Why Your Hours Don't Add Up Anymore

An hour is a strange thing to sell.

Nobody wants the hour itself. You want what comes out of it: the contract, the tax return, the audit opinion, the report on your desk. The hour is just how the bill keeps count.

For a long time, that was close enough, because harder work took more hours and the bill rose right along with it. Clients grumbled. And paid.

Now picture a first draft that used to fill a whole morning, and a tool that turns it out before your coffee goes cold. Someone still has to read it. Check it. Fix it. Sign it.

So what goes on the bill? The morning? Or the coffee?

That isn't a riddle anymore. Across several industries, the people who set the rules, and the people who pay the bills, have started to write about it. Lawyers' ethics bodies. A tax authority. Audit regulators. A government that buys consulting and translation. A federal appeals court.

They don't all give the same answer. That's the interesting part.

Here's the short version, as I read them. If you bill by the hour, the time a tool saves goes to the client. If you price the job, the answers differ. And several of them treat the time spent checking the tool's work as real work.

The Meter Counts the Minutes

Start with the hour. Virginia's Legal Ethics Opinion 1901, on reasonable fees and generative AI, was approved by the Supreme Court of Virginia on November 24, 2025. It says it in its first paragraph: time-based billing, such as hourly fees, can only be based on the actual time spent on a task. It calls that clear.

The State Bar of California's practical guidance on generative AI, from 2026, says hourly billing must reflect the time actually spent on the client's matter, while recognizing that generative AI may reduce the time required for certain tasks. Time spent. Not time saved.

Think of a taxi meter. It counts minutes, and it doesn't care that your driver knows a shortcut through the back streets. Take the shortcut, and the fare drops. That's the deal.

On an hourly bill, as I read these documents, the saving goes straight to the client. The tool shortens the trip. The meter shows it.

So what can go on the meter? California gives two examples. A lawyer may charge for the actual time spent crafting or refining what goes into the tool, its inputs and prompts. And for the time spent reviewing and editing what comes out.

Read that second example twice. The check is on the meter. Hold onto that, because it comes back later in this chapter from a direction you might not expect.

Same Fee, Less Time

Now take the meter away.

Plenty of professional work isn't sold by the hour at all. It's sold by the job, at a price agreed before anyone starts, whatever the work turns out to take. A flat fee.

Lawyers increasingly want guidance on fees that aren't hourly, Virginia's opinion says. Its rule says a lawyer's fee shall be reasonable, and it lists eight factors. In value-based billing, the opinion says, the tension lies between two parts of the first one: the time and labor required, and "the skill requisite to perform the legal service properly."

AI can pull the time and labor down. Generative AI can dramatically reduce the time and labor required for certain tasks, the opinion says, such as drafting routine documents, conducting preliminary research and analyzing large volumes of data. Then it turns. In that same sentence, it says it would not be reasonable to conclude that a lawyer is ethically required to reduce or limit the fee based solely on that factor.

The skill factor? It might actually increase. Using AI effectively, the opinion says, could require specialized knowledge: to prompt the tool, verify its output, supplement it and work it into competent legal work product. Knowing when and how to use AI tools takes judgment. Judging what they produce takes expertise. The opinion calls both of those "valuable services," the kind a lawyer can reasonably be paid for.

Here's the line I'd underline.

It is "not per se unreasonable," the opinion says, for a lawyer to charge the same non-hourly fee for work done with AI as for work done without it. Same fee. Less time. Not unreasonable in itself.

Look at the word non-hourly. The whole thing hinges on it.

Any fee must still be reasonable in light of all the rule's factors, the opinion says. Over time, it allows, the market rate might drop as efficiency improves dramatically. Even so, the rule should not require a lawyer to "surrender any benefit" from those efficiency gains, as long as clients keep getting value from the lawyer's output.

There's a price for that freedom. Take a fee that rests mainly on the lawyer's skills and the value of the expected final product, rather than on time spent or on reaching a fixed end point in a case. Then the lawyer must make sure its basis is adequately explained to the client, the opinion says. That could be especially important, it says, when the tool has substantially cut the time spent on the matter.

