AI and Billing: Your Firm Is Not Selling Hours. It Is Selling Embedded Judgment

Imagine that every time someone bought a new license for Microsoft Word, Microsoft assigned a team of engineers to rewrite the product from scratch.

That would be crazy. Customers are not paying for the fresh labor of engineers each time they open a new document. They are paying for a product that contains decades of design decisions, bug fixes, feature improvements, security work, user feedback, and accumulated engineering judgment.

Law firms have their own version of accumulated judgment. A family law firm has seen hundreds of divorce petitions. A property tax firm has challenged thousands of assessments. A corporate firm has negotiated the same clauses across countless NDAs, SaaS agreements, purchase agreements, and employment contracts.

The billable hour makes each matter look as if it begins from zero. AI reveals that it does not.

That is why the most important billing question in the age of AI is not simply whether firms should keep billing by the hour or move to value pricing. The deeper question is what law firms are actually charging clients for.

Are clients paying for human time? Are they paying for lawyer supervision of AI output? Or are they paying for something more valuable: the firm’s accumulated judgment, embedded in templates, checklists, playbooks, prior matters, and AI workflows?

AI does not just make legal work faster. It exposes the fact that much of a law firm’s value was never the hour itself. The value was the judgment hidden inside the firm’s way of doing the work.

The standard argument: AI breaks the pyramid

The standard argument starts with the familiar law firm pyramid.

At the top are a relatively small number of equity partners. They own the firm, bring in clients, supervise matters, and share in the firm’s profits. Beneath them are non-equity partners, counsel, senior associates, junior associates, staff attorneys, paralegals, and other legal professionals. The lower you go in the pyramid, the more people there are.

The billable hour fits this structure neatly. Partners bill at high rates. Associates bill at lower rates. Paralegals bill at still lower rates. The partner’s economic power comes from leverage: one partner can supervise multiple people below them, and those lawyers and legal professionals generate billable revenue that exceeds their compensation and overhead.

In the traditional model, work flows down the pyramid. Partners handle client relationships, strategy, supervision, and high-stakes judgment. Associates handle research, drafting, analysis, discovery, and deal execution. Junior associates and paralegals handle much of the repetitive production work: document review, first drafts, summaries, chronologies, diligence, cite checking, and routine filings.

That division of labor also supports the way lawyers talk about value. Partners are paid for judgment. Associates are paid for analysis and execution. Paralegals are paid for lower-cost legal support. The client receives one invoice, but the invoice reflects a hierarchy of timekeepers.

AI threatens this model, according to the standard argument, because many of the tasks most exposed to AI sit near the bottom of the pyramid¹. Generative AI can summarize documents, produce first drafts, compare versions, extract issues, classify clauses, and organize large volumes of information. Those are the kinds of tasks that historically generated hours for junior associates, staff attorneys, and paralegals.

If AI can compress that work, the base of the pyramid shrinks. Some commentators have suggested that the traditional pyramid may give way to something more like a diamond: fewer junior lawyers and paralegals at the bottom, more emphasis on mid-level and senior lawyers, and partners still at the top.²

That story is partly right. If a firm’s profitability depends on large numbers of junior lawyers billing large numbers of hours, AI is an obvious threat. Toby Brown, a former chief practice management officer at Perkins Coie, modeled this problem by reviewing anonymized law firm time entries. He hypothesized that AI would have its greatest impact on “drafting and reviewing,” which accounted for 47 percent of revenue in his sample. Even using conservative assumptions, a 5 percent reduction in partner hours and a 20 percent reduction in non-partner hours produced a 13 percent revenue decline and a 7 percent reduction in profit margins.³

For a law firm partner, that is a frightening conclusion. It suggests a future in which the firm bills fewer hours, generates less revenue, and can no longer rely on junior lawyers and paralegals to produce the same economic leverage.

But the standard story is incomplete. It still imagines AI as if it were a new machine installed at the bottom of the pyramid.

Why the pyramid picture is misleading

The pyramid story assumes something like this:

Partners provide judgment. Associates provide analysis. Paralegals provide routine production. AI provides cheaper routine production.

That picture is comforting because it preserves the hierarchy. AI may shrink the bottom of the pyramid, but the top remains secure. The grunt work gets automated. The judgment work remains human.

But AI does not fit neatly into that picture.

AI can help with drafting, summarization, document review, and intake. But it can also help with issue spotting, risk classification, negotiation positions, strategy memos, client counseling preparation, and identifying when a supposedly routine matter is not routine at all.

The better point is this: AI does not replace one layer of the pyramid. AI cuts across the pyramid.

We should be careful about assuming that AI will automate only the work lawyers already consider low-status. Much of what lawyers call judgment is also pattern recognition: knowing which facts matter, which clauses are dangerous, which cases are routine, which risks require escalation, which arguments have worked before, and which client goals should control the strategy.

