Dan Klein and Dan Roth have spent years making AI that does not guess. The two co-founders of Scaled Cognition (Klein, a UC Berkeley professor who built the Berkeley NLP Group, and Roth, a serial AI entrepreneur who sold his last company to Microsoft) announced a $100 million Series A led by Khosla Ventures on June 25. The round values the company at approximately $750 million, according to the Wall Street Journal, and includes strategic investment from Genesys.
The money is for one thing: proving that AI can stop hallucinating.
"The problem isn't resources or effort, it's architecture," Roth said in the funding announcement. "You could have an interaction that was spectacular, think the singularity is here, and then look at the data and discover the system was making grievous errors."
Scaled Cognition's answer is APT (Agentic Pretrained Transformer), a model the company says delivers the conversational quality of frontier LLMs while guaranteeing policy-adherent, hallucination-free output. The key claim is architectural: reliability is engineered into the model, not bolted on as a safety wrapper. APT is smaller, faster, and cheaper than frontier models, and available for VPC or on-premise deployment, letting enterprises own their AI stack without ongoing dependency on third-party providers.
The approach stands apart from the dominant strategy in enterprise AI today. Most vendors take a frontier model (GPT-5, Claude 4, Gemini 3) and wrap it in guardrails, prompt templates, and output filters. Scaled Cognition argues that post-hoc safety cannot fix an architecture that hallucinates by design. "The way to quickly get into the market is to take a frontier model and put a layer on top," said Vinod Khosla, whose firm led the round. "What Scaled Cognition did was develop a different approach, then combine it with the best of LLMs. That took more research and more developmental risk."
Genesys, the cloud contact-center platform serving 8,000 organizations, is already using APT inside its Genesys Cloud platform for agentic virtual agent capabilities. The deployment is early but concrete: Scaled Cognition says its model is on track to automate more than one billion customer service interactions over the next twelve months. Fortune 500 customers across financial services, healthcare, telecom, and insurance are in production.
The founding story matters here. Klein and Roth previously built and sold Semantic Machines, one of the first agentic AI companies, to Microsoft in 2018. That gave them a front-row seat to the gap between what AI demoed and what it delivered in production. "The biggest reliability challenge isn't the mistakes that look wrong — it's the ones that look completely correct," Klein said. "If you want AI to take real actions on behalf of customers, that's the problem you have to solve."
The larger play is the $600 billion business process outsourcing market. Enterprises are beginning to insource what they once outsourced: customer service, IT support, HR, finance — replacing third-party managed services with AI workforces they own and control. Scaled Cognition is building the infrastructure that makes that possible. Whether the architecture delivers at scale is the open question the market will answer over the next twelve months.