Scaled Cognition

Scaled Cognition: a foundation model built for customer service agents that follow the rules

Scaled Cognition is a Mountain View AI lab that trains APT (Agentic Pretrained Transformer), a foundation model designed specifically for building and running customer service AI agents at enterprise scale. Rather than prompting a general-purpose LLM and hoping it stays on script, APT is architected so that policy compliance is enforced at the model level: responses are verified against company policies and brand standards, and the company positions the system as hallucination-free for production customer interactions.

It is aimed at enterprises where a wrong answer has real consequences — banking and finance, retail, travel and hospitality, telecom, healthcare, insurance, and the BPOs that serve them. Teams integrate via a standard RESTful API and can build agents through no-code, low-code, or SDK paths, then validate them with GenAPI simulation tooling, test case generators, and customer simulators before anything touches production.

Key Features

• APT, an agentic foundation model purpose-built for customer service rather than adapted from a general chat model
• Architecturally enforced policy compliance the company describes as unable to be broken by prompt tricks
• GenAPI simulation for testing agents against synthetic customers without production risk
• Agent validation, drift detection, and continuous production monitoring
• No-code, low-code, and SDK deployment options behind a RESTful API
• Designed for regulated, high-stakes workflows: bank balances, medical records, flight changes, insurance claims

Ideal Use Case

Scaled Cognition fits large enterprises and CX platforms that want autonomous customer-facing agents but cannot tolerate hallucinated refunds, invented policies, or off-brand answers. Typical adopters are contact center operations in financial services, telecom, travel, and healthcare replacing scripted bots with agents that can actually take action under policy.

How Scaled Cognition differentiates

Where competitors like Sierra and Decagon build agent layers on top of third-party frontier models, Scaled Cognition trains its own. The company raised a $100M Series A led by Khosla Ventures in June 2026 at a reported ~$750M valuation, is in production with Fortune 500 enterprises, and counts Genesys — whose cloud platform serves more than 8,000 organizations — as both a customer deploying APT for virtual agent capabilities and an investor. Founders are CTO Dan Klein, a UC Berkeley AI professor, and CEO Dan Roth, who previously sold an early agentic AI company to Microsoft.

FAQ

What is Scaled Cognition? An AI lab building APT, a foundation model designed for policy-adherent, reliable customer service agents at enterprise scale.

What is APT? Agentic Pretrained Transformer: a model trained for agentic customer service tasks with policy compliance enforced architecturally.

How much does Scaled Cognition cost? Pricing is enterprise sales-led; there is no published self-serve pricing. Contact the company for a quote.

Who uses Scaled Cognition? Fortune 500 enterprises in financial services, healthcare, telecom, and insurance, plus Genesys, which deploys APT in its cloud CX platform.

tl;dr

Scaled Cognition trains APT, a foundation model purpose-built for customer service agents that follow policy. Backed by a $100M Khosla-led Series A and deployed by Genesys and Fortune 500 enterprises.

Related

Looking for more options? Browse the AI/ML Models directory or read our best AI models listicle. Scaled Cognition is also tracked on Crunchbase.