Plato, a San Francisco startup that builds simulated workplaces for training and testing AI models, raised a $16 million Series A led by Nava Ventures, the company announced on October 28. Gradient Ventures, which led Plato's seed round, joined along with Transpose Platform and Mustafa Suleyman, whom the announcement identifies as Microsoft AI's CEO.
The Plato funding brings its total to $19 million, which means about $3 million came in before this round. Nava founder Freddie Martignetti joins Plato's board.
Robert Farlow, the CEO, and Pranav Putta, the CTO, founded Plato in 2025. The company sells to AI labs that want models to practice multistep work in business software, such as placing an order or updating a record, where a mistake in a live system would spend real money or change real data.
How Plato's simulations work
Plato rebuilds the software a job runs on, with copies of tools such as Jira, Google Sheets and Salesforce, and fills them with mock data from people who do that job. Models practice inside the copies through reinforcement learning, attempting tasks and getting feedback, and each simulation can be reset and run again. Plato says it works with domain experts on jobs in customer support, people operations, finance, accounting and healthcare administration.
The company grades how a model reaches a result as well as the result. A purchase that breaks the budget does not count as a success, and its checks look for models that exploit flaws in a simulation to score well.
“We build evaluations that reflect what the job actually requires, so developers can distinguish a high score from work done correctly,” said Putta.
On October 27, the day before the round was announced, Plato published a benchmark called Computer Work Val v1.0, made of 300 tasks contributed by people earning more than $100,000 a year. The best of the five frontier models it tested passed 58.5% of the tasks on the first try, on average. Finance and accounting is the benchmark's largest domain, with 75 of the 300 tasks, according to Plato's website.
Why Nava led
Martignetti said in the announcement that Nava was first drawn to how the founders build realistic environments at scale, and then came to see the company differently.
Plato is really a data lab, working on the frontier of the field to support the world's best model development.Freddie Martignetti, founder of Nava Ventures
Martignetti described Farlow and Putta as “deeply talented researchers and developers from Waterloo and Georgia Tech.” According to the announcement, Farlow previously built software at Yutori for labeling the data that computer-use models are trained on, and Putta trained computer-use models at MultiOn.
What the money is for
Plato said the money will go toward gathering more real-world work experience: it plans to pay professionals to rebuild their own jobs as simulations and to run AI agents on real tasks. The company is also working on simulations that run over longer stretches of work, in which models deal with simulated people.
