Where Firm Leaders Are With AI in 2026

One firm with a written economics test before any tool is built or bought said that automations which passed the test repeatedly failed to show up in the metrics that matter to the partner group. One group reported a thousand hours a year saved, but nobody can verify what those hours were used for. One firm three years into an audit platform could not prove meaningful savings.

Across every conversation, no leader named a destination for freed hours. Better margins, more revenue, developing the bench, or time back for people: each firm can choose, and zero had any plan for governing the additional capacity. Nor had any firm built a way to detect whether the hours were real. The closest was one partner who measures total hours per person against budget, to bring the highest-hour people down, but it was the only one.

Capacity that nobody aims tends to disappear whether a machine created it or not.

Over half of the firm leaders described their firm’s AI posture as deliberate restraint, and every one of them named a specific refusal with a reason attached. Not being the first firm to try a new tool. Capping the firm at one or two pilots at a time across every line of business. Signing no large contract until the economics are validated. Insisting on a firm-wide approach before picking tools. Dropping tools that turned out to duplicate what an existing license already did. Declining to chase a vendor’s claim.

Strategic advisors, chief AI officers at two Top 100 firms, and several association leaders looking at other firms describe some of the same behavior as being frozen. One AI consultant described a managing partner who had known for two years that his firm needed outside technology leadership and had done nothing about it, partly because admitting the gap to his own people felt worse than living with it. From the outside, restraint and freezing look the same. From the inside they can feel the same too.

At firm after firm, the most AI-capable people were the people with the least authority to change anything. About 61% of firm leaders described this first-hand. At one firm the interns were all over the tools one summer and gone in the fall, with nothing captured. At another, an intern with a cloud subscription and a few weeks built working tools and demonstrated them to the group her managing partner called the most resistant in the firm, the group changed its approach, and when her internship ended the firm lost what she had built. At a third firm, the heaviest use of a tool landed in a department the tool was never bought for.

The tenured middle is often where the resistance sits. Ten-year staff who say they already do not have enough time to do their job, so learning a tool leaves even less. A consultant described a client firm with a self-taught star builder the firm will not trust with any part of the AI strategy.

One caution: tool fluency and judgment are different things, and the firm sells judgment. Judgment stays with the tenured. Fluency sits with the less tenured. The problem is that most firms have no way to move fluency up without moving judgment down, and some of the fluency walks out the door every August.

Ask a firm leader what makes AI urgent and the answer is prevalence. Every conference. Every peer group. Every headline. Every vendor email. One managing partner named not wanting to fall behind as his own reason for moving. One leader described their firm’s chief executive coming back from conferences asking why the firm was not doing the far faster audits that vendors said other firms were doing.

Peers are the channel firms trust most: association rooms, alliance peer groups, conference hallways, a weekly call among a dozen managing partners. About half of everyone we spoke with named the peer channel as where trust and information come from. The same channel carries claims nobody has validated. When leaders did check, many of the claims did not hold. A vendor named a firm as a client before any contract existed. Products were sold that duplicated what an existing license already did. And peers years into a platform told one practice leader that the ROI had not arrived, because the work moved upstream into cleaning the data the platform needed (work the vendor had promised would not be needed).

Inside the firm, the buying itself often had no gate. One operations leader described two overlapping AI tools bought within weeks of each other by different people, the overlap discovered after both contracts were signed. Other leaders described a subtler version: every platform the firm already runs now ships AI, so the choice gets harder each quarter and nobody owns it.

The leaders who went looking for help found candidates quickly. An association chief executive said she could pick up the phone and reach people who do AI work well. No one we spoke with described a shortage of technical experts or AI thought leaders.

However, about 40% of firm leaders described, first-hand, not knowing how to evaluate the options in front of them, and others described the same gap in their peers. Partners struggling to find someone with both technical and commercial expertise, when most vendors lean heavily to one side. A managing partner without a technology background who needs to learn from others because he cannot put the pieces together alone. A chief executive asking who the person is who should own AI strategy, and knowing it is not her. An operations leader who struggles to connect tool usage to the strategic plan.

When leaders did trust outside help, the bar was high, but surprisingly human. Chief AI officers and the firm leaders furthest ahead named change management as critical, and described treating AI as a change in behavior and process more than a change in technology. Several named a working demonstration on the firm’s own work. AI fluency came a distant third.

One hypothesis sits behind this finding and is labeled as such: the firms describing gates, builds, and sequencing were the ones with a named owner of the technology decision, a partner hired for it, a chief operating officer with the mandate, an AI committee that reports to the partners, or a cross-disciplinary team working inside leadership’s guardrails.