Partnering with Edra: Context for Agents at Scale

Ex-Palantir duo Eugen Alpeza and Yannis Karamanlakis have a pitch that sounds almost too simple: stop trying to teach AI agents how your company works, and just let them read the evidence you've already generated. That's the whole premise behind Edra, the startup Sequoia just announced it's backing, and it's aimed squarely at one of enterprise AI's ugliest bottlenecks.

Here's the problem they're circling. Every business, even two competitors in the same industry, runs on its own tangle of escalation paths, workarounds, and tribal knowledge that mostly lives in employees' heads rather than any manual. Plug a general-purpose model into that mess and it knows nothing. Fixing that usually means armies of forward-deployed engineers, months of documentation, and consultants billing by the hour — and the whole exercise has to be repeated every time a process shifts, which in most companies is constantly.

Edra's answer is to skip the documentation step entirely. The system mines support tickets, emails, chat logs, and internal records to build a knowledge base that reflects how the company actually operates, not the org chart's fantasy version of it. It keeps updating itself as people work, and — this is the part Sequoia's Luciana Lixandru leans on hardest in her writeup — it's legible. You can inspect exactly what the system has learned and why, a pointed contrast with opaque fine-tuning pipelines where nobody can explain a model's behavior after the fact.

Alpeza and Karamanlakis aren't newcomers to this specific pain point. Alpeza spent seven years at Palantir building its U.S. commercial business, including the AT&T deployment, before running the launch of Palantir's AI Platform in 2023. Karamanlakis became Palantir's first Forward Deployed AI Engineer, a role the two invented together to get LLMs out of demo purgatory and into actual production — he'd previously built a recruiting search tool that lifted a staffing firm's placement rate by 129%. They've known each other 13 years since university and always planned to eventually start something jointly.

So far Edra's traction is concentrated in IT service management and customer technical support, two areas where the underlying data — tickets, logs, chat transcripts — is dense enough to make the self-learning approach actually work, and where the cost of slow resolution is obvious enough that buyers move fast. Sequoia says early customers are expanding usage quickly, though as with any funding announcement from the investor itself, the claims come without independent numbers attached.