The Pulse
Plato Raises $16 Million to Train AI on Simulated Jobs
Plato announced a $16 million Series A on October 28, bringing its total funding to $19 million. The San Francisco startup builds simulated workplaces where AI models can practise tasks and be evaluated without changing live business record
AI.info Team · Written with AI · fact-checked ·
Plato has raised $16 million to let AI models practise work tasks in simulated environments, bringing the San Francisco startup’s total funding to $19 million. The company announced the Series A on October 28, saying it will expand its library of simulated jobs and pay professionals to recreate their work. Nava Ventures led the round, joined by Gradient Ventures, Transpose Platform and Microsoft AI CEO Mustafa Suleyman. Nava Ventures founder Freddie Martignetti will join Plato’s board.
Practice without changing live records
Plato builds simulated versions of workplace software and workflows where AI models can attempt tasks, receive feedback and try again. The company says the environments can be reset and reused, unlike live systems where an agent could spend real money placing an order or alter actual administrative records. The setup is intended for AI developers training and testing models before they operate in production.
“Mastering real complex work comes from experience doing those tasks,” said Robert Farlow, Plato’s co-founder and CEO. “We're building simulations that capture the tools, decisions, and unexpected situations those tasks involve, so AI models can gain experience in safe, contained environments before developers put them to work.”
Plato grades the method, not just the result
The company says its simulations cover customer support, people operations, finance, accounting and healthcare administration. AI labs use them for reinforcement learning: models attempt tasks and receive feedback intended to improve later attempts. Plato is also developing scenarios with longer sequences of work and interactions with simulated people.
Plato’s evaluations check both whether a model finishes a task and whether it follows the rules while doing so. A model that completes a purchase over budget, for example, should not pass simply because it placed the order. The company also says its tests look for models exploiting weaknesses in a simulation to obtain a high score.
The funding will add people and real-world tasks
Plato plans to spend the new capital on expanding its library of work experience. Professionals will be paid to recreate their jobs in simulated environments, while AI agents will be deployed on real tasks to collect additional experience. The announcement does not give a timetable for that work or report model-performance results from Plato’s environments.
Farlow previously built an annotation platform at Yutori to support training computer-use models, according to the announcement. Co-founder and CTO Pranav Putta has worked on computer-vision, robotics and reinforcement-learning research, as well as training computer-use models at MultiOn. Their backgrounds connect Plato’s product to a specific challenge for agent developers: teaching models to follow workplace constraints across multistep tasks, not merely produce a plausible answer.
A $19 million company, with results still to show
The $16 million Series A follows a seed round led by Gradient Ventures, the company said. Plato was founded in 2025 and is based in San Francisco. Its announcement describes the intended training and evaluation system, but does not disclose named AI-lab customers, commercial terms or measured gains in model performance.
For now, Plato’s plan combines expert-built simulations with experience gathered from agents working on real tasks. The funding will support that expansion; the company has not yet published results showing how well skills learned in its simulated workplaces transfer to live jobs.