Think of the fixed fare from the airport. Price set at the curb. Light traffic or heavy, you pay it, and nobody hands you a refund because the roads happened to be clear.

And nobody charges you extra when they weren't.

The Tax Authority's Different Answer

Now cross from law to tax. Here, as I read it, the answer sounds different.

On June 24, 2026, the IRS issued introductory guidelines for responsible AI use in federal tax practice (Issue Number 2026-19). Under the rule against unconscionable fees in practice before the IRS, the guidelines say AI can reduce research and drafting time. So billing clients for manual labor or time that was not actually spent, or double billing for AI-assisted tasks, may violate that rule, depending on the facts. Its examples of such facts are a noticeable pattern across clients, or the size of the billing differences.

Here's where, as I read them, the guidelines and the opinion part ways. Cost savings should be passed on openly, the guidelines say, with billing practices that reflect the efficiencies gained from using generative AI. Practitioners should disclose the AI activities performed, in general or specific terms as needed. And they should "fairly credit to the client's account any cost reductions."

Put the two side by side.

Virginia says the rule should not require a lawyer to surrender any benefit of the efficiency gains on a non-hourly fee, as long as clients keep getting value. The IRS says cost savings should be passed on openly. Its guidelines never mention flat fees, and that line names no kind of fee.

One says you needn't give the saving up. The other says pass it on.

Here's how I read the split. Both warn against billing time that wasn't actually spent. Beyond that, they test different things: Virginia asks whether a fee is reasonable, and the IRS rule forbids a fee that is unconscionable. And the IRS guidelines say "should" where they speak of passing savings on, and "may violate" where they speak of the rule. That's guidance with a warning in it, not a fee schedule.

As I read them, there's a third line, between those two. Lawyers may consider alternative billing structures, such as flat fees, for more predictable assignments, California's guidance says, and may consider building the time savings created by AI use into their competitive rates for such work. The saving shows up in the price, not as a refund.

If I sold work to both kinds of client, I'd write the answer into the engagement letter before the first job. Which way the saving goes. And why.

The Buyer Rewrites the Price

So far, the rule-makers. Now the buyer.

The UK government has committed to halving central government spending on consultancy, its Consultancy Playbook says. The Playbook, version 2.0, updated on November 6, 2026, is guidance from its own commercial function to the government's buyers. It has a section on AI.

AI is increasingly being used in consultancy, the Playbook says, and that could influence, among other things, how consultancies are structured, what they deliver, and the pricing models used to pay for them. The traditional "pyramid" model of consultancy may change, it says, as AI automates certain tasks typically done by junior consultants.

Then it gets to price. Time and materials is a commonly used model, it says, where the supplier charges for actual time worked, at rates agreed in advance, typically day rates by grade. However, as AI continues to drive efficiencies for consultancies, the Playbook says, the market is shifting away from pure time-based pricing. So buyers should carefully define the outputs and outcomes they want, as a move towards fixed pricing.

That's the buyer saying it out loud. On a fixed price, the meter becomes the supplier's problem. The Playbook still keeps a place for the meter: time and materials may be appropriate at the start of a project, it says, when the scope and the outcomes wanted are less clear.

Nor does a fixed price leave the hours behind. Suppliers are typically able to provide one, the Playbook says, by estimating how long the work will take and adding some contingency. And the Playbook doesn't pretend a fixed price is free of risk. A fixed price can mean lower transparency of costs, it says, with limited understanding of how the price breaks down. Output-based pricing needs the buyer to check that the work meets the deliverables before it pays. And no pricing mechanism is risk free.

It also tells buyers they can ask bidders, as part of the invitation to tender, to declare any intention to use AI in the procurement and in producing the advice or outputs.

