In 1873, a salesman named Frank Shepard, who worked for a Chicago legal publisher, began selling lawyers stickers. They were gummed, perforated sheets (sold as “Shepard’s Adhesive Annotations”) printed with lists of every later case that had cited a given decision. You tore them apart and pasted them into the margins of your reporters, so that the book itself would warn you if the case on the page had been questioned or overruled.
A convenience, plainly. Nobody needed the stickers; lawyers had practiced for centuries without them. But the product spread, the stickers became bound maroon volumes, and the brand became a verb. Somewhere along the way, without any court or committee ever quite announcing it, the convenience hardened into an obligation. By the time our grandparents’ generation practiced law, “failure to Shepardize” was shorthand for professional negligence. The tool’s trade name had become the name of the duty.
Last week I wrote about the Model Context Protocol, the plumbing that now lets an AI assistant verify a brief’s citations against multiple independent databases in about a minute. That piece was about how the plumbing works. This one is about what the plumbing does to the standard of care. Because Shepard’s stickers were not a one-time story. They were the first run of a pattern that has repeated in every generation since, and it is repeating again right now, on a deadline. Convenience, then advantage, then norm, then duty. The ratchet turns one way. It does not turn back.
A tugboat off the Jersey coast
The rule that turns the ratchet is older than computers, and it has nothing to do with law practice. In March 1928, two tugs lost their barges in a gale off the New Jersey coast. A working radio receiver would have caught the storm warnings; the tugs didn’t carry them, and (this was the heart of the defense) neither did most of the industry. Radios on tugboats were new. Custom didn’t require them.
Judge Learned Hand’s answer, in The T.J. Hooper, 60 F.2d 737, 740 (2d Cir. 1932), is one of the most consequential paragraphs in American tort law. “Indeed in most cases reasonable prudence is in fact common prudence; but strictly it is never its measure; a whole calling may have unduly lagged in the adoption of new and available devices. It never may set its own tests, however persuasive be its usages. Courts must in the end say what is required; there are precautions so imperative that even their universal disregard will not excuse their omission.”
Read that twice, substituting your own calling. The industry’s uniform practice was no defense, because the device was cheap, available, and would have prevented the loss. A profession does not get to vote on its standard of care, and it does not get to vote no by unanimously declining the new tool.
“Nothing strategic or tactical about ignorance”
Apply that to lawyers and you get the duty to research, and the case every malpractice course teaches, Smith v. Lewis, 13 Cal. 3d 349 (1975). A divorce lawyer failed to claim his client’s community interest in her husband’s retirement benefits. The California Supreme Court affirmed the malpractice verdict against him, and its description of the lawyer’s conduct still stings. He “dogmatically asserted his theory, which he was unable to support with authority and later recanted.” He hadn’t researched the question. He had simply been sure.
The standard the court applied is the load-bearing language. A lawyer is “expected, however, to possess knowledge of those plain and elementary principles of law which are commonly known by well informed attorneys, and to discover those additional rules of law which, although not commonly known, may readily be found by standard research techniques.” Id. at 358. Even where the law is genuinely unsettled, “an attorney assumes an obligation to his client to undertake reasonable research.” Id. at 359. And the court reached back to a Ninth Circuit line it described as brutal, because it is. “There is nothing strategic or tactical about ignorance.” Pineda v. Craven, 424 F.2d 369, 372 (9th Cir. 1970).
Notice what the Smith formulation actually does. The duty is not pegged to any particular tool or any fixed quantum of effort. It is pegged to what “may readily be found by standard research techniques” (a phrase with a technology dial built into it). What was readily findable in a county law library in 1967 is not what was readily findable on Westlaw in 1995, and neither resembles what is readily findable in 2026 by a lawyer whose research tools answer questions in plain English. The doctrine was never frozen. It was designed to move, and the thing that moves it is whatever tools become standard. (The principle itself was already old when Smith applied it. The citations in that paragraph run back to 1918.)
The clicks nobody heard
In 2012, the ABA amended Comment 8 to Model Rule 1.1 to say that competence requires keeping abreast of “the benefits and risks associated with relevant technology.” The Ethics 20/20 Commission’s report was at pains to say the amendment imposed no new obligations. It was merely “a reminder.” A reminder that the majority of states have since adopted, that ethics opinions now cite as the foundation of an affirmative duty of technological competence, and that no lawyer can today disclaim.
