Agents in law firms: the brake isn't the technology, it's billable hours, unpredictable costs and training juniors

By The Information · November 9, 2026.

The legal sector is trying to move from chatbots to AI agents—systems that carry out parts of a workflow on their own. According to a piece from The Information Partnerships, in which Laura Mandaro, its editorial director, interviewed three industry executives, adoption is growing fast but at uneven rates. The speakers were Isabel Parker (chief innovation officer at White & Case), Jillian Burnett (VP of Engagement at Legora) and Sunita Verma (CTO of Ironclad). Bear in mind that two of the three work for vendors of these tools, so their view is not neutral.

The clearest use case is non-disclosure agreements (NDAs). According to Verma, her clients want lawyers to spend less time on them and more on strategic deals. She says that in some cases the savings have allowed companies to bring in-house work previously handled by outside legal providers. She gives no figures in the piece.

At White & Case, according to Parker, lawyers are comfortable using AI for basic tasks such as summarizing documents, but most resist a broader rollout of agents. The firm does use client-specific "skills" (reusable instructions given to the agent) that capture preferences such as tone or who should appear on contracts and communications. It is a way of turning institutional knowledge that used to circulate by email into something shared.

Burnett points to two sources of friction. The first is AI fatigue: a lawyer who has just been trained on one assistant doesn't want to give up more billable hours to learn another tool. The second is noise when many people create skills; she proposes innovation committees and senior partners to decide which ones become firm standards.

The third problem is money. With vendors shifting to consumption-based pricing, Parker says, tech spending is "unpredictable" and firms are not prepared to account for it. Burnett says Legora helps forecast and monitor that consumption. The baseline metrics she cites are efficiency, speed and reduced risk, which for some clients means keeping headcount steady while absorbing more work and for others means responding to clients faster.

**Our take.** The most interesting part is not what gets automated, but what firms choose not to automate. The billable-hour model explains the resistance: a lawyer doesn't push back on AI out of fear, but because learning a tool costs time the firm bills for. As long as the business is built on selling hours, every efficiency gain collides with the revenue model. That tension, more than technical capability, will set the pace.

The most valuable passage is Parker and Verma's on junior staff. White & Case sets aside opportunities for them to learn rather than automating every task, and Ironclad pairs senior and junior developers to pass on judgment. It is the same pyramid that, broadly, is narrowing at the base across the legal sector: if agents absorb the routine work that used to train juniors, who will be the senior partner fifteen years from now? Whoever fails to solve this will save money today and lack judgment tomorrow. That is our opinion; the source does not frame it in those terms.

On cost, the warning applies more broadly: consumption-based pricing shifts to the client a risk the vendor used to bear, and forces firms to measure outcomes rather than licenses. As industry context, the same is happening in other sectors adopting agents, where governance and trust lag behind capability.

In the short term, this means an uncomfortable transition: routine tasks disappearing, workdays being reorganized and firms torn between gaining efficiency and protecting their talent pipeline. In the long term, we see a better horizon: cheaper, faster access to legal services for those who cannot afford them today, and professionals devoted to judgment and client relationships, which is what AI does not replace. Getting there requires firms to invest in training people, not just in buying tools.

One final caveat: the material is a promotional conversation with vendors, with no quantitative data on savings, adoption or return. Claims about results are statements by the parties involved, not independent measurements.