A&O Shearman’s AI investment turns on the value of lawyer time

On November 5, the Financial Times reported a $1.1 billion A&O Shearman AI commitment over the next five years, largely involving time: practices would devote at least 3% of billable time to development. The amount is FT reporting, not independently confirmed here by a new official announcement. FT report

The distinction matters. Reading the headline as a cash budget for software would miss the economic experiment. A professional partnership is redirecting part of its scarce earning capacity toward tools that might be used repeatedly. Whether that becomes valuable institutional capital depends on the work that survives after the development hours have been recorded.

What the official record establishes

The firm's April 2025 announcement already described specialist agents and a share of subscription or usage revenue. It provides background for the strategy, not verification of the new investment figure. Official partnership announcement

This commercial structure creates two possible returns on the same expertise: better delivery inside the firm and software income outside it. It also creates tension. Knowledge useful in a particular client matter must be separated from material that may legitimately become a reusable product. A marketable workflow needs clear rights, maintained sources and defined limits, not simply a collection of successful prompts.

Legal and product teams therefore need to decide together which judgments can be standardized, which should remain suggestions, and which exceptions must return to a lawyer. Writing experience into a procedure tests whether it still works when separated from its original author.

A time commitment needs its own accounting

Lawyer hours have several values. Salary is a cost, the billing rate is a price, and revenue forgone depends on whether alternative client work actually existed and would have been collected. These are different quantities. A time-based investment should therefore not automatically be compared with a technology company's cash capital expenditure.

The management problem is to protect the allocation. If a busy practice loses development time whenever a deal arrives, the programme becomes spare-time experimentation. If it must meet an hours target regardless of useful output, it becomes an internal compliance exercise. Partners need credit for workflows that colleagues actually adopt, with review costs and maintenance included.

This is an economic interpretation of the reported structure, not a claim about the firm's undisclosed internal incentives. The same issue appears whenever a professional-services organisation asks its highest-paid experts to document knowledge that previously made those experts indispensable.

The unit of value is an accepted work product

Consider a loan-review workflow. Finding a clause quickly helps, but the valuable output also identifies the applicable agreement version, detects exceptions and shows the evidence supporting its conclusion. A fluent summary that needs a senior lawyer to repeat the entire investigation may save little.

A useful scorecard would measure total time to an accepted result, material errors found during review, failures discovered after delivery, and the cost of keeping the workflow current. It should distinguish routine documents from unusual jurisdictions and heavily negotiated terms. Average speed can conceal the cases where the risk is concentrated.

The firm reported $3.7 billion of revenue and $1.6 billion of pretax profit for its financial year ended April 2026. Those figures establish business scale; they do not isolate AI's contribution or demonstrate a return on this new programme. FY26 results

Pricing and responsibility move together

Hourly billing makes efficiency commercially awkward: doing the same task faster can reduce billed revenue. Fixed fees can let a firm retain part of a productivity gain, but only if the agreed scope is stable and the client values the result. Software subscriptions introduce a different obligation: users expect continued updates, support and a product that works beyond its creators' desks.

None of these models removes professional responsibility. The ABA's Formal Opinion 512 discusses competence, confidentiality, supervision and reasonable fees in the context of its Model Rules. It is a useful US reference, not a single binding rulebook for every jurisdiction in a global firm. ABA opinion

Operationally, responsibility needs a named owner at each handoff. Who can approve a changed source set? Who investigates conflicting authorities? Who signs off when an agent cannot resolve an exception? Review must have enough time and evidence to challenge the output. A nominal human approval step can otherwise become a rubber stamp.

The longer test is institutional learning

There is also a training trade-off. Junior lawyers learn through work that AI may accelerate. Removing repetitive tasks without replacing their educational function could weaken the next generation's ability to spot errors. A better design would use saved time for supervised comparison of outputs, disputed clauses and deliberately difficult cases.

The investment thesis ultimately rests on reusable judgment: knowledge that can travel across teams while retaining clear limits and accountability. The indicators to watch are sustained use, lower review-adjusted delivery costs, transparent client pricing and a credible process for updating or retiring failing workflows. The headline establishes ambition. These operational results will establish whether the partnership has built an enduring asset.