Join us at our next event - Inside: Union Square Ventures (June 10)
AI agents are huge news - yet enterprises are still hesitant to deploy them in mission critical or sensitive environments. It’s an inherent paradox: how do you give agents autonomy, while still maintaining control?
Enter Veris AI. Instead of developers having to observe, evaluate and fine-tune agents *after* setting them free in the wild, Veris AI gives them a simulation of the real world to test and train their AI agents.
I sat down with Mehdi Jamei, Founder and CEO of Veris AI, to talk about the business and the announcement earlier this week of its $8.5 million seed round. Check out the video interview above to hear about it from his perspective, and read on to learn more from mine.
The Vision: Autonomous Enterprise Agents
Veris is on a mission to enable enterprises to build fully autonomous workforces. The company is building a platform where organizations can train highly specialized AI agents that are tailored to specific enterprise tasks. Instead of relying on static, prompt-driven models, Veris is creating a dynamic environment where agents learn by doing, experimenting, and interacting within a sandboxed simulation that closely mimics real-world conditions.
What sets Veris AI apart is their focus on simulation-driven learning. By allowing agents to operate in a high-fidelity, risk-free environment, Veris AI ensures that these agents can learn from feedback and improve iteratively—without the risk of real-world failures. As Mehdi explained, “The ideal way for agents to learn is to truly put them in the real world and let them run and give them feedback from the environment to learn.”
Whereas the real world runs along a linear timeline, Veris enables agents to simultaneously run hundreds or even thousands of scenarios in parallel, and with immediate feedback from the simulated environment.
Testing, observing and evaluating agents can be a rather time-consuming, manual and risky process. With Veris, there’s no need to worry about response times from people in different time zones, or having to program an agent on how to deal with every distinct scenario it might encounter. And in a test environment, there’s no need to worry about what it will do with sensitive financial or health data. Instead, Veris uses generative AI to design variations on scenarios so that agents can train on and learn from these scenarios until they exhibit predictable, reliable behavior.
Why the Approach Matters
Most enterprise AI solutions today are still stuck in the era of manual fine-tuning and prompt engineering, which limits their adaptability and scalability. Veris AI’s approach is fundamentally different. By providing a library of simulated tools and user personas, they enable enterprises to create customized environments that reflect their unique workflows and challenges. Agents interact with these simulated tools and users, generating thousands of decision trajectories that can be used to train and refine the agents over time.
This iterative, simulation-based training is not just theoretical. Mehdi shared an example of a sourcing agent they built for a manufacturing client. The agent was able to search for suppliers, interact with them via email, and even use Slack to clarify details—all while learning from each interaction and improving its performance. This level of automation and adaptability is exactly what enterprises need to be able to confidently deploy agents in higher stakes environments.
The Team and the Road Ahead
Mehdi and his co-founder Andy bring deep expertise in agentic systems, given their deep research backgrounds and systems-building at both AI/ML startups and large companies like Google. They understand that the current tooling is not sufficient for the next generation of AI systems, and they are committed to building the infrastructure to support it.
With $8.5 million in fresh funding, Veris AI is now poised to grow its team of researchers and engineers in New York and invest in compute resources. The company is also planning to contribute back to the ecosystem by open-sourcing components of their platform, ensuring that the broader AI community can benefit from their innovations and accelerate agent development.
Why I Invested
I’ve gotten to know Mehdi over the past 1.5 years and come to really appreciate his deep perspective on AI technology as it changes at a rapid pace. I’ve also seen the way he operates, moving through tough challenges decisively. As it became clear that there is a huge business need for a better way to evaluate agents, Mehdi and Andi were developing this very forward-thinking approach to that problem, in a way that only they could. Giving AI agents on-the-job training is not just a technical breakthrough—it’s a fundamentally new contemplation of how enterprises can leverage AI to achieve agentic autonomy.
In the immediate future, we’re about to see the next generation of AI agents emerge and realize a greater potential.
Upcoming Lynx Collective Events
June 12 - Inside: Union Square Ventures
June 18 - NYC VC Investors - Summer Rooftop Mixer (VCs only)
July 2 - AI Builders Lunch w/Next Wave NYC
Share this post