Automating Oral Argument (adamunikowsky.substack.com)

A practicing Supreme Court advocate ran an experiment asking whether a “robot lawyer” could handle real-time oral argument. He fed his case briefs and precedents into Claude 4.0 Opus, prompted it to play an appellate advocate, and had it answer the actual questions posed by Justices in his Williams v. Reed argument. He then used ChatGPT’s Advanced Voice Mode and an ElevenLabs v3-generated appellate voice (derived from his own voice) to produce audio, splicing AI answers into the real Court question audio. The pipeline—document context + LLM responses + voice synthesis + minimal editing—produced a complete AI-only oral argument with essentially no prep time required. The result: the AI’s answers were clear, responsive, and often tactically clever; it never stumbled, knew the record, and handled hostile hypotheticals (e.g., an irrelevant Twenty-First Amendment prompt) with composed, multi-point replies. The author argues AI’s speed, short-latency reasoning, and immunity to human frailties make it likely to outperform most human advocates on the short timescales of oral argument. Limitations remain—LLM hallucinations and occasional misinterpretation—yet these are mitigable by loading source documents and careful prompts. The experiment suggests courts should consider permitting AI-assisted or autonomous oral advocates, while policymakers must reckon with authentication, malpractice, and deepfake risks as voice synthesis and LLMs rapidly improve.