With apologies to, but in everlasting admiration of, Dorothea Lange (1895-1965)
In January, we wrote that over the next three years AI tools would move from standalone apps into the tools lawyers already use, and that even "late adopters" would encounter AI during normal work whether or not their firm had an official AI program.
That three-year forecast took about nine months to come true.
AI is now part of the research platforms, practice-management systems, word processors, and email programs that small and midsize law firms use every day. Adoption is no longer the question. In Clio's 2026 survey of U.S. legal professionals, 75% of small firms reported using AI. But only 31% reported a related increase in revenue, according to the North Carolina Bar Association's review of recent surveys.
The more important gap is not between firms that use AI and firms that do not. It is between firms that manage AI and firms that merely use it.
Most small firms are on the wrong side of that gap. The 2026 ILTA Technology Survey found that 57% of firms with fewer than 50 lawyers have a formal generative AI policy, compared with 91% of firms with more than 700 lawyers. Only 11% of smaller firms require AI training before lawyers get access to the tools, according to a summary of the survey results. Across the profession, the 2026 Legal Industry Report found that only 9% of legal professionals say their firm has a written AI policy that is actually enforced.
In other words, AI has arrived in most small firms without the management system needed to make it safe or profitable.
What has changed since January?
Four developments have reshaped the practical AI agenda for small and midsize firms.
AI is in the tools, whether the firm chooses it or not. Lawyers no longer have to go looking for AI. It appears as a summarize button, a drafting assistant, or a research feature in software the firm already licenses. A firm with no AI policy still has AI use. It simply has no control over it.
Agentic AI has arrived. The newest tools do more than answer questions. They plan and carry out multi-step tasks, such as gathering documents, drafting, checking, and revising, with less human direction at each step. Use is still early. Fewer than 20% of organizations report widespread agentic AI adoption, but about half are planning to use it or considering it, according to the Thomson Reuters Institute. The American Bar Association’s Law Practice magazine expects midsize and even small firms to follow the early adopters in 2026 and beyond. Agentic tools raise a new management question: What may a machine do without a lawyer's decision?
Courts have lost patience. Researcher Damien Charlotin's AI Hallucination Cases database now lists 2,202 court and tribunal decisions worldwide involving AI-fabricated or misrepresented material. In the U.S., entries involving lawyers rather than self-represented litigants rose from 14 in 2024 to 207 in 2025 and 361 so far in 2026, according to an analysis of the database. The penalties have grown too. A federal judge in Oregon sanctioned two lawyers $110,000 for 23 fabricated citations and eight invented quotations, which Fortune described as the largest AI-hallucination penalty in American legal history.
Large firms are not immune. In April, Sullivan & Cromwell apologized to a federal bankruptcy judge for a filing containing AI hallucinations. The firm told the court that it had comprehensive AI policies and training requirements, but that those policies were not followed and its secondary review process did not catch the errors, Reuters reported.
The lesson for smaller firms is not that AI is too dangerous to use. It lies in the assumption many law firms make: that written policy is the same as a management system. A policy not built into how work actually gets done will fail under deadline pressure, even in the best-resourced firms in the world.
Regulators are converging. Professional regulators around the world are applying existing duties to AI rather than writing new rulebooks, and their expectations are remarkably similar. ABA Formal Opinion 512 addresses competence, confidentiality, communication, supervision, and fees, and cautions that boilerplate consent in an engagement letter is not adequate for putting client confidences into generative AI tools. In England and Wales, the Bar Standards Board's AI guidance took effect May 18, 2026, and the Law Society updated its generative AI guide in June 2026, according to a summary of recent U.K. regulations. In Canada, the Law Society of Ontario expects lawyers to have a human independently verify the output and to inform clients when AI use may affect their interests, the outcome, or the cost, as Innlaws explains. Singapore's Ministry of Law published a guide for using generative AI in the legal sector in March 2026.
Across these jurisdictions, four duties keep recurring: verify the output, protect confidential information, supervise the people and tools doing the work, and be candid with clients. Firms that build their AI management around those four duties will be well positioned almost anywhere they practice.
Five management problems for 2027
In January, we identified five opportunities AI offered small firms: efficiency, client service, competitiveness, value-based pricing, and a tech-confident culture. Those opportunities are still real. But in 2027, five management problems will decide whether a small firm captures them.
