Who Owns Your Firm's Expertise Now?

What is "Who Owns Your Firm's Expertise Now?" about?

This article examines the under-discussed IP and governance implications when law firms encode institutional knowledge into vendor-controlled AI platforms—specifically Harvey's Workflow Builder.

A&O Shearman made an announcement in July that should have gotten more attention than it did.

The firm is building custom workflows with Harvey AI and reselling them to clients on a SaaS subscription model, splitting revenue with Harvey. The press coverage focused on the business model innovation. Law firms becoming software companies. Service-as-Software. The future of legal.

Nobody asked the question that occurred to me immediately: What happens to those workflows if A&O Shearman decides to leave Harvey?

The Revenue-Sharing Arrangement

Let me be clear about what's happening here. A&O Shearman—one of the largest law firms in the world, 3,000+ lawyers across 40 offices—is encoding its institutional expertise into Harvey's platform. Thirty years of M&A playbooks. Partner judgment on due diligence. Client-specific knowledge accumulated across thousands of transactions.

Then they're packaging that expertise as workflows and selling access to clients. Harvey takes a cut of the revenue.

The framing from both parties emphasizes innovation. "Building the future of legal services." "First-of-its-kind partnership." All true. But there's a contractual reality underneath the press release, and I'd be surprised if it favors the law firm.

Who Owns the Workflow?

I've been trying to understand who actually owns what in these arrangements.

The firm provides:

• Precedent documents and templates
• Partner-developed playbooks and checklists
• Client-specific institutional knowledge
• The "secret sauce" that justifies $1,500/hour rates

Harvey provides:

• The Workflow Builder platform
• LLM infrastructure and model access
• Integration with legal databases (LexisNexis partnership)
• The execution engine that runs the workflows

The resulting workflow is a fusion of both. It's not just the firm's expertise—it's been encoded into Harvey's proprietary system, structured to work with Harvey's model architecture, integrated with Harvey's data sources.

I haven't seen A&O Shearman's enterprise agreement with Harvey. But based on standard software vendor contracts and Harvey's positioning as a platform, I'd expect the IP terms look something like: Harvey owns the platform and all derivative works built on it. The firm licenses access.

The "we just provide the platform" defense might apply to generic tools. It doesn't hold when the platform's entire value proposition is encoding firm-specific expertise.

The Lock-In Math

Harvey is publishing metrics that should give law firm managing partners pause.

From Pereyra's thread on Workflow Builder traction:

• May 2025: ~50 workflows per week
• July 2025: 300+ workflows per week
• Trajectory: 6x growth in two months

Do the math. 300 workflows per week × 52 weeks = 15,600+ workflows per year across Harvey's customer base. Each workflow encodes firm-specific judgment. Each workflow becomes harder to replicate outside Harvey's platform.

This is the inverse of the Hollow Firm problem. That piece examined how AI automation threatens the junior talent pipeline by eliminating the grunt work that trains the next generation of partners. This is the flip side: firms are capturing institutional knowledge, but they're capturing it into a vendor's platform.

The switching cost isn't migration. It's rebuilding institutional memory.

A firm that builds 10,000 workflows over three years and then wants to move to a competitor would need to:

1. Recreate every workflow from scratch on the new platform
2. Re-train all users on new interfaces
3. Re-integrate with firm systems
4. Accept months or years of lost productivity

By design, this is prohibitively expensive. Harvey's $8 billion valuation isn't predicated on model quality—models commoditize. It's predicated on workflow lock-in.

The Governance Vacuum

There's a governance question here that firms haven't answered.

Who decides what gets encoded?

In most BigLaw firms, Workflow Builder access goes to Innovation Teams and Knowledge Management groups. These are typically mid-level attorneys and staff—not partners. They're building workflows that encode partner judgment, but the partners themselves may not be reviewing what gets built.

Who audits workflow quality?

A workflow that runs 500 times per week and makes a subtle error has now propagated that error 500 times. Traditional quality control assumes a human reviews each work product. Workflow automation assumes the workflow itself is correct. But who verified that assumption?

What happens when a workflow fails?

I wrote about the HITL Firewall and the importance of human-in-the-loop supervision for high-stakes AI applications. Harvey includes "Review Nodes" that pause workflows for partner approval. But those review nodes are optional. Firms decide whether to include them.

When—not if—a workflow produces work product that leads to malpractice, the liability chain is unclear. The lawyer signed off. But the lawyer was reviewing AI output, not original work. The workflow was built by KM staff. The platform was provided by Harvey. The model was trained by OpenAI.

Everyone can point at someone else.

What Partners Should Be Asking

If I were a managing partner at a firm deploying Harvey Workflow Builder, I'd want clear answers to these questions:

On IP:

• Do we own the workflow logic, or just license it?
• Can we export workflows to a competing platform?
• If we leave Harvey, what happens to the workflows we built?
• Does Harvey have the right to use our workflow patterns to train models or improve services for other firms?

On governance:

• Who approves new workflow creation?
• What's the quality review process for workflows before deployment?
• How do we audit workflows for errors after they're running?
• Is there a partner-level owner responsible for AI governance?

On liability:

• When a workflow produces defective work product, who's liable?
• Does Harvey indemnify us for workflow errors?
• Does our malpractice insurance cover AI-generated work product?
• What's the supervision standard for reviewing AI outputs?

The "I Worry That..." Moment

Here's what concerns me.

Law firms are so excited about the efficiency gains—30% time savings on contract analysis, 300+ workflows per week, revenue-sharing with clients—that they're transferring their most valuable asset to a vendor without negotiating ownership.

Institutional expertise is the only moat a law firm has. It's not the office space. It's not the brand (fungible across BigLaw). It's certainly not the hourly billing model (under pressure everywhere). It's the accumulated judgment of partners who've seen thousands of deals, know where the landmines are, and can spot issues that junior associates miss.

That expertise used to exist in partners' heads. Now it's being encoded into Harvey workflows. And Harvey—not the firm—controls the platform those workflows run on.

I don't think Harvey is acting in bad faith. They're building a platform and offering value. But the incentives don't align. Harvey benefits from lock-in. Firms benefit from portability. Only one of those parties is writing the enterprise agreement.

What to Watch

This situation will clarify over the next few years. I'm watching for:

The first defection attempt. When a major firm tries to leave Harvey after building thousands of workflows, we'll learn what those contracts actually say. Expect litigation—or a settlement that tells us everything we need to know about leverage.

IP disputes. If Harvey uses workflow patterns from Firm A to improve suggestions for Firm B, there's an argument that's misappropriation of firm expertise. The first lawsuit will set precedent.

Bar association guidance. State bar ethics committees will eventually issue opinions on AI vendor governance. California and New York tend to lead. The question of whether delegating workflow creation to KM staff satisfies supervision requirements remains open.

Insurance exclusions. Legal malpractice insurers are watching AI adoption closely. If they start excluding AI-generated work product from coverage—or pricing risk differently based on AI governance practices—that will force firms to take governance seriously.

The Uncomfortable Truth

Harvey Workflow Builder is genuinely innovative. It's letting law firms automate complex legal work at a velocity that would be impossible with traditional software development. The 300+ workflows per week metric is remarkable.

But innovation and governance aren't opposites. The firms deploying Harvey most aggressively should be asking the hardest questions about ownership, portability, and control. They're not.

The legal profession has spent a century building expertise. It's now encoding that expertise into a platform owned by a VC-backed startup with an $8 billion valuation to defend. The VCs expect a return. That return comes from lock-in.

Someone should ask who owns what before it's too late to negotiate.