Scaled Cognition, a Mountain View-based AI lab building Large Action Models for enterprise customer experience, has raised $100 million in Series A funding led by Khosla Ventures at a roughly $750 million valuation. The company’s APT-1 model is architected from the ground up to eliminate hallucinations through deterministic action prediction rather than probabilistic text generation. The capital will expand the research team and accelerate enterprise deployments.
Reliability Focus Amid Agentic AI Surge
The timing aligns with a broader push toward agentic systems that can execute real customer service workflows. Sierra AI has raised $1.58 billion total, while Decagon secured $481 million. Ada raised $200 million. Scaled Cognition differentiates by building its own foundation model instead of wrapping existing frontier LLMs with guardrails.
Hallucination Risks Block Enterprise Adoption
Enterprises report actual hallucination rates five times higher than their estimates because models sound confident even when wrong. A single error in healthcare or banking can trigger regulatory or financial consequences. Current solutions rely on post-hoc prompt engineering that fails under repeated identical scenarios.
Architecturally Enforced Determinism
APT-1 predicts actions and enforces policies at the model level rather than adding correction layers afterward. It achieves Pass 100 reliability, returning the same correct answer across 100 identical executions. The platform includes a simulation environment to test agents against millions of backend scenarios before deployment.
As Dan Klein, CTO, noted:
"Reliability is engineered into the architecture of our models, not bolted on after the fact."
Khosla Leads with Strategic Validation
Khosla Ventures led both the seed and Series A rounds, with Vinod Khosla joining the board. Genesys participated as a strategic investor and integration partner. The round signals conviction that architectural reliability, not incremental guardrails, will determine which systems reach production scale.
Vertical Model Labs Target $600B BPO Market
The AI agents market is projected to grow from $10.9 billion in 2026 to $182.9 billion by 2033. Scaled Cognition targets the $600 billion global BPO market by offering enterprises sovereign AI that runs in their own VPC or on-premises. Competitors such as Sierra, Decagon, and Ada wrap general-purpose models rather than training domain-specific architectures.
Founders Bring Prior Exit and Academic Depth
Dan Roth and Dan Klein previously built Semantic Machines, which Microsoft acquired in 2018. Klein leads the Berkeley AI Research Lab. Their experience with conversational AI informs the focus on deterministic execution over general intelligence.
Next Steps Include Research Expansion
The company plans to grow its team of roughly 25 researchers and engineers while scaling deployments with existing enterprise customers and the Genesys partnership.