The Great AI Filter Is Here: Not All Firms Will Get Through

The AI filter is already forming

AI capability is spreading quickly across legal work. In the 2026 Future Ready Lawyer survey, 92% of legal professionals reported using at least one AI tool, while 62% reported saving 6% to 20% of their weekly time. The survey is global and vendor-produced, not a PI-specific benchmark, but it captures the direction of travel: access is becoming ordinary.

Results remain uneven. Thomson Reuters' 2026 research found that nearly 80% of the lawyers it studied believed their practices had a clear AI plan, yet fewer than half felt confident that their practice area could succeed as AI became more integrated.

That is the filter. Some firms convert access into a new operating capability. Others remain aware, licensed, and strategically concerned while daily work changes very little. Falling behind may first appear as weaker margins, slower response, staff overload, or poorer client service rather than a dramatic collapse. Over time, those differences compound.

Seeing the disruption does not make the switch easy

Clayton Christensen's theory of disruptive innovation is often reduced to a warning that leaders failed to notice new technology. His harder insight was that well-run organizations can see a threat and still respond too slowly because their resources, processes, customers, and measures of success were built for the existing business.

In Christensen's strict theory, a disruption usually begins in a new or low-end market and improves until it can challenge incumbents. Not every AI product follows that path. The organizational lesson still matters: a promising technology can require economics and working methods that the existing firm is poorly designed to support. The Christensen Institute's theory library emphasizes designing systems and aligning the organization around the problem being solved.

A successful firm has accumulated habits that once made it successful. Cases move through familiar roles. Information follows familiar channels. Compensation rewards familiar outputs. Vendor contracts reinforce existing boundaries. Changing one component is easy. Changing the relationships among those components is much harder.

The missing capability is architectural knowledge

César Hidalgo's The Infinite Alphabet: And the Laws of Knowledge examines how knowledge grows, moves, and decays. In his course built around the book, Hidalgo connects learning curves and disruptive innovation with the difficulty of moving between them. One reason for that difficulty is architectural knowledge.

Rebecca Henderson and Kim Clark explored this problem in their 1990 study of architectural innovation. They studied companies making the equipment used to print tiny circuit patterns on semiconductor chips. Established manufacturers understood the underlying technology. Yet some struggled when new designs changed how familiar parts worked together.

Their distinction was simple. Component knowledge is knowing how each part works. Architectural knowledge is knowing how those parts depend on one another. A company could retain its expertise in the parts while losing its advantage when the connections changed.

The difficult discovery was that the old connections also lived inside the organization. Teams knew whom to consult. Engineers knew which information mattered. Testing routines looked for familiar problems. These habits made the company efficient at building the old product, but could cause it to overlook problems in the new design. Hiring capable people or buying new equipment did not automatically change those habits.

Applied to a PI firm, the idea is easy to picture. A lawyer knows how to evaluate a case. An intake specialist knows how to speak with a frightened caller. The CRM stores the lead. The firm's architecture determines whether the right facts reach that lawyer while the caller is still deciding whom to hire.

That legal example is an application of the study's insight. Henderson and Clark studied manufacturing, and their findings do not establish which law firms will succeed with AI. They give us a useful explanation for why knowing the technology can coexist with difficulty changing the organization.

A firm can understand every AI tool on the market and still lack the knowledge required to rebuild its own workflow around them.

A faster intake response can preserve a slow intake system

Suppose a PI firm adds an AI agent that answers after-hours inquiries in seconds, gathers facts, and creates a lead. The component works exactly as promised.

The inquiry can still be lost. The CRM may create a duplicate. No one may own the morning callback. The agent may flag a high-value case without alerting the right lawyer. Staff may distrust the summary and repeat the interview. Marketing may still count the lead without learning whether it became a wanted, signed, and retained case.

The firm automated the first step while preserving every broken connection around it. This is why AI cannot repair a broken PI workflow by itself.

Speed at the front exposes delays further along. The experienced employee who used to keep everything moving may have been compensating for gaps nobody had written down. Once a machine takes over part of the work, the firm discovers how much coordination depended on that person's memory and judgment.

What the filter selects

As more firms gain access to capable AI, the ability to learn together becomes more valuable. Some firms will produce faster individuals. Others will turn individual discoveries into methods the whole firm can use and improve.

In the first firm, useful prompts, corrections, and techniques stay with the people who discover them. When they leave, much of the learning leaves too. In the second, an employee's discovery changes a shared workflow. An error changes the instructions. A lawyer's correction clarifies where judgment belongs. The next matter benefits from what happened in the previous one.

This is also why legal AI implementation remains difficult. A Stanford Law School review of legal AI identifies firm structure, access to high-quality proprietary data, privacy, and integration with existing workflows among the enduring constraints. Better models do not automatically remove those organizational conditions.

PI firms have a useful economic starting point. Under a contingency model, reducing administrative effort can improve the economics of a case without reducing the hours available to bill. But an incentive to become efficient does not itself create the capacity to change. The owner still has to reconsider familiar roles, tolerate a period of learning, and make room for staff to question procedures that once worked well.

The title's warning is a competitive argument, not a forecast of mass closures. Losing ground may look ordinary at first: slower callbacks, more staff effort per case, inconsistent updates, less room to invest. If another firm keeps learning how to reduce those costs while improving service, the distance can grow. The model may be available to both firms. The accumulated ability to use it well takes time to build.

One practical implication

Start with one workflow and follow its connections. Notice where ownership changes, information is copied, judgment enters, and exceptions arise. Make those relationships clear before introducing AI, then assign an owner and measure the result. A small firm can often bring everyone involved into the same conversation.

This is the reasoning behind AI transformation one workflow at a time. The first project improves a piece of the business. It also teaches the firm how to change the next one.

The great AI filter will reward firms that can learn as an organization. Buying the technology is only the admission price.

Research basis

The knowledge framework draws from César A. Hidalgo's The Infinite Alphabet: And the Laws of Knowledge and his course on collective learning. The distinction between component and architectural knowledge comes from Rebecca Henderson and Kim Clark's 1990 paper, Architectural Innovation. The disruption framing uses the Christensen Institute's theory resources.

Legal-industry context comes from the 2026 Thomson Reuters Stand-out Lawyers Survey, the 2026 Wolters Kluwer Future Ready Lawyer survey, and the Stanford Law School paper on legal AI. Survey findings are directional and are not specific forecasts for personal injury firms.

Frequently asked questions