Are We Training a Generation of Lawyers Who Can't Think Without AI?

According to a study published by The Decoder detailing a National Bureau of Economic Research (NBER) working paper, artificial intelligence dramatically speeds up work for early-career professionals, but it comes with a hidden cost. In a three-month field experiment across eleven U.S. intellectual property law firms involving 133 patent attorneys, researchers found that junior lawyers using Google Labs' unreleased "InFlow" assistant produced significantly better patent drafts.

However, when the AI was taken away during unassisted review tasks known as redlining, juniors showed no lasting performance gains. In fact, their independent results split wider into poor and erratic scores, while only senior attorneys retained their edge. The data shows that baseline expertise is necessary to transform AI-assisted practice into durable, internal skill.

What happens when the safety net is removed?

The experiment used a clever design to separate immediate output from actual learning. Attorneys first drafted patent claims with AI, then evaluated existing drafts without it. While juniors closed the performance gap against seniors during the AI-assisted drafting phase, they stumbled completely during the unassisted review. Instead of fixing legal flaws, many junior lawyers merely spotted problems and left notes that read like instructions to an AI assistant.

This cognitive outsourcing mirrors findings in other technical fields. Studies in software development and consulting show that professionals who rely entirely on AI during their learning curve score worse on follow-up knowledge tests. Entry-level workers traditionally build their foundation by tackling textbook knowledge on the job—precisely the domain where AI now replaces human effort.

How to manage AI use in professional routines

If your team relies on generative tools for technical drafting or legal work, productivity gains on Monday morning can mask severe skill erosion over three years. To protect your workflow, reorganize supervision and training with clear boundaries:

• Enforce unassisted practice: Schedule regular drafting and review sessions where junior staff work entirely without AI to test independent judgment.
• Audit feedback loops: Ensure that when an AI tool suggests a correction, human operators explain the underlying legal or technical principle rather than just accepting the output.
• Balance delegation: Do not assign 100% of routine drafting to automated agents; keep early-career workers engaged in the core reasoning steps.

The verdict: who should change their workflow?

Law firms, tech companies, and professional service providers must stop treating AI as a pure efficiency multiplier. While it cuts drafting time instantly, failing to mentor junior staff outside the safety net of automated prompts creates a fragile workforce. Senior professionals retain their advantage because they already possess deep baseline expertise; juniors who skip that grueling phase may never build it.