The author distinguishes between current AI deployments—session-based "copilots" designed to generate answers to individual prompts—and what enterprises actually need: stateful systems embedded into workflows that execute decisions, adapt to changing conditions, and maintain compliance. The proposed shift moves AI from advisory tools to "systems of action" that own outcomes rather than merely suggest them, with persistent memory, continuous learning, and real-world operational constraints built in.
The timing reflects broader disillusionment with enterprise AI adoption in 2026. Attorneys should monitor this emerging consensus on AI implementation failures, particularly as it affects contract review, compliance automation, and litigation support tools. Organizations deploying generative AI for legal work should assess whether their systems track decisions and outcomes across matters, maintain context between sessions, and enforce firm-specific constraints—or risk joining the 95 percent that fail to generate measurable returns.