may take. Mintive brings those questions into a management layer alongside model routing and cost controls, rather than stopping at the ability to create another agent.
An agent identity need not copy the employee's
A central component manages agent identities and permissions that can differ from those of the employees they serve. This distinction is at the heart of the offering: an employee and an agent acting for that employee do not necessarily need the same access. A management system can give the agent boundaries suited to its task even when the employee has broader responsibilities.
For example, a task requiring access to particular information does not necessarily need every action the employee is allowed to perform. This illustrates the reason to separate permissions; it is not a description of a particular customer deployment. The practical question is whether the organization can express that difference in its settings and follow activity under the agent's assigned identity.
Mintive offers this management as part of a wider platform. The company says setup takes only a few clicks. That is its claim about getting started, not an independent measurement of how long an organization needs to connect every system and permission. Initial activation and the surrounding organizational policy are different parts of implementation.
Two gateways for different decisions
The platform's MCP gateway sets which cloud services agents can access and which actions they can perform. In practical terms, it handles the relationship between an agent and tools outside the model. Access therefore involves more than connecting to a service; it also concerns what that connection lets the agent do in the workplace.
An LLM gateway sits alongside it, routing requests to different models and allowing cost caps. This is a different job: determining the path of a model request and its associated spending. A tool gateway and a model gateway are not two names for the same function. One concerns action in business services; the other concerns the models that help the agent do its work.
Connecting those gateways to agent identity explains the product's structure without turning it into a blanket security promise. Assigning permissions, routing requests and limiting costs are separate mechanisms. Their presence on a feature list does not establish that every agent decision will be correct or every organizational setting appropriate. It shows the control points Mintive aims to give IT teams.
For organizations evaluating the platform, that division also sharpens the demonstration questions. They can ask separately to see how tool access is set, how a model is selected and how an identified agent's activity is displayed. This tests the described capabilities against a defined need without assuming that a general description covers every workflow in the same way.
Monitoring agents and longer tasks
Another component, Agent Monitor, is intended to track agents across the company. The platform also includes tools for building agents that perform long, proactive tasks. Together, these extend the story beyond the initial connection: organizations need to understand which agents remain active after creation, particularly when work continues beyond one short request.
A long-running task raises different management questions from a one-off chat response. Who is responsible for the agent, what permissions does it hold and how is its activity followed throughout the task? These are questions about organizational design, not claims that a security incident occurred. Mintive presents monitoring and identity within the same management environment as tools and models.
The company launched in February 2026 and reports more than 50 enterprise customers. Named organizations include Ro, Modern Treasury, University of Michigan, Coursera, Apax and Stability AI. The customer count comes from the company. The list does not detail deployment size at each organization, so it should not imply that every component is used identically by every customer.
The customer count provides a reference point for the reported scale of activity, but does not itself disclose revenue per customer or the duration of each relationship.
The founders and investors
Mintive CEO Jiquan Ngiam was Coursera's founding engineer and later worked on neural networks at Google Brain. CTO Vijay Vasudevan co-created TensorFlow while at Google Brain. Those backgrounds connect learning systems with technology infrastructure and provide context for the team now building an agent-management layer.
Alongside the funds are prominent individual investors. TechCrunch named Frederic Kerrest, an Okta co-founder; Jeff Dean, a Discovery Loop co-founder and former Google Chief Scientist; and Andrej Karpathy, whom it identified as an OpenAI co-founder. These are details about funding participants. Their participation is not a product-test result and does not replace assessing the platform's fit for an organization.
The $7 million round and $11 million cumulative figure refer to different things: the newly reported investment and all capital raised to date. Even when presented together, the amounts should not be added as separate rounds. The distinction places the new transaction within the company's financing history.
Mintive belongs to the emerging category of AI-agent-access governance, but its funding is a separate event from other companies' rounds. Its proposition combines agent identity, tool access, model use and operational monitoring. For IT teams, the combination's value depends on translating a defined policy into settings they can inspect and test, not merely on how many agents the organization can run.