1. Shadow AI. When the firm does not provide approved tools, lawyers and staff use whatever is convenient, often free consumer chatbots. That creates confidentiality risk the firm cannot see. It can also create privilege risk. Fortune reported that a judge in Manhattan ruled that a defendant who used a general-purpose AI chatbot to help prepare his case had waived attorney-client privilege (Fortune). Forbidding the use of AI does not solve shadow AI. Better-approved tools, made easier to use than the alternatives, do.
2. Verification at scale. Checking a single AI-assisted memo is easy. Checking every AI-assisted document produced by every lawyer and staff member, every day, under deadline pressure, is a process problem. Who verifies what? How thoroughly? How is the check recorded? If the answer is "the lawyer who signs it will catch any problems," the firm is relying on exactly the safeguard that failed in the cases now filling the sanctions databases.
3. Agentic workflows. As AI tools begin acting on their own, firms need to define the points where a human decision is required. Examples include before anything leaves the firm, before a client's confidential information is sent to a new system, and before a deadline is calculated or calendared. These decision points should be designed in advance, not discovered after an error.
4. Turning time saved into value. If AI cuts a 10-hour task to two hours and the firm bills hourly, the firm has just reduced its revenue by 80% on that task. Clients increasingly expect to share in those savings. About 90% of legal dollars still flow through hourly billing arrangements, according to the Thomson Reuters Institute, while major bank clients are pressing their law firms to pass along AI savings, the Financial Times reported. The firms that benefit will decide deliberately where saved time goes: into fixed-fee and other alternative pricing that lets the firm retain some of the efficiency, into faster and more responsive service, or into capacity for new work. Clio reports that firms with wide AI adoption are nearly three times more likely to report revenue growth than firms that have not adopted it, and that most of those firms credit improved operations rather than the technology alone.
5. Vendor and data risk. Every AI tool is also a contract. Does the vendor train its models on the firm's data? Where is client information stored, and for how long? What happens to it if the firm changes vendors or the vendor is acquired? Small firms often accept standard terms of service without review. For tools that touch client information, those terms deserve the same scrutiny the firm would give any other arrangement involving client confidences.
A quality system, not a policy
More than 30 years ago, Walker Clark members helped pioneer continuous process improvement and total quality management in legal practice. Those disciplines are the right framework for managing AI, and arguably the only framework that scales to a small firm.
W. Edwards Deming's third point, "Cease dependence on inspection," is especially relevant. Relying on a final proofread to catch AI errors is inspection. It is expensive, unreliable under pressure, and, as the Sullivan & Cromwell episode shows, it can fail even in firms with strong policies. The better approach is to build quality into the process itself. That means designing workflows where verification happens at defined steps, where the tools and sources are controlled, and where errors are captured, analyzed, and used to improve the process.
For a small firm, that might look like this:
Plan. Choose two or three workflows where AI can make a measurable difference, such as intake, first drafts of routine documents, or research memos. Map how the work is done today and define the points where a human must check or decide.
Do. Run the workflow with approved tools, clear verification steps, and a simple record of who checked what.
Study. Measure results each month: time saved, error rates, rework, turnaround time, write-downs, and client feedback.
Act. Adjust the workflow, update the firm's AI playbook, and decide whether to extend the approach to another workflow.
This is not a large-firm program scaled down. It is the kind of disciplined, incremental improvement that small firms are very well suited to, because decision-makers are close to the work.
What will clients ask in 2027?
Clients are already asking their lawyers whether and how they use AI. In 2027, the questions will become more specific. Which tools do you use on my matters? How do you protect my information? Who checks the work? And if AI makes the work faster, what does that mean for my bill?
Firms that can answer those questions clearly, candidly, and consistently will turn AI governance into a competitive advantage. As we noted in January, more than 20 years of Walker Clark research with law firms’ clients shows that responsiveness and partner availability are the two highest-value service indicators. Well-managed AI can improve both. Poorly managed AI puts client trust at risk.
We will address the client side of this conversation, including disclosure, consent, and what AI means for fees, in an upcoming Worldview post on engagement letters.
How can Walker Clark help you?
For more than 30 years, Walker Clark members have advised law firms, corporate law departments, and government legal agencies on integrating new technology into their practices and managing the changes it creates. To learn how we might be able to help your practice, use the secure email contact link at the bottom of this page to schedule a complimentary, strictly confidential consultation with a senior member of our team.
This post is for general information and is not legal or ethics advice. Lawyers should consult the professional-conduct rules, ethics opinions, and court requirements that apply in their own jurisdictions.