The same government buys translation. On the page for its language services agreement, the government's buying agency lists, among the benefits for buyers, the opportunity to achieve cost and time savings using technological solutions such as machine translation, automatic speech recognition software and emerging AI solutions. Further down, it says direct awards will be priced based on maximum agreement rates. Buyers should consider whether further competition gets better value for money.

As I read that page, the savings are on offer. They don't arrive by themselves. A buyer who wants them may have to compete for them.

Nobody Bills You for the Truck

There's a second line on most bills. Not the fee. The costs.

Who pays for the tool itself?

California draws a clear line. Subscription fees for AI tools that give general office help, such as drafting assistance, research or document review, typically count as overhead. It likens them to library upkeep or general computer systems. So they should be absorbed within the lawyer's fee, rather than charged to clients separately.

A builder doesn't put the company pickup truck on your invoice, even though it drove to your job every day. The windows ordered for your house? Different story.

California makes room for the windows. A lawyer may charge clients for AI costs incurred specifically for the client's matter that go beyond general office operations. Its examples are per-use fees for specialized AI tools, and the costs of custom AI implementations developed for a particular matter.

Then come three conditions. The fee agreement must clearly disclose that AI costs may be billed separately. The charges must reasonably reflect the lawyer's actual cost. And no markup or profit element may be added without the client's informed written consent.

Here's the test I'd use. Would the firm have paid for this tool if this client had never walked in the door? If yes, it's the truck. If not, it may be a window.

When the Software Sends Its Own Bill

Health care turns the question around.

In most of this chapter, a professional uses a tool and the question is what to charge for the professional's time. In health care, the tool's own work can be the service a payer pays for.

On November 4, 2026, the Centers for Medicare & Medicaid Services issued its final rule on Medicare's payments for hospital outpatient and surgical center services in 2027. Its fact sheet speaks of software-based technologies that diagnose disease and support clinical decisions. It renames them, from Software as a Service to Software as a Medical Service.

And it says how Medicare will pay for them, for now. It assigns them to its New Technology payment groups, as an interim step toward a broader payment policy.

Interim. Toward. Those are the agency's own words, and I'd take them at face value. As I read them, Medicare hasn't settled how to pay for software's work either.

One limit matters here. The fact sheet speaks of software and algorithmic analyses. It doesn't use the word AI.

A Faster Wrench Is Not Evidence

Back to the professions, and to a courtroom.

The first fee award in the case was $114,663.11.

It came out of a trade-secrets lawsuit in federal court in Colorado. Three of the defendants were dismissed early, because the court found it had no personal jurisdiction over them. When a tort case is dismissed before trial on a defendant's motion to dismiss, a Colorado statute awards that defendant its reasonable attorney fees. They asked for theirs, and the district court awarded the full amount they had moved for.

The plaintiffs appealed. One of their arguments was that the defense lawyers should have billed fewer hours, because they relied on AI and other online research tools.

If the tool is faster, why isn't the bill smaller?

On October 30, 2026, the Tenth Circuit answered. The plaintiffs, it said, "identify no evidence" that the defense lawyers' use of AI resulted in excessive billing.

The court didn't brush the idea aside. It is possible, it said, that a firm's use of AI and other online research tools may reduce the hours required for research. That could lower the fee the court starts from. Then came the other half. As the magistrate judge had explained, ethical rules require lawyers to verify the cases they cite.

Even assuming AI and other research tools might make lawyers more efficient in the long run, it said. Even assuming they might, in some circumstances, reduce the hours needed to litigate a case. The plaintiffs still had not shown that the defense's use of AI made the hours actually spent unreasonable.

The award stood. All of it.

You can't knock down a plumber's bill because you've heard there's a faster wrench. You'd have to show that this job took longer than it should have.

And look at what the magistrate judge had weighed, among other things: that the defense lawyers tracked their time to the tenth of an hour, did no block billing, reduced hours where appropriate, and had extra work because the plaintiffs amended their pleading. Nothing on that list is about which tools were used.

When Hours Get Harder to Judge By

Now the audit world.

