Structural Decisions for Legal AI

I enjoyed joining a panel moderated by Ann Marie Lane at iManage, with Michael Israel, CIO, Kraft Group, and Kamal Dhawan, Client CTO/ATS at Microsoft, at an invite-only afternoon program on the structural decisions AI is forcing in legal, co-hosted by Microsoft, iManage, and Harvey.

We benefited from open framing by John James at iManage, organized around the concept of "IA before AI" (Information Architecture before Artificial Intelligence). The well-informed audience kept us on our toes with a lively Q&A, and as a result we didn't make it through all our planned topics — so I wanted to circle back to those here.

FUTURE LAW DEPARTMENTS — Looking ahead to 2030, what does the legal team of the future look like?

For most corporate law departments, revenue is not legal-services-based, and they are under pressure to stay lean. GenAI can help achieve that goal.

With ministerial work reduced, legal will also have more time to focus on high-value, proactive advice and counsel.

Tech skills will become more critical for paralegals, who will monitor and audit GenAI apps that provide more self-service to business clients.

The Legal Engineering role — which I hired into Hearst for the first time last year — is becoming more routine as we turn prototypes into robust, tailored applications. Law.com had an excellent article last week on law firms partnering with legal technology companies. That option is equally available to law departments with a tech-forward legal operations function and/or legal engineering services. At Hearst, we had a multi-year partnership with iManage to work on features that informed the INSIGHT product.

OUTSIDE COUNSEL — Surveys suggest many in-house teams expect to rely less on outside counsel because of AI. Where has that actually played out for you, and what types of work should always stay with outside counsel?

Specialty work for which the company doesn't have consistent internal demand will remain with outside counsel.

To the extent there is overflow work or secondments for leaves of absence, that work will come down.

Corporate law departments were created in the 1980s and 1990s to counter the high costs of law firms, so to the extent GenAI-first law firms arise, we could see a reduction in inside counsel.

SHIFTING RESPONSIBILITY — What's a task your legal team owns today that it probably won't own in three to four years?

A lot of the work involved in maintaining policies and answering routine questions will become more self-service through regulatory scanning and chat-based applications.

Contract review not requiring escalation will also become more self-service.

It's less that legal won't own the work, and more that legal will build and monitor the applications while being less involved in routine ministerial tasks day-to-day.

There is also new work that may have been beyond available bandwidth before for which law departments are now building solutions.

TOOLS — There's an AI tool for nearly every step of the legal workflow now. What is AI actually doing for in-house teams today, and is that the highest-value use of it?

In the iManage Knowledge Work Benchmark Report, there was a use-case slide, and in my experience both Hearst and current clients are going deep into that list — the top eight resonated. The top three are:

• Tabular review for M&A due diligence, contract extraction, and discovery
• Creation of playbooks that are more quickly derived from precedents and applied
• Regulatory scanning that can suggest redlines to policies, templates, and playbooks based on local, federal, and foreign jurisdiction changes

The most difficult work involves multi-step processes. Legal Engineering, building alongside the attorney subject matter expert (SME), is best equipped to resolve multi-step obstacles and fine-tune applications as underlying models shift results.

GOVERNANCE, PART 1 — What does robust governance look like in practice — the policies, guidelines, and frameworks needed to support that shift?

First, you want employees on enterprise-grade tools. If you make the bar too high to access enterprise licenses, employees will turn to unsanctioned tech, which expands the company's risk profile.

In my experience, companies are good at addressing external-facing privacy issues in their policies, in part because the media is filled with dramatic failure examples. If the company hasn't already done so, it needs to resolve how it approaches Article 4 of the EU AI Act, which took effect August 2nd, since it assigns responsibility both to those who develop applications on top of GenAI models and to the underlying model providers themselves.

At times, more consideration could be given to internal-facing confidentiality and auditing capability. For example, companies often set zero retention on company-wide tools to address discovery concerns, whereas legal tools need the ability to audit responses and correct as needed.

GOVERNANCE, PART 2 — How should organizations think about their data and systems architecture as AI adoption grows?

You need to make sure apps surface results only to people who should have access to the underlying data they draw from. If you work on a Microsoft or iManage platform — whether via an MCP connection or natively — the underlying security is retained. But if you pull information from a secure repository into another tool, you can lose that security, just as you would if you emailed the information out.

Data enrichment (like iManage provides via INSIGHT, and will make available on its next-gen tool set for general availability October 1) applies taxonomies consistently and enables GenAI tools to more quickly identify the source data or precedent most on point for the request.

GOVERNANCE, PART 3 — Where do you see legal departments getting governance wrong: too loose, too rigid, or focused on the wrong risk?

As noted above, being over-restrictive may encourage unsanctioned use.

The tools are evolving so quickly that a typical cybersecurity review done once at purchase is not sufficient. You have to set security around the work task, not focus solely on the tool. (Colin Levy of Malbek has an excellent LinkedIn article on this topic, “AI Governance Has to Survive the Technology”)

Following traditional paths leads to outdated positions quite quickly. Leveraging GenAI scanning tools to recommend updated positions and re-review tool documentation can help reduce human effort.

Next week, I'll share a few highlights from the second panel on what differentiates leaders in the space.