Gimmick or desk skill? What a New York general counsel can actually test.

Two finalists is an interview. A department already in the tens is a training plan. A New York general counsel who collapses those two numbers onto a single scorecard is not testing a skill. That person is inventing a census the calendar will not hold.

Sartori’s research program maps nearly 1.5 million lawyer profiles globally and has run quarterly market surveys since 2019. The New York in-house findings on this page sit on that instrument. We have worked the New York in-house market for more than ten years, for listed issuers, sponsor-backed platforms, asset-manager legal desks and DFS-regulated financials. Over the trailing three years the New York in-house files ran on a four-to-seven-month clock, with a median of 16 working days from offer to signature. The unit of that work is the open seat. It is not the department.

The Bureau of Labor Statistics counted 90,980 lawyers in the New York-Newark-Jersey City metropolitan statistical area in the May 2023 Occupational Employment and Wage Statistics, at an annual mean of $213,420. That is the wage-and-salary stock across firms, companies and government. It is not a count of people a chief legal officer can put through a live writing exercise on Tuesday. BLS does not publish a Standard Occupational Classification for “prompt engineer”; the Handbook still prices lawyers (national median $151,160 as of May 2024, 864,800 jobs in 2024) and, if the company hires a non-lawyer builder, software developers (national median $133,080 as of May 2024). There is no public pay band for the skill the spec is reaching for. The comparable measured bands are the in-house title the hire will hold, and the developer seat if the company is actually buying a builder.

CLOC’s 2025 State of the Industry report, built on Harbor’s 2024 wave of 186 organizations, put overall median attorney headcount at 44, on a median legal department of 77 FTEs. Attorney headcount climbs with company revenue; the bands sit in the chart below. A general counsel filling one counsel seat against a desk that already numbers in the tens is not short of lawyers to “test.” That person is short of interview hours. The Thomson Reuters / Georgetown 2026 Report on the State of the US Legal Market, released 7 January 2026, recorded that nearly 90 percent of general counsel report resource limitations preventing the strategic impact their organizations expect. The scarce resource is the hour, not a prompt-syntax census.

Of those 286 general counsel, deputy general counsel and heads of legal in the same New York cohort, 204 said every practicing lawyer who would use a sanctioned tool needed to be able to verify output. Only 41 said they would score prompt authorship as a gate on a commercial or contracts counsel hire. That is the split this page is about. The first number is a duty the ethics opinions already impose. The second is a craft that belongs on a handful of hybrid seats. One hundred and twelve of the same 286 said they had never sat a live writing exercise with a counsel candidate at all. The fluency line on the spec is running ahead of the interview the company has actually designed.

Of 24 closed New York in-house searches over the trailing three years, 9 were commercial or contracts counsel, 6 privacy or cybersecurity, 4 funds or product-regulatory, 3 employment or HR legal, and 2 legal-operations or knowledge hybrids. A scored writing exercise appeared in 3 of those 24 files: both hybrids, and one privacy seat where the general counsel wanted to watch the candidate refuse to paste a DPIA into a public model. The other 21 files were practicing-lawyer seats. Syntax was not the work. Judgment was.

A general counsel at a New York-headquartered global asset manager told Sartori they could sit two finalists with a live writing exercise, and that scoring the rest of the desk was a fantasy of the requisition, not of the calendar. That is the testimony that matches the count. The in-house and general counsel search we run is built on that unit: one open chair, a designed exercise, a human score. It is not a department-wide certification.

Local Law 144 of 2021 took effect on 1 January 2023; the Department of Consumer and Worker Protection began enforcement on 5 July 2023. Employers and employment agencies in New York City may not use an automated employment decision tool unless an independent bias audit was completed within the prior year, a summary of results is publicly available, and notice goes to City-resident candidates or employees at least 10 business days before use. An AEDT is a computer-based tool that uses machine learning, statistical modeling, data analytics or artificial intelligence and that substantially assists or replaces discretionary decision-making in hiring or promotion screening. It applies if the job is in a New York City office, is fully remote but tied to a City office, or the employment agency is in the City. Civil penalties run $500 to $1,500 per day. DCWP enforces the audit and notice rules; discrimination claims go to the City Commission on Human Rights.

A general counsel sitting across the table with a red pen is not an AEDT. An automated ranker of prompt quality is. The interview design is therefore a Local Law 144 question before it is a skill question. The State Comptroller’s audit of DCWP’s enforcement, Report 2024-N-6, December 2025, covering July 2023 through June 2025, found DCWP’s 311 complaint routing ineffective; despite only two AEDT complaints in the period, DCWP did not test whether intake worked. DCWP surveyed 32 company websites and found one non-compliance issue; auditors reviewing the same set identified at least 17 potential violations. Companies that automated any slice of legal hiring in New York City should expect counsel time on audit design and website summaries in 2026. They should not expect a new prompt-occupation title to appear because of it.

Among larger companies, 26 percent of chief legal officers in ACC’s 2025 CLO Survey cited concern about AI regulatory enforcement, behind antitrust at 41 percent in that cut. Twenty-three percent of organizations had been subject to a regulatory investigation or enforcement action in the prior year. The RAISE Act, signed 19 December 2025 and effective 1 January 2027, covers frontier-model developers with more than $500 million in revenue, creates an oversight office inside DFS, and gives the Attorney General civil penalties of up to $1 million for a first violation and $3 million for a subsequent one, on IAPP’s 22 December 2025 read of the signed text. That is a product and regulatory file at covered developers, plus the DFS RAISE seats already posted in New York City. It is not a mandate to test every corporate counsel candidate on prompt syntax.

The capital cycle on the tools is moving faster than the job architecture. Harvey raised $200 million at an $11 billion valuation on 25 March 2026, as Reuters reported that day, and $550 million at a $15.5 billion valuation on 9 September 2026. Those rounds buy agents and vendor legal-engineering teams. They do not create a Census code, and they do not open attorney seats. Harbor’s 2025 sample is the other half of that sentence: departments are buying governance faster than they are hiring lawyers.

The finding that does not flatter us sits in the mandate file. In 5 of the 9 commercial and contracts searches, the company added an unsourced “AI fluency” screen after the first round. Three of those five required a second shortlist, because the parlor trick filtered out commercial lawyers who had never authored a production prompt and who were otherwise the hire. Those five files ran to the long end of the four-to-seven-month window. The two legal-operations hybrids that designed the exercise before the first round closed inside four months. We had to rerun slates we should have underwritten against a designed exercise, not against an adjective. New York in-house counter-offer incidence on this line is 28 percent; the candidates who drew them were the ones who had already sat a competing company’s verification exercise, not the ones who had written a clever prompt in ours.