The AI Staffing Problem Isn’t What We Expected

In March 2023, Goldman Sachs estimated that generative AI could automate 44% of tasks performed in legal occupations.

The headline number drew enormous attention because the implication seemed obvious: if technology could do nearly half of the work, law firms would eventually need far fewer lawyers.

Three and a half years later, that hasn’t happened. As of September 2026, the U.S. Department of Labor reports U.S. legal-sector employment at a record 1.249 million jobs, up 7.6% over five years. While AI use has increased rapidly, widespread lawyer displacement has not followed.

Three years is still a short window, especially with AI developing this quickly.

The 44% figure became misleading when it was treated as a direct proxy for headcount.

The more immediate staffing problem is uncertainty: AI is making it harder to know how many lawyers a given amount of legal work will require.

Legal hiring already shows signs of that uncertainty. According to NALP, firms with more than 500 lawyers hired 7.5% fewer entry-level associates from the Class of 2025, their first decline since 2014. Summer associate hiring has declined as well. In a recent Citi survey, two-thirds of large firms said they expect fewer first-year associates in 2028 than in 2025.

Normal hiring cycles and broader economic conditions also factor in. AI adds another reason firms making hiring decisions today have less certainty about how many junior associates they will need several years from now.

A Second Forecast

Law firms have always had to forecast demand. Leaders consider economic conditions, client activity, competitive behavior, and other factors when deciding how aggressively to hire.

Those forecasts have always been imperfect, but firms historically had a reasonably stable relationship between demand and staffing. If a practice expected a certain volume of work, it could reasonably estimate the lawyer hours needed to handle it.

AI adds another forecast: how much attorney time will that work require?

Consider an M&A practice that expects transaction volume next year to look much like this year. That may provide a reasonable view of demand without providing the same clarity on staffing needs.

If AI reduces the time required for diligence and document work, including first drafts, fewer associate hours may be needed. Negotiation and client counseling—the work that depends more heavily on judgment—may change much less. Gains will also vary from one matter and client to the next, depending on how effectively each team uses the technology.

No single number will tell a managing partner how much more productive AI has made the firm. Significant gains in one category of work may coexist with almost no change in another.

Greater efficiency could also create more work. Earlier this year, I wrote about why AI may increase legal demand rather than reduce it. If some legal services become faster and less expensive, clients may use lawyers for work they previously handled themselves or chose not to pursue.

The percentage of tasks AI can automate therefore tells us surprisingly little, on its own, about how many lawyers a firm will need. Even if Goldman’s 44% ultimately proves accurate, we still won’t know what it means for headcount until we know what happens to the capacity AI frees up.

Talent and Technology Move at Different Speeds

I also wrote earlier this year about the tension between the traditional law firm talent model and increasing volatility in client demand. That problem becomes more complicated when the productivity of the lawyers doing the work is changing at the same time.

Law firms make talent decisions years in advance. They recruit associates well before they know what demand will look like. Lawyers take years to develop. Lateral hiring is expensive, and successful practices depend on teams that build client knowledge and expertise over time.

AI can change materially in months, which means firms must make long-term hiring decisions while one of the variables affecting those decisions moves much faster.

A firm hiring aggressively into strong demand could discover two years later that technology has reduced the attorney hours required to handle a meaningful portion of that work. Conservative hiring based on expected AI efficiencies carries its own risk if demand grows, client expectations expand, or the technology has little effect on the practices driving growth.

The cost of getting it wrong is increasing. In the second quarter of 2026, according to Thomson Reuters Institute’s Q2 2026 Law Firm Financial Index Analysis, direct expenses, which include compensation and staffing costs, rose 8.3% and overhead increased 7.7%, both well ahead of demand growth.

As those costs rise, permanent hiring decisions carry greater financial implications, particularly when future demand and productivity are difficult to predict.

Reduce the Cost of Uncertainty

Measuring where AI is actually reducing hours and changing staffing needs can help firms make more informed decisions. But even that has its limits because the technology keeps changing.

Firms can also create more options.

Permanent lawyers will remain the core of successful firms. Client relationships, expertise, and institutional knowledge still take years to build. At the same time, firms have opportunities to complement their permanent teams with experienced flexible attorneys as needs evolve.

Flexible legal talent gives firms another way to structure their teams, bringing in experienced attorneys who work alongside permanent team members and contribute to ongoing work. This allows firms to expand their teams while maintaining flexibility as they evaluate how demand and productivity may evolve.

We’re seeing this uncertainty in our own conversations with law firm leaders. Many want to add experienced attorneys to their teams now while continuing to evaluate their longer-term permanent hiring needs as demand and productivity evolve.

I’ve described flexible talent before as a means to buy time and information. In this environment, I think one aspect of its value can be stated even more simply: it reduces the cost of a bad forecast.

The legal industry has spent much of the past three years debating how many lawyers AI will replace, and that debate will continue.

Meanwhile, law firms must make talent decisions now, long before anyone knows the answer.

AI may change how much legal work clients want and how much lawyer time it takes to do that work. Those effects will vary across practices and won’t happen on the same timetable.

The practical challenge is building a talent model that combines the continuity of permanent teams with the flexibility to adjust as the answer takes shape.

The firms best positioned for this transition will be the ones that can afford to be wrong.