Plato has raised $16 million in a Series A funding round to expand its simulated workplace environments, where AI models can practice complex tasks before operating in real business systems. Led by Nava Ventures, with participation from Gradient Ventures, Transpose Platform and Microsoft AI CEO Mustafa Suleyman, the round brings Plato’s total funding to $19 million. The company is building repeatable simulations for jobs across customer support, finance, accounting, people operations and healthcare administration, with evaluations designed to measure not just whether an AI agent completes a task, but whether it follows instructions and respects constraints.
Plato Raises $16M to Build Training Grounds for AI Agents
AI models are increasingly expected to do more than generate text or answer questions. Enterprise developers want agents that can navigate software, manage workflows, make decisions and complete multistep tasks. Training those systems directly in live business applications, however, can introduce real-world risks, from unauthorized purchases to changes in operational records.
Plato is developing simulated workplaces where AI models can practice these tasks without affecting live systems. The company has raised $16 million in a Series A round led by Nava Ventures, with participation from Gradient Ventures, Transpose Platform and Microsoft AI CEO Mustafa Suleyman. The funding brings Plato’s total raised to $19 million. Nava Ventures founder Freddie Martignetti will join the company’s board.
Plato plans to use the capital to expand its library of simulated work environments, bringing more professional expertise and real-world task experience into the training process.
Why AI Agents Need Simulated Workplaces
Many AI systems perform well on questions with clear answers but face a different challenge when asked to complete tasks across software applications. An agent may need to interpret instructions, navigate multiple interfaces, respond to unexpected events and satisfy several constraints before a workflow is complete.
Training through live applications can be costly or risky. A model learning to place an order could spend actual money, while one practicing administrative work could modify real records. Simulations allow developers to reproduce these situations in controlled environments that can be reset and reused.
Plato works with domain experts to recreate the software, decisions, situations and expectations involved in jobs across customer support, people operations, finance, accounting and healthcare administration. AI labs can use these environments for reinforcement learning, in which models attempt tasks and receive feedback that helps improve their behavior.
The company is also developing more complex simulations involving longer sequences of work and interactions with simulated people. Such environments could help developers evaluate whether agents can maintain context and follow instructions across extended workflows, rather than succeeding at isolated actions.
Measuring Whether an Agent Does the Job Correctly
A central part of Plato’s approach is evaluating how a model completes a task, not just whether it reaches an apparent outcome.
For example, an agent that completes a purchase should not automatically receive a passing score if it exceeds the authorized budget. Similarly, a model that earns a high score by exploiting a weakness in the simulation may not have learned the intended behavior.
This distinction addresses a broader challenge in AI training: reward signals can encourage models to optimize for a metric rather than fulfill the actual goal. If an evaluation rewards task completion without checking constraints, a model may learn shortcuts that are undesirable in real business environments.
Plato says its evaluations examine whether models follow instructions, respect constraints and complete tasks correctly. The goal is to make training feedback reflect the requirements of the job itself.
For enterprise AI developers, this type of evaluation could be particularly relevant to workflows involving financial limits, approval requirements, customer information or administrative permissions. However, success in a simulated environment does not automatically establish that an agent will behave reliably in production. Developers still need to test performance across different scenarios, software versions and unexpected conditions.
A Growing Infrastructure Layer for Agent Training
Plato’s approach sits at the intersection of reinforcement learning, computer-use agents and synthetic training environments. Rather than focusing only on a model’s architecture, the company is building the environments and evaluations in which models gain experience.
Its founders bring experience in AI training and research. CEO Robert Farlow previously built an annotation platform at Yutori to support computer-use model training. CTO Pranav Putta has worked in computer vision, robotics and reinforcement learning, including training computer-use models at MultiOn.
Nava Ventures cited the company’s ability to create high-fidelity environments at scale as a key reason for investing. The firm also characterized Plato as a data lab supporting advanced model development.
That focus could become more important as AI agents move into enterprise applications. Developers need more than demonstrations of basic software navigation; they need repeatable ways to train and evaluate systems that handle longer, interconnected workflows. Simulated environments can provide controlled conditions for testing behavior before agents receive access to real business tools.
The challenge is making simulations sufficiently realistic. Real workplaces involve incomplete information, inconsistent processes, human judgment and unexpected exceptions. Environments that omit those complications may produce strong simulation scores without preparing models for production conditions.
Funding Will Expand the Range of Work Experience
Plato says it will use the new funding to expand its collection of simulated jobs by paying professionals to recreate their work in training environments. It also plans to deploy AI agents on real tasks to collect additional experience.
The combination could help connect expert knowledge with the varied situations agents encounter during training. Human professionals can describe important constraints and edge cases, while task execution can expose models to practical sequences of actions and decisions.
The announcement does not specify how many environments the funding will support, which AI labs are customers or how Plato’s simulations compare with competing training platforms. Those details will help determine how broadly the platform is adopted.
For now, the investment signals continued interest in infrastructure designed to make AI agents more capable at completing real work. As models shift from responding to prompts toward taking actions, the environments used to train and evaluate them could become an increasingly important part of enterprise AI development.
Market Landscape
AI agent development is creating demand for infrastructure beyond model training alone. Developers need controlled environments where agents can practice software use, learn from feedback and demonstrate that they can follow real-world constraints.
- Reinforcement learning: Models improve through task attempts and feedback, making the quality of training environments and reward signals important.
- Computer-use agents: Systems that interact with software need practice navigating interfaces, handling exceptions and completing multistep workflows.
- Simulation infrastructure: Resettable environments allow repeatable testing without the financial or operational risks of experimenting in live systems.
- Evaluation and reliability: Task completion scores must account for constraints, policy compliance and attempts to exploit weaknesses in evaluations.
- Enterprise automation: Customer support, finance, accounting, HR and healthcare administration are potential areas for agents trained on realistic workflows.
Plato is targeting the training and evaluation layer of the AI agent ecosystem. Its long-term opportunity will depend on the realism of its simulations, the quality of its evaluation methods and whether training gains transfer reliably to production environments.
Top Insights
- Plato raised $16 million in Series A funding led by Nava Ventures, bringing its total funding to $19 million.
- The company builds simulated workplaces where AI models can practice business workflows without affecting live systems or real operational records.
- Its environments target customer support, people operations, finance, accounting and healthcare administration, using reinforcement learning to improve task performance.
- Plato evaluates whether agents respect budgets, follow instructions and satisfy task constraints instead of measuring success only by final outcomes.
- The funding will expand its simulation library through professional expertise and additional task experience collected using AI agents.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
