SimpleClosure, a startup that helps struggling companies shut down, has launched a new product allowing defunct businesses to sell their accumulated code, Slack messages, emails, and workspace data to AI companies, according to The Verge, citing original reporting by Forbes.
The primary buyers are a new category of AI infrastructure company that The Verge calls “reinforcement learning gyms.” These firms “specialize in using defunct company data to build simulated environments where AI agents can practice navigating real workplaces,” according to The Verge.
The RL Gym Category
Reinforcement learning gyms construct realistic simulated enterprise environments where AI agents train by performing real tasks: navigating a CRM, responding to support tickets, writing and reviewing code, processing internal communications. The quality of an agent’s training depends directly on how realistic the simulation data is. Synthetic data can approximate workplace patterns, but real Slack threads, authentic engineering emails, actual Jira tickets, and genuine code with bugs and fixes produce higher-fidelity training environments.
The category is already attracting significant capital. The Information reported on April 12 that an unnamed RL gym startup has reached a $750 million valuation, buoyed by AI labs’ appetite for training data.
SimpleClosure’s Position
SimpleClosure already handles the legal and operational complexity of shutting companies down. The new data monetization layer adds a revenue channel for the shutdown process: shuttered companies can sell their data archives, potentially recouping some capital for founders and investors during dissolution.
The supply is consistent. Thousands of startups shut down every year, each sitting on years of authentic enterprise operations data: real decision trails, genuine human communication patterns, actual engineering workflows. That data had no commercial value after shutdown until now.
The Training Data Bottleneck
The offering addresses a specific constraint in AI agent development. Frontier model training has largely consumed the publicly available internet. Enterprise operational data, the kind that teaches an agent how a real company actually functions, is scarce and difficult to license from active businesses due to confidentiality, compliance, and competitive concerns.
Dead companies have none of those constraints. The employees have moved on, the products no longer exist, and the data sits idle. SimpleClosure’s insight is that this idle data is precisely what agent training infrastructure needs, and that the company shutdown process is the natural point of sale.
For teams building production AI agents for enterprise workflows, the quality of training and evaluation data is a primary differentiator. RL gyms represent the emerging infrastructure layer for that data, and SimpleClosure is positioning itself as a supply channel for the raw material.