The $1.1 Billion AI Bet Is Mostly Lawyer Time.
The easiest way to misunderstand A&O Shearman's new $1.1 billion AI commitment is to picture a procurement department writing a very large check. The firm announced something more consequential on November 5: most of the stated value will come from its practices committing at least 3 percent of their time to developing specialist AI tools and services. Reuters reported that practice leaders will assign lawyers to dedicated projects, rather than requiring every lawyer to record exactly 3 percent. The distinction matters. This is not chiefly a software purchase. It is a decision to redirect scarce professional labor away from today's matters in the hope of building tomorrow's reusable capability.
The headline number is an opportunity-cost number
A&O Shearman says it will commit $1.1 billion over five years and that the majority of the investment comes from every practice devoting at least 3 percent of its time to specialist AI development. That wording should be preserved. The announcement does not say the firm will spend $1.1 billion in cash on licenses, vendors, or data centers. It values the hours that lawyers and other professionals will not spend on ordinary production because they are building something intended to be used more than once.
Reuters added an important operational detail. With nearly 4,000 lawyers, the firm is not assigning an identical quota to every person. Practice leaders will place lawyers on dedicated projects, and 17 projects have been active since July. The commitment is therefore a portfolio decision made at the practice level. Some lawyers may spend much more than 3 percent of their time on a project, while others may spend none.
That makes the number more credible, but not automatically more valuable. A billable hour forgone today is only an investment if the resulting asset improves future work. Otherwise it is simply an expense with unusually elegant accounting. The proper comparison is not between $1.1 billion and another firm's press release. It is between the matters, training, client contact, and collections that the time could have produced and the durable capability that the projects actually create.
Law firms have long invested partner and associate time in precedents, training, knowledge management, pitches, and pro bono work. What is new here is the scale, the explicit valuation, and the premise that practice expertise can be encoded into tools clients and lawyers will use repeatedly. The firm has put a price on moving legal work from memory and individual craft into shared systems.
The firm is trying to turn judgment into infrastructure
A&O Shearman's stated theory is that generic AI becomes more useful when it is combined with deep practice and sector expertise. That is a defensible position. General-purpose models can summarize, draft, classify, and retrieve. They do not arrive with a firm's preferred clause positions, escalation thresholds, client-specific risk tolerances, regulatory taxonomies, or hard-earned sense of which exceptions matter.
The difficult work is not writing a clever prompt. It is deciding what the system should know, what evidence supports that knowledge, when a result needs human escalation, how the output will be tested, and who will update the system when law or market practice changes. A useful asset may include a decision tree, benchmark set, clause library, structured taxonomy, validation protocol, and maintenance owner. The model is only one component.
The firm points to ContractMatrix, its contract review and negotiation platform, as evidence of demand. It says client uptake increased ninefold during the last year and reports that one client rated its output 8.5 percent higher than the next competing provider in a blind evaluation. Those are the firm's claims, not an independent market audit. Still, they illustrate the intended result: not merely faster internal research, but a repeatable service that clients can evaluate against alternatives.
This does not mean legal judgment has become a commodity. It means parts of the path to judgment can be made explicit and reusable. A tax lawyer who identifies recurring issue patterns, a finance lawyer who defines acceptable clause variants, or a litigator who builds an evidence-classification protocol is not surrendering expertise. The lawyer is converting expertise into an institutional method. Whether that method remains sound depends on what the firm does after launch.
Three percent needs an internal constitution
A time commitment this large cannot run on enthusiasm alone. Every practice needs a way to decide which problems deserve investment. High frequency helps, but frequency is not enough. The work should also be sufficiently stable to encode, expensive enough to improve, testable against a meaningful standard, and capable of reuse across matters or clients without collapsing important distinctions.
The incentive system matters just as much as the project rubric. A lawyer asked to build a reusable tool is giving up some combination of billable credit, current matter experience, client visibility, and near-term revenue. If the firm praises innovation publicly but promotes and compensates people through the old production metrics, the 3 percent commitment will be absorbed by the people least able to refuse or reduced to administrative time that does not change how the practice works.
