Harvey Buys Aida to Give Its AI Agents Firm-Specific Context

Harvey bought Aida for its people. A Glean co-founder and a team of search and machine learning engineers will now build the layer that decides what Harvey's agents know about your firm, so if you buy Harvey, the question for your next renewal is which internal systems that layer will index and who owns what it builds. If you are one of Aida's sales customers, nobody has told you what happens to your product, and you should plan for a migration.

Harvey announced the deal on October 29, 2026. Aida's co-founders, Piyush Prahladka (a Glean co-founder) and Atish Das Sarma, "will join Harvey's product and engineering organization," Harvey says. Latham & Watkins advised Harvey and lists no price. It is Harvey's fifth acquisition of 2026, after Benchmark in July and Guardrails AI in September.

---

What Harvey Says It Bought Aida For

Harvey bought Aida to speed up one project: helping each customer firm build its own ontology. Harvey's announcement says the Aida team "will advance Harvey's work to help each firm build its own ontology," which it describes as "critical context that makes every agent, model, and product the organization uses more effective," and adds that "firms that own theirs will stand apart." Artificial Lawyer carries the first two lines as well.

Harvey CEO Winston Weinberg framed the problem this way: "Legal teams have decades of valuable knowledge locked away in documents and systems that are difficult to search and connect." His next line: "Unlocking that knowledge securely will fundamentally change how they work." Law.com's summary puts it in engineering terms: Aida "will support the engineering of Harvey's context layer and how it interacts with its AI agents."

Harvey's post never defines "ontology." In enterprise AI the word usually means a structured model of a firm's entities and the relationships between them (clients, matters, people, documents, prior positions) that an agent can query, so it does not have to rediscover those links from raw text on every request. Harvey also writes, per Artificial Lawyer, that "agents are already capable of far more than they do today. What stops them is that they don't know enough." Read that as the strategy: if every legal AI vendor can rent similar models, Harvey wants to compete on how much of your firm's context its agents can reach.

The team fits that plan. Prahladka co-founded Glean, whose whole product is permission-aware search across enterprise systems. Das Sarma previously led engineering, data science and growth teams at Patreon, Apple and X, according to Legal IT Insider. Aida itself was a sales tool. Its pitch was capturing context automatically from the systems sales reps already use, and that is the same problem Harvey has to solve inside a law firm.

What Happens to Aida's Customers?

Nobody has said, and that silence is the most important fact for Aida's customers. Harvey's announcement, the Latham note, and the Law.com and Legal IT Insider reports carry no statement on whether Aida's product continues, merges into Harvey or shuts down. As of October 31, Aida's own site shows no acquisition notice.

What Aida's customers connected matters more than its size. The site describes a "Digital Twin" for each sales rep that runs as a Chrome extension with email drafting, a Slack app, a desktop meeting recorder and a mobile app, with "every call, email, Slack thread, and CRM note captured automatically." Simplify's company profile lists syncs with Salesforce, HubSpot, Gmail, Zoom and Gong, a Palo Alto base, 11 to 50 employees and a 2024 founding. That is a lot of OAuth grants and recorded meetings sitting with a vendor whose team now works on legal AI.

Harvey's own record gives you the comparison. When it bought Benchmark in July, its announcement said "the Harvey and Benchmark teams are working together on a thoughtful plan to support existing Benchmark customers with continuity of service as a top priority." Benchmark served asset managers, a market where Harvey says it already works with more than 125 firms. Aida's sales buyers are not a Harvey market, and the Aida post has no such sentence. In September, Harvey took the Guardrails AI founders and team into the same product and engineering organization. We covered that deal at the time: Guardrails' hosted Hub had already been retired before the team moved.

To be fair to Harvey: a missing continuity line is not a shutdown notice, and a two-day-old deal often gets its customer letter later. But a team acquisition outside the buyer's market usually ends one way. Klaviyo's purchase of Agency came with a dated product shutdown, and Adobe's Rilo deal left customers holding a Slack token that never expires.

