What the $285B AI Crash Means for Planning Technology

What Happened

Anthropic released 11 starter plugins for Claude Co-work, their enterprise AI workspace. One was a legal contract review plugin—a structured markdown file, roughly 200 lines, that could review contracts with reasonable accuracy.

Within 48 hours, Thompson Reuters lost 16% of its market capitalisation (its largest single-day decline on record). RELX, which owns LexisNexis, dropped 14%. LegalZoom fell 20%. Private equity firms with legal tech holdings lost approximately 10% each.

The total damage: roughly $285 billion in market value, erased in two days.

Why It Mattered

The crash wasn't caused by the plugin being better than existing legal research tools. It exposed a structural vulnerability: per-seat licensing breaks when AI agents replace human users.

Thompson Reuters sells Westlaw at roughly $200 per user per month. If one AI agent can do the preliminary research that previously required ten paralegals, nine seats of revenue disappear—even though the underlying legal data remains valuable.

The market wasn't pricing in whether the plugin was good. It was pricing in what happens when AI agents become the primary consumers of professional knowledge databases, and those databases are priced for human users.

The Pattern: UI-First vs Data-First

The companies that lost value shared a common architecture: they built interfaces designed for humans to navigate, charged per human who navigated, and assumed the interface was the product.

But in an agentic world, the interface is irrelevant. AI agents don't need search bars, dropdown filters, or paginated