The $285 Billion Crash Wall Street Won't Explain Honestly. Here's What Everyone Missed.

A 200-line open-source prompt template for legal contract work helped trigger a rapid repricing of enterprise software—about $285 billion in market value—because it made one uncomfortable truth impossible to ignore: the per-seat “pay for every human who touches the tool” pricing model is cracking under AI automation. The legal plugin for Anthropic’s Claude Co-work (released Jan. 30) can triage NDAs, flag non-standard clauses, and generate compliance summaries. It shipped as readable, structured markdown with roughly 200 lines of workflow logic and a disclaimer that outputs require review by licensed attorneys. The market reaction was swift and brutal: Thompson Reuters fell 16% in a day, RELX (parent of LexisNexis) dropped 14%, and LegalZoom sank about 20%, with private-equity-linked software names also sliding.

But the markdown file itself wasn’t the real cause. It functioned more like a spotlight than a spark—revealing that the industry’s financial foundation was already under strain. Enterprise software has long relied on per-seat SAS licensing, where revenue scales with headcount and forecasts depend on humans logging in and performing tasks. AI agents threaten that assumption because they can execute knowledge work without the same number of billable or licensed “touches.” The transcript argues that Wall Street’s earlier hesitation wasn’t about whether AI could do the work; it was about whether the market had priced in how quickly AI changes the economics of access.

Several signals were already pointing to a valuation reset. Software’s forward price-to-earnings multiple had compressed sharply—from around 8x to about 2x—matching the largest four-month valuation compression since the dot-com bust. Earnings season was also described as unusually weak, with software companies missing revenue estimates at rates not seen since the post-COVID correction. In that context, the Anthropic release didn’t start the fire; it made the burning visible.

The transcript also pushes back on a common counterargument: that AI will replace software. Jensen Huang’s claim at a Cisco AI Summit—that AI doesn’t replace software but runs on it—gets credit for being directionally right, but the market debate is framed as different: not whether software is needed, but whether buyers will keep paying for software in the same way. The analogy is print media: content survived the internet, but the access-and-bundling business model collapsed. Similarly, proprietary data and workflow “moats” may remain valuable, while the pricing layer over them becomes harder to justify.

A key example of that shift is a quieter operational event: KPMG pressured Grant Thornton UK to pass on AI-driven cost savings from audits. Grant Thornton resisted on grounds that high-quality audits rely on expert human judgment and that fees reflect people costs. KPMG’s response—lower prices or face a new auditor—worked. Grant Thornton’s international audit fees reportedly fell from $416,000 in 2024 to $357,000 in 2025, illustrating how AI becomes leverage in fee negotiations even without fully replacing staff.

The transcript concludes that survival depends on whether SAS incumbents rebuild for an “agentic-first” world—charging for value tied to data and accountability rather than per-seat access. It warns that bolting AI onto existing interfaces won’t be enough, because the deeper crisis is resource allocation: engineering teams maintaining one-size-fits-all platforms may not have time to build agentic workflows. The same dynamic is extended to individuals: using AI as a bolt-on (summaries, proofreading, unchanged workflows) may feel productive, but it doesn’t address the structural shift required to keep up as capabilities accelerate every few days.