Sequoia has backed Edra, a startup that solves the cold-start problem for enterprise AI agents. General-purpose models deployed inside a company begin with zero institutional knowledge. Edra fixes this by ingesting data the company already produces: support tickets, emails, logs, chat histories. It builds a living, editable knowledge base that reflects how the business actually operates, not how a process document says it should.
The founders are Eugen Alpeza and Yannis Karamanlakis, both former Palantir. Eugen spent seven years there, including launching the AI Platform under CEO Alex Karp and leading the AT&T deployment, one of Palantir's largest. Yannis was the first Forward Deployed AI Engineer at Palantir and previously shipped a recruiting search engine that lifted placement rates 129% for a staffing firm. The two have known each other for 13 years. Their first production wins are in IT service management and customer technical support, two domains where operational data is dense and agent errors are costly.
The technical detail worth reading for: Edra is not a black-box fine-tuning approach. The knowledge it extracts is transparent and editable. Users can inspect what the system learned and correct it. That auditability is the entire argument against generic RAG pipelines and opaque model updates. The full piece details how the system improves autonomously as agents use it, which is the architectural bet Sequoia is underwriting.
[READ ORIGINAL →]