In April 2026, audit oversight authorities from many countries met in Paris, at the plenary meeting of the International Forum of Independent Audit Regulators. Japan's Financial Services Agency reported on it in its June newsletter.

Audit networks, the report says, are putting substantial money into technology to make audits more efficient and sophisticated. Speakers widely acknowledged that the ultimate professional judgments and responsibilities in the audit process should continue to reside with auditors.

In a panel session with members of an external advisory group to one of the forum's working groups, two points were made about fees. Participants shared the awareness that audit fees not commensurate with the costs and effort required may adversely affect audit quality. And it was noted that, as technology continues to improve the efficiency of audit work, it may become increasingly difficult to assess whether audit fees are appropriate based on traditional metrics such as audit hours.

That's a regulator reporting what was said at a meeting. Not a rule. But read it next to the Tenth Circuit.

The court still measured a fair bill in hours, because that's the method its law gives it. At the regulators' forum, it was noted that it may become increasingly difficult to assess audit fees by traditional metrics such as audit hours. And participants there shared the awareness that audit fees that don't match the costs and effort required may hurt audit quality.

So the question isn't only who keeps the saving. It's what you count when the count may mean less than it used to.

The Part the Tool Can't Sign

Put the documents side by side, and look for where they meet.

California lets a lawyer bill the actual time spent reviewing and editing what the tool produces. Virginia says the skill a fee rewards might increase, because using AI well could require specialized knowledge, including to verify its output. The IRS guidelines say any tax professional using generative AI must carefully review all documents the technology crafts. The Tenth Circuit repeats the magistrate judge's point that ethical rules require lawyers to verify the cases they cite. At the regulators' forum, speakers widely acknowledged that the ultimate judgments should stay with the auditors.

Different industries. One meeting point. The check.

Here's how I read it. The tool can take much of the drafting off the clock, and for the client that's real progress. The check stays on it. It may even be where much of the skill now sits.

Tools can help with the checking too. They can't own it. In its section on competence, California's guidance says a lawyer's professional judgment cannot be delegated to AI. It remains the lawyer's responsibility at all times, the guidance says.

So bill that judgment honestly by the hour, or price it openly as part of a flat fee. Either can work, as I read these documents. Hiding it doesn't.

Your Hours, Your Clients

One limit matters here, and I'd rather say it before someone else does. These documents differ in weight: an ethics opinion approved by a state's supreme court, a State Bar committee's guidance, a tax authority's guidelines, a government's guidance to its own buyers, a buying agency's page, a payment rule, a regulator's report of a meeting, and a court's ruling on one AI argument against a fee award, reviewed for abuse of discretion.

And some industries are missing. No bank supervisor, manufacturer's regulator, notary's or engineer's body speaking about AI and fees in its own documents could be found that this chapter could use. When they write, this chapter will need them.

Still, I think the logic travels. Any firm that sells its time, whether it bills by the hour or by the job, can expect the same question.

Picture your own firm a year from now.

Your team closes the year-end files faster than it ever has. The first drafts come from a tool. Your people spend their hours on the checking, the hard calls and the talks with clients. The invoices go out.

Then a client's finance director calls. They've read about AI. They want to know why this year's bill looks just like last year's, when the work got faster.

What do you say?

If you bill by the hour, the answer is in your time records. They show where the hours went. The checking is on the meter. Show it.

If you charge a flat fee, the answer should already be in your engagement letter: what the fee covers, and how the tool figures into it. Which tool costs go on the bill, if any. And, if you practice before the IRS, whether the saving is credited to the client.

If neither is true, that phone call is going to be hard.

Your hours don't add up anymore. They never quite did. The hour was always the meter, never the work.

The faster tool isn't a threat to a fair fee. The surprise is. So price the judgment, and show the check. And I'd say so before the work starts.

Do that, and the tool needn't shrink what you're worth to a client. It gives you more time for the part they're really paying for.

How much of your team's time now goes into checking what a tool produced? Does your fee say so? Does your client know?