Mintentif raises $11 million, a $7 million led by Cowboy Ventures, backed by Jeff Dean & Andrej Karpathy, to develop identity platform built on Google Brain

With the rapid integration of software agents within organizations, the core engineering challenge has shifted from the language models' ability to generate answers to securing the permissions that allow the machine to execute cloud actions. In this critical space, the platform “Mintentif” announced that it has raised $11 million in total funding, including a newly announced seed round of $7 million led by Cowboy Ventures, with participation from Koto, Operator Collective, and Hassle Fund, alongside angel investors such as Frederic Christ, co-founder of Octa, Jeff Dean, co-founder of Discovery Loop and former senior Google scientist, and Andrej Karpathy, co-founder of OpenAI.

The platform is built on technical expertise led by Jiqwan Najim, CEO and the first production engineer at Coursera and former neural-network developer at Google Brain, together with his co-founder and head of technology, Vijay Vasudevan, one of the contributors to the TensorFlow platform during his time in the same lab. The company offers a dedicated software suite for governing AI agents and managing their digital identities, including a custom gateway for the model-context protocol that allows IT teams to define which cloud services the agent may access and which operations it may perform. The platform also provides a gateway for language models to route queries and set spending caps, along with a system to monitor agent activity across the organization’s departments and enable them to carry out long-term autonomous tasks.

Separating the Agent’s Identity from Employee AccountsIt forms the core pillar of the service, as the platform grants each agent authorizations that are completely independent of the personal permissions of the employees they assist, thereby closing the loophole of uncontrolled privilege escalation. Despite fierce competition from infrastructure giants such as Amazon, Microsoft, OpenAI and Anthropic, as well as specialized platforms like LangChain, New Relic and Replit and dozens of startups, the platform bets on simplifying complex deployment environments. Najim explained that the system frees technology teams from dealing with sandbox configurations, test-bed architectures, or memory management, turning operation into concise steps, which has helped the company attract more than fifty enterprise customers since its product launch in February, including Modern Treasury, Abacus, Stability AI, and the University of Michigan.

The round carries a striking financing philosophy praised by Ellen Lee, managing partner at Cowboy Ventures, who noted that protecting institutional infrastructure requires governance and precise control before rapid expansion. Lee explained that the platform’s management avoided chasing inflated valuation rounds that similar companies in the sector have pursued, citing extraordinary rounds such as Prime Intel’s $130 million raise for a valuation approaching $1 billion. The venture fund’s leadership believes that building control tools step by step gives technology products operational robustness that shields them from the failure risks faced by entities that rely solely on fast-cash inflows.

Defining Software Agent Permissions Redraws Technical Spending PrioritiesIn Middle-East markets, particularly within banks, telecom companies, and government agencies in the Gulf, Egypt, and the Levant that are undergoing rapid shifts toward procedural automation, relying on isolated test environments no longer meets corporate audit requirements; assigning autonomous tasks to agents without independent identity gateways creates a risk of exposing sensitive data, especially when the tool’s behavior exceeds the permissions of the supervising employee. This shift compels IT directors in Riyadh, Abu Dhabi, and Cairo to incorporate context-protocol gateways and model-cost limits into technical procurement clauses, and to move local team skills from merely running software to configuring delegation policies and real-time monitoring of autonomous system behavior.

Mintentif’s experience reveals that the race for enterprise AI has entered a phase of structural discipline, where the allure of generative interfaces recedes in favor of control engineering and compliance. As more tasks are assigned to agents within work environments, technology leaders will face a clear equation: either build a strict supervisory wall that separates employee identity from tool authorization, or incur heavy security and financial costs generated by unmanaged digital process chaos.