---

Why This Matters if You Buy Harvey

For Harvey buyers, the deal means the next part of the product wants access to more of your internal systems. A firm-owned ontology gets built from your document management system, matter records, email and knowledge bases. An agent can only use context it is allowed to read, so the capability Harvey is buying and the data access it will ask for are one request.

Harvey has the scale to make that request widely. Artificial Lawyer reported in September that 80% of Am Law 100 firms use it, along with in-house teams at five of the Fortune 10, and that it raised $550 million at a $15.5 billion valuation. Competitors are buying their way to the same place. Non-Billable counts five 2026 acquisitions at Legora as well, and it names Lume AI and Hexus as Harvey's other two deals this year.

Harvey's own phrase, "firms that own theirs," is the right test, and it is a contract term you can check today. Ownership has three parts:

1. The map itself. Can you export the entities, relationships and classifications Harvey derives from your data, in a documented format, if you leave?
2. Where it runs. Is the index tenant-isolated, and does it inherit the access rules of your document management system, or does it hold its own copy with its own permissions?
3. What trains on it. Does anything derived from your matters improve models or features that other customers use?

We made the same point in our Harvey vs CoCounsel vs Legora comparison: with legal AI, what you pay for is the grounding layer, because the underlying models are increasingly interchangeable. Harvey has already switched models once to protect its margin. Its context layer is the part that will stay put, so that is where your lock-in sits.

The Glean Precedent for Permissions

The team Harvey bought knows the hardest part of this problem from enterprise search: mirroring source-system permissions exactly. Our Glean vs Copilot vs Dust analysis argued that buyers should choose on the permission model before ranking quality. A context layer has the same exposure, with stakes a law firm adds on top: ethical walls, client confidentiality and privilege. An ontology that links a partner to a matter they are walled off from leaks that link to every agent that queries it, even if the agent never opens the document.

There is a cost question as well. Building relationship graphs over a large corpus is expensive. Our GraphRAG vs vector RAG cost breakdown put graph indexing at roughly 1,000 times the cost of a vector index. Harvey prices by seat. Ask whether ontology building is in the seat price or will arrive as a separate SKU.

---

What to Do Next

This Week (Aida customers):

1. Email your Aida account contact and ask three things in writing: is the product continuing, on what date does service end if not, and how do you get a full export of recordings, transcripts and drafted emails.
2. Pull the list of OAuth grants Aida holds in Google Workspace or Microsoft 365, Slack, Salesforce or HubSpot, and Gong, with your identity team. Name an owner for revoking each one on the day service ends.
3. Check your order form for change-of-control, assignment and data-deletion clauses, and diarize the notice periods.

This Month (Aida customers):

1. Stop expanding seats or rolling Aida's Chrome extension and desktop recorder out to new teams until you have a written continuity statement.
2. Shortlist a replacement for meeting capture and CRM updates, and run it alongside Aida for one sales pod so a sudden end of service does not cost you a quarter of forecast data.

Before Your Next Harvey Renewal (law firms and legal departments):

1. Ask Harvey for its roadmap for the ontology work: which source systems it will connect first, and whether it runs inside your tenant.
2. Write in portability for derived data (the entity map, not just your uploaded documents) and a clause barring cross-customer training on it.
3. Have your knowledge management and risk teams test any context connector against your ethical-wall configuration before it indexes a single matter.

The Bottom Line

Harvey is assembling a platform by acquisition. Benchmark added asset-management workflows, Guardrails added agent controls, and Aida adds engineers to build firm-specific context. Enterprise search went through a similar phase, and the vendors that held a customer's index were the hardest to replace at renewal. If you are a Harvey customer, settle in the contract who owns the map before it gets built. If you are an Aida customer, start your export this week.

Continue Reading

• Guardrails' Hub Died Aug 25. Harvey Bought the Team.
• Harvey vs CoCounsel vs Legora: Buy the Library, Not the Model
• Harvey Swapped Models After Agents Sank Its Margin to -50%
• Klaviyo Bought Agency. The Product Dies August 31.
• Stilla Promised Continuity. Meta Bought It for WhatsApp.
• Glean vs Copilot vs Dust: Buy the Permission Model