Credit also has to survive launch. Who receives recognition when a tool is adopted by a thousand users? Who is responsible when the law changes? Does a project have a budget for evaluation, security review, training, support, and retirement, or only for the exciting build phase? Can a practice stop a project that has senior sponsorship but weak evidence of value? Without clear answers, a portfolio of 17 projects can become 17 permanent obligations.
The best internal measure will not be hours committed. It will be useful work produced and safely reused. That includes adoption by the intended users, time saved without loss of quality, error and override rates, the number of matters benefiting from the asset, maintenance cost, client satisfaction, and whether lawyers can explain when not to rely on it. A percentage is a funding mechanism. It is not a result.
Governance is part of the product
A&O Shearman says a qualified lawyer remains responsible for legal work supported by AI and that governance committees will oversee deployment. Those commitments are not compliance decoration. They are features of the service. A client cannot meaningfully assess an AI-assisted legal product without knowing who is accountable for the result, what information the system used, how the output was tested, and what happens when the system encounters a condition outside its design.
The Solicitors Regulation Authority made the same point in its August warning on AI misuse. Using an AI system does not transfer or diminish a solicitor's professional responsibility. The SRA highlighted human oversight, verification, confidentiality, security, supervision, and compliance with court and regulatory obligations. Its February compliance guidance also calls for leadership oversight, risk and impact assessments, policies, training, and continuing monitoring.
That framework becomes more important when a tool moves from one lawyer's experiment to firm infrastructure. Version control matters because a result may change after a model, prompt, data source, or policy update. Evaluation records matter because anecdotes about good outputs cannot establish reliability. Access boundaries matter because practice knowledge and client information have different permissions. Escalation rules matter because the system must identify circumstances in which speed is less important than stopping.
The lawyer-led label is useful only if lawyers perform those functions. Subject-matter experts cannot simply contribute content at the beginning and lend their title to the finished interface. They must define the standard, review the evidence, own the exceptions, and remain involved when performance drifts. Responsibility without continuing authority is only a liability allocation.
Clients should judge the output, not the budget
The announcement will invite a predictable comparison among large firms: who has committed the largest number, licensed the most products, or logged the most AI activity. A&O Shearman says its lawyers recorded more than 558,000 AI-supported activities in the year ending September 2026, with 93 percent of lawyers and 96 percent of partners using AI tools. Those figures show broad use. They do not, by themselves, establish quality, savings, or client value.
Clients should ask narrower questions. Which work changed? What was the baseline? How often did a lawyer correct or reject the output? Did cycle time fall? Were errors detected before delivery? Is the benefit reflected in price, predictability, scope, or a new service the client could not previously buy? Who maintains the system, and what happens when its assumptions no longer hold? A large investment can support good answers. It cannot substitute for them.
Firms should be equally skeptical of their own numbers. Activity counts reward use even when use adds an unnecessary step. Time-saved estimates can become fictional if nobody measures the full review burden. Adoption can be high because a tool is mandatory rather than useful. The most revealing metric may be whether lawyers voluntarily return to the system after they understand its limits.
This is why the announcement matters beyond one global firm. A&O Shearman has made an unusually explicit claim that lawyer time can be treated as capital, converted into reusable infrastructure, and eventually returned as better legal service. That is a serious management thesis. It may prove right. But the verdict will not be found in the $1.1 billion headline. It will be found in the tools the firm keeps, the projects it kills, the judgment its lawyers preserve, and the value clients can actually see.
Sources and further reading
Primary and industry sources used to support this page. External guidance should be reviewed in context and for your jurisdiction.
- A&O Shearman, AI strategy announcementThe firm's November 5, 2026 announcement of its five-year commitment, practice-time allocation, usage figures, ContractMatrix claims, and governance position.
- Reuters, A&O Shearman's $1.1 billion AI planIndependent reporting on the practice-level allocation, dedicated project teams, 17 active projects, and the firm's approximate lawyer count.
- Solicitors Regulation Authority, warning on AI misuseThe regulator's August 2026 statement on professional responsibility, human oversight, confidentiality, security, and supervision.
- Solicitors Regulation Authority, AI compliance tipsThe regulator's February 2026 guidance on leadership, risk assessment, policies, training, and monitoring.