In 2016, Florida went first in the nation and made it mandatory, with three hours of technology CLE per reporting cycle. It also added new technology language to the comment to Rule 4-1.1. Continuing education must now include “an understanding of the benefits and risks associated with the use of technology.” The Florida Supreme Court’s opinion adopting the requirement, In re Amendments to Rules Regulating the Florida Bar 4-1.1 and 6-10.3, 200 So. 3d 1225 (Fla. 2016), called the changes “straightforward amendments.” Formal notice was published inviting objections. The opinion records the response in one sentence. “The Court did not receive any comments.” A new professional duty for every lawyer in the state attached, and not one of us so much as wrote a letter.
And this Monday, June 15, 2026, brought the latest click. Amended Rule 2.515 requires everyone who signs a Florida court filing to certify that the authorities cited in it “exist and are accurately cited.” I wrote about its interpretive problems in May. For present purposes the point is simpler. It is the first Florida rule of general practice written in the shadow of generative AI, and it converts citation verification from good hygiene into a signed representation, sanctionable when false. From Shepard’s stickers to a compelled certification took 153 years, and every step of the way was described, accurately and misleadingly at once, as nothing new.
The click happening now
Which brings us back to the plumbing. Last week I described an agent cross-checking every citation and quotation in a draft against two independent databases, in a minute, for pennies. That workflow is the radio receiver on the tugboat. It is new. It is not yet custom. Most of the calling has not adopted it, and a lawyer who skips it today violates no rule and deviates from no norm.
That is exactly what the masters of the Montrose and the Hooper could have said in 1928.
Nobody is going to mandate the tool, and nobody needs to. The ratchet turns because the tool changes what the old words mean. Reasonable research. Readily found. Practical verification. When confirming that a quotation actually appears in an opinion took an afternoon in the stacks, a missed misquote was an understandable casualty of practice economics. When it takes one instruction to an assistant that any lawyer can run, the same miss starts to look like the tug that sailed without the radio. The precaution became imperative the moment it became trivial. Rule 2.515 doesn’t require you to use AI verification. It is studiously tool-agnostic. It just requires you to certify the thing AI verification is the only cheap way to be sure of. The rule sets the certification; the technology sets the price of making it honestly; and the gap between them is where the standard of care now lives.
Two consequences follow, and neither is comfortable.
First, the ratchet binds the lawyers who never adopt the tool. The standard of care is comparative. It asks what a reasonably prudent practitioner does, and once cheap verification is common, the lawyer who declined it is no longer being measured against 2020. You can refuse the technology. You cannot refuse the standard it resets. A whole calling may have unduly lagged.
Second, the symmetry. Generative AI created the fabricated-citation problem; agentic verification is its cure; and it is the cure, not the disease, that becomes compulsory. Most commentary still frames AI as a risk for lawyers to manage. The history reads differently. AI is also becoming a duty, one click at a time, each adopted without comment, each sold as a reminder of obligations we already had.
Where the ratchet stops
Honesty requires the caveat, and Hand supplied it himself in the passage’s opening clause. In most cases, reasonable prudence is common prudence. Courts do not measure lawyers against every gadget on the market, and no Florida judge will sanction a lawyer next month for failing to wire an AI agent into her drafting workflow. Standards of care move at the speed of custom, which is slow, and proportionality still matters. The diligence owed in a slip-and-fall answer is not the diligence owed in an appellate brief.
But 2.515 is a warning that the slow path is not the only path. A standard of care usually hardens over decades of verdicts. A rule amendment hardens overnight, statewide, by per curiam opinion. Florida has now shown it will move citation integrity by rule rather than wait for the malpractice cases. And rules, unlike customs, have effective dates.
What Frank Shepard actually sold
The stickers are long gone. The maroon volumes became flags in a database; the flags are becoming agents that read the cases themselves. Every form of the product was sold the same way Shepard sold the first one, as a convenience, a time-saver, an edge. And every generation of lawyers who bought the convenience discovered, a little too late to object, that they had bought a duty.
I closed last week’s piece by observing that protocols outlast products. The corollary is the thesis of this one. Duties outlast both. The tools will keep changing names. The ratchet only knows one direction.
The authorities: The T.J. Hooper, 60 F.2d 737, 740 (2d Cir. 1932); Smith v. Lewis, 13 Cal. 3d 349, 358–59 (1975); Pineda v. Craven, 424 F.2d 369, 372 (9th Cir. 1970) (quoted in Smith); ABA Model Rule 1.1 cmt. 8 (2012); In re Amendments to Rules Regulating the Florida Bar 4-1.1 and 6-10.3, 200 So. 3d 1225 (Fla. 2016); In re Amendments to Florida Rule of General Practice and Judicial Administration 2.515 (Fla. 2026), effective June 15, 2026. Part one of this series is A Lawyer’s Guide to the Model Context Protocol.
Next in this series: The Library Has No Librarian, on who vouches for the servers a lawyer connects, and what the confidentiality rule already requires.