Announcing Our Investment in Edra

Ontology was the silent word of the 2010s. A key insight from Palantir’s $350B+ journey: raw data alone isn’t enough to understand how a business operates. You need to map the nouns and verbs of a business—the objects, their states, their relations—on top of the data to navigate the intricacies of any operationally complex organization. The ontology was the grammar of the enterprise. Building an ontology is an immersive, painstaking process of extracting knowledge from the minds of experts across a company.

Similarly, raw data alone is not enough to power AI agents. An AI agent requires clear and detailed instructions to effectively execute tasks in a complex organization. It requires top talent and deeply customer-minded engineers to write those clear instructions to automate workflows or deploy AI agents reliably.

In every organization, the real knowledge of how tasks get done is hidden in inboxes, ticket histories, and logs. A sales discount gets approved in Slack. An IT escalation gets resolved in an email thread. A procurement inquiry gets answered by someone who's been there fifteen years and just knows.

The next generation of enterprise platforms will use this data to recreate processes from tribal knowledge and build instructions for AI agents. The companies that codify operational knowledge and simplify agent deployment will own the most valuable asset in the era of intelligent systems.

We've backed the team building it.

Introducing Eugen & Yannis 

We first met Eugen Alpeza a bit over two years ago when he gave a talk at an AIPCon. We were immediately impressed. 

Eugen and his co-founder, Yannis Karamanlakis, created the Forward Deployed AI Engineering team at Palantir, and spent years solving problems for organizations at the largest scale. They saw firsthand the time and effort it took, not just to build the ontology, but to also translate raw data into clear instructions that could actually drive automation. They understood that in a future where every organization deploys AI agents, not every agent can come with a team of forward deployed engineers writing and maintaining instructions for it.

Their insight was that rather than using stale documentation filled with contradictions and outdated information to provide instructions for AI agents, the data from tickets, messages, and logs could build a reliable playbook with clear and detailed instructions covering how processes are lived every day.

We spent a year learning alongside them while they were still at Palantir, watching the landscape shift, seeing them think through problems in infrastructure and post-training that most investors weren't yet asking about. When they left, they showed us their first demo: they'd taken raw data from an inbox and inferred why someone handled their email the way they did—the implicit logic, the conflicting choices, the priorities. From that, their system generated a reviewable playbook in plain text. It was the first time we'd seen anyone attempt to codify the "why”.

They've since expanded far beyond.

The Problem

Every enterprise maintains a knowledge base. Almost none function as a reliable source of truth, let alone as instructions for AI agents.

Articles go stale within weeks of publication. The gap between what's written and what's actually done widens with every interaction. 

Take one of Edra’s early adopters, a global multinational consumer brand. They were outsourcing their IT service management process and wanted to reduce their operational cost by automating ticket resolutions. But with limited insight into how the process was running, it was simply impossible to automate.

The failure mode is predictable: a third of incoming tickets have answers somewhere in the knowledge base, but the answers don't match the question closely enough, or they're out of date, or the agent doesn't trust them. The documentation describes what should happen in the general case. It can't tell you what happened in this specific case -- the context and the details of the exception, and the precedent that justified the deviation.

An outdated, inconsistent knowledge-base is hard for human agents to navigate, let alone an AI that doesn’t have any additional context and can’t ask follow ups.

The Edra Approach

Edra takes the unstructured exhaust of daily work and extracts the actual process in a playbook you can review, edit, and update.

Edra builds a living playbook of how an organization actually operates. The system learns not from the existing documentation, but from thousands of resolved conversations where someone figured out what to do.

The results surface what knowledge is missing, what's stale, and what exists only in the heads of tenured employees. Edra shows where documentation matches behavior and where it should be deleted entirely. 

From there, Edra becomes operational. When an email arrives, a draft response appears with suggested resolution and relevant precedent. When a ticket opens, a plan of action materializes. And when the decision logic is well-understood, the system handles resolution end-to-end.

Traction

Edra builds a Living Playbook by connecting to a customer’s existing systems in minutes, ingesting standard operating procedures, tickets, and communications without manual configuration. From there, it continuously learns from actual employee behavior and surfaces suggested improvements. Unlike static documentation or knowledge bases, Living Playbooks evolve as the business changes.

The pattern is proving out across industries where process knowledge determines competitive advantage.

Marosa, a Spanish provider of VAT compliance technology for large multinationals, maintains over 1,500 knowledge base articles to answer complex tax questions for more than 1,200 enterprise customers. Edra ingested 70,000 Outlook messages alongside the existing knowledge base and built a Living Playbook in seven days, surfacing 96 edits and generating 130 new articles in a four-week pilot. After going live, 25% of inbound queries were being handled automatically, with that number climbing week over week. 

An agricultural data platform signed a two-year enterprise contract in five weeks. Edra ingested their support history and surfaced five hundred pages of knowledge that had never been documented.

A global fashion retailer provided 400 knowledge base articles and 20,000 conversations spanning technical support, merchandising, and warehouse operations. The system identified which articles needed revision, which should be retired, and which needed to be written from scratch directly from resolution patterns. Their review burden dropped by half. 

Edra enables organizations to create operating leverage. The use cases extend beyond IT and support. The same engine now powers sales enablement, learning from call transcripts to build searchable precedent for revenue teams, and other operational functions. Anywhere work is digitally captured and resolution requires judgment, the playbook can be extracted and systematized.

That surface area is enormous.

Edra x 8VC

Today, Edra announced Series A led by Sequoia. Building on our $6 million seed, they have now raised over $30 million in funding. We've had conviction in Eugen and Yannis since before they left Palantir, and we're proud to have partnered with them from the earliest days.

Eugen and Yannis represent a rare combination: pattern recognition forged at the frontier of enterprise complexity, and the conviction to pursue a problem most people didn't know existed. 

If you want to learn more, or if you’re interested in joining their team, visit edra.ai.