That does not mean AI can replace the lawyer’s professional responsibility. It cannot. A lawyer still must understand the client’s goals, make the final recommendation, communicate the risks, and stand behind the result.

But it does mean that the line between “routine work” and “judgment work” is not as clean as the pyramid suggests. Some judgment is live judgment, exercised by a lawyer in the moment. Some judgment is embedded judgment, accumulated over years and captured in templates, checklists, playbooks, intake questions, and workflows.

This is where the industrial revolution analogy can mislead us. We often talk as if AI will do to law what machines did to factories: replace repetitive manual labor while leaving human creativity and judgment untouched. But AI is not only a machine for reducing drudgery. It is a machine for applying patterns, classifying risks, generating options, and making predictions from context.

In other words, AI does not merely automate the grunt work. It can also help automate parts of the intelligence that law firms have built up over time.

That is why the future of AI and billing cannot be understood only as a fight over junior associate hours. The more important question is whether a firm can take its accumulated legal knowledge and turn it into a repeatable, lawyer-supervised workflow.

What are clients actually paying for?

Clients do not care whether a lawyer hand-crafted a document from scratch. They care whether their legal problem is solved competently, quickly, safely, and at a fair price.

If clients could solve a legal problem safely by putting quarters into a vending machine, many would do it. Clients do not buy lawyer effort for its own sake. They buy relief from a problem.

So what should lawyers charge for in the age of AI?

Option 1: Pay for time

The traditional answer is simple: clients pay for time. A partner bills one rate. An associate bills a lower rate. A paralegal bills a still lower rate. The invoice reflects the number of hours each person spent on the matter.

That model becomes harder to defend as AI compresses legal work. If a document that once took three hours to draft can now be generated in three minutes, the firm cannot honestly pretend that the old amount of human labor still occurred.

To be clear, lower production costs do not necessarily mean lower law firm profits. Jevons paradox may apply.⁴ If AI lowers the cost of legal services, more clients may be able to afford help. A family law firm might make less per routine filing but serve many more clients who otherwise would have gone without a lawyer.

But as a theory of value, “pay for time” is under pressure. AI makes it painfully obvious that the client was never really buying hours. The client was buying a result.

Option 2: Pay for supervision

A second answer is that clients will pay for lawyer supervision. On this view, AI does much of the drafting, summarizing, reviewing, and issue spotting, while lawyers supervise the output.

There is truth in this model. Lawyers must remain responsible for the work. Someone has to check the AI output, identify mistakes, communicate with the client, and decide whether the work is safe to use.

But as a complete theory of legal value, the supervision model is too thin.

It turns lawyers into quality control inspectors. The AI becomes the factory, and the lawyer stands at the end of the line checking samples before they leave the floor. That is an important role, but it is not the whole role of a lawyer.

It also risks preserving the worst habits of the billable hour. A firm could replace paralegal and associate time with AI, then simply move the invoice upward by charging premium partner rates for “AI supervision.” That would change the staffing model without really changing the client value proposition.

The client would still be left asking: what exactly am I paying for?

Option 3: Pay for accumulated skill and experience

The better answer starts somewhere else.

A law firm’s value is not merely the time spent on a matter, and it is not merely the final act of supervising AI output. The deeper value is the firm’s accumulated skill and experience.

A good law firm does not approach its thousandth divorce petition, property tax protest, NDA review, or asset purchase agreement as if it were starting from scratch. It brings to that matter everything it has learned from prior matters: the questions to ask, the risks to flag, the language to use, the traps to avoid, the exceptions to escalate, and the choices that tend to matter most.

In the age of AI, the firms that thrive will be the firms that can convert that accumulated knowledge into repeatable AI workflows.

That is the real shift. The client is not paying for tokens. The client is not paying for fake human hours. And the client is not paying only for a lawyer to inspect machine output at the end of the process.

The client is paying for the firm’s experience, captured in a workflow, delivered efficiently, and backed by lawyer responsibility.

Embedded judgment: the real product

Embedded judgment is legal experience converted into a reusable system.

It shows up in intake questions, templates, checklists, playbooks, risk flags, escalation rules, county-specific practices, preferred clauses, negotiation positions, and review standards.

This is the accumulated value of a law firm. It is also the part of legal work that can be converted into an AI workflow.

Consider a family law firm. A divorce petition workflow does not merely insert names into a form. It asks the right intake questions. It identifies issues involving children, property, venue, service, family violence, temporary orders, separate property, and name changes. Most importantly, it knows when a routine petition is no longer routine and should be escalated to a lawyer.

Or consider a property tax firm. A property tax protest workflow does not merely generate a protest letter. It identifies comparable properties, flags valuation anomalies, classifies the strongest arguments, builds the evidence packet, suggests negotiation positions, and escalates weak or unusual cases.

Or consider NDA triage. An NDA workflow does not merely redline text. It applies the firm’s playbook. It classifies risks, marks unacceptable provisions, suggests fallback language, and escalates unusual clauses.

The point is not that AI replaces a junior associate or paralegal. The point is that the firm has turned its accumulated knowledge into a system.

That is why the Wilson Sonsini example is so important.⁵ The firm did not simply ask a generic AI tool to “review this contract.” It built a workflow around a human-created playbook. The AI applied the firm’s rules to mark up contracts, and a lawyer reviewed the result before it went back to the client.

That is embedded judgment.

The client is not paying for the AI model. The client is not paying for the tokens. The client is not paying for a lawyer to reinvent the wheel by hand. The client is paying for the firm’s judgment, captured in a workflow, improved over time, and backed by lawyer responsibility.

How this changes billing

This is where Texas Opinion 705 matters.⁶

The opinion is clear on one point: lawyers cannot use AI efficiency as a way to bill fictional time. If a lawyer bills by the hour, the lawyer may bill for the actual time spent using the AI tool, refining the output, and checking the work. But the lawyer may not bill for the time that AI saved.

As Opinion 705 puts it, “A lawyer may not, however, charge hourly fees for the time that was ‘saved’ by using the generative AI program.”

That rule is essential. If AI drafts a document in minutes, a firm cannot pretend that a junior associate spent three hours drafting it by hand. Lawyers should not bill fake human hours for machine work.

Opinion 705 also addresses a narrower question: when can the lawyer pass through the cost of the AI tool itself? If the lawyer pays per use for a particular generative AI program, the lawyer may be able to collect that expense from the client if allowed by law and accepted by the client. But the lawyer generally may not recover more than the amount actually incurred and paid to the AI provider.

That makes sense as far as it goes. A generic AI subscription may look like overhead. A matter-specific, per-use tool may look more like a reimbursable expense, depending on the fee agreement and client consent.

But that does not answer the deeper billing question.

The client is not paying for “AI.” The client is paying for the legal service delivered through the workflow.

This distinction matters. Suppose a firm builds a divorce petition workflow based on years of family law experience. The workflow uses structured intake questions, firm-approved templates, issue flags, escalation rules, county-specific filing practices, and lawyer review. The client is not paying for the tokens used by the model. The client is not paying for the software subscription. And the client is certainly not paying for imaginary hours that no human worked.

The client is paying for a lawyer-supervised legal service that the firm has learned how to deliver efficiently.

The same point applies to an NDA triage workflow, a property tax protest workflow, or a discovery response workflow. The AI tool may be part of the delivery system. But the value is not the tool by itself. The value is the firm’s accumulated knowledge, captured in the workflow and backed by lawyer responsibility.

So it helps to separate four different things that the billable hour used to bundle together:

The mistake is to think that Opinion 705 leaves only two choices: bill the client for human review time, or treat AI as overhead. That is too narrow.

The better answer is that lawyers should not bill for AI as if it were a timekeeper. They should not bill for hours that were never worked. And they should be careful about passing through tool costs.

But firms can still price the legal service they deliver.

If a firm has built a reliable, lawyer-supervised workflow that produces a faster, cheaper, and better client result, the billing conversation should shift from “How many hours did this take?” to “What legal service are we delivering, what risks are we taking responsibility for, and what is that service worth to the client?”

This is not a license to bill clients for time that AI saved. Texas Opinion 705 is clear that a lawyer billing by the hour may not charge for hours that were not actually expended simply because AI made the work faster. Embedded judgment is different. It is not a fictional time entry. It is the basis for pricing a defined legal service: a lawyer-supervised workflow built from the firm’s templates, checklists, risk flags, playbooks, and experience. The ethical line is crossed when a firm uses AI to reduce its labor and then pretends the old labor still occurred. The better approach is to be candid about the service being offered, price it transparently, and make clear that the client is paying for the legal result, not for imaginary human effort.

Artisan firm versus AI Workflow firm

The old law firm model is a medieval cobbler model.

A client brings in a problem. The lawyer pulls up a stool. The lawyer studies the leather, cuts the pattern, stitches the pieces, and produces a handcrafted legal product. The more time the lawyer spends cutting, stitching, and polishing, the more the client pays.

That model still shapes how lawyers talk about value. Partners do the “judgment” work. Associates do the “analysis” work. Paralegals do the “routine” work. AI, in this story, merely replaces the bottom layer of the craft shop.

That is the wrong way to think about AI.

The future is not a slightly more efficient cobbler shop where AI cuts the leather and the partner inspects the shoes. The future is an AI Workflow firm.

An AI Workflow firm does not ask, “Which human used to do this task?” It asks a different question: “What is the best system for producing this legal result?”

That system may include structured intake, templates, rules, playbooks, document automation, client portals, lawyer review, escalation paths, and quality checks. The point is not that some parts are “low judgment” and some parts are “high judgment.” The point is that the firm’s intelligence has been redesigned as a repeatable workflow.

Consider two firms.

Law Firm A follows the artisan model. Every matter is treated as if it begins from scratch. Every document is treated as a fresh act of legal craftsmanship. The firm justifies its value by pointing to effort: hours spent, drafts revised, lawyers involved, process followed.

Law Firm B follows the workflow model. It identifies the legal results it produces again and again. It converts its experience into structured, lawyer-supervised AI workflows. It continuously improves those workflows as the firm learns from more matters. The firm justifies its value by pointing to the quality, speed, consistency, and reliability of the result.

Clients will not continue paying artisan prices for work that an AI Workflow firm can deliver faster, cheaper, and just as well. The winning firms will not be the ones that merely automate paralegal tasks. They will be the ones that rethink the whole service around the workflow.

The point is not to automate the grunt work.

The point is to automate the intelligence.

Takeaway for Texas law firm partners

We are living through a rare moment in the history of the legal profession.

For decades, law firms could organize themselves around hours. More time meant more revenue. More people beneath each partner meant more leverage. More manual work meant more billable entries on the invoice.

AI changes that equation.

The firms that move first will have a chance to do something extraordinary: convert their accumulated skill, experience, templates, checklists, playbooks, and judgment into repeatable AI workflows. Those firms will be able to deliver legal services faster, cheaper, more consistently, and at greater scale.

That is not a threat to good lawyers. It is an opportunity for them.

A law firm’s most valuable asset is not the number of hours it can bill. It is what the firm knows. It is the pattern recognition built from hundreds or thousands of matters. It is the partner’s instinct for what matters, what does not matter, what can go wrong, and how to solve the problem before the client even knows the danger exists.

The question is whether that knowledge remains trapped inside individual lawyers, or whether the firm turns it into a system.

Texas law firm partners should be asking hard questions now:

1. What do we know that other firms do not?
2. Which parts of our work repeat across matters?
3. Where do our templates, checklists, and playbooks already contain judgment?
4. Which tasks are expensive because they are truly hard, and which are expensive only because humans still perform them manually?
5. Where could an AI workflow create a faster, clearer, and less stressful client experience?
6. Which parts of our practice could become fixed-fee products if we redesigned the workflow from the ground up?
7. What would we build if we stopped defending the old model and started designing the next one?

The firms that wait will not simply be preserving tradition. They will be giving faster-moving competitors time to package the same services in a better way.

Clients will not keep paying artisan prices for work that another firm can deliver through a lawyer-supervised AI workflow. They will move toward firms that solve their problems faster, explain their pricing more clearly, and deliver reliable results without pretending that every matter must be handcrafted from scratch.

The firms that win with AI will not be the firms that merely automate grunt work.

They will be the firms that automate their intelligence.

Footnotes

1. Nancy B. Rapoport & Joseph R. Tiano, Jr., Fighting the Hypothetical: Why Law Firms Should Rethink the Billable Hour in the Generative AI Era, 20 Wash. J. L. Tech. & Arts (2025), https://digitalcommons.law.uw.edu/wjlta/vol20/iss2/2. ↩

2. Nancy B. Rapoport & Joseph R. Tiano, Jr., Fighting the Hypothetical: Why Law Firms Should Rethink the Billable Hour in the Generative AI Era, 20 Wash. J. L. Tech. & Arts (2025), https://digitalcommons.law.uw.edu/wjlta/vol20/iss2/2. ↩

3. Roy Strom, Law Firms’ AI Nightmare Is Fewer Billed Hours and Lower Profits, Bloomberg Law (May 16, 2024), https://news.bloomberglaw.com/business-and-practice/law-firms-ai-nightmare-is-fewer-billed-hours-and-lower-profits. ↩

4. Jevons Paradox, Wikipedia, https://en.wikipedia.org/wiki/Jevons_paradox (last visited June 16, 2026). ↩

5. Roy Strom, Law Firms’ AI Nightmare Is Fewer Billed Hours and Lower Profits, Bloomberg Law (May 16, 2024), https://news.bloomberglaw.com/business-and-practice/law-firms-ai-nightmare-is-fewer-billed-hours-and-lower-profits. ↩

6. State Bar of Texas Professional Ethics Committee, Opinion No. 705, at 5 (Feb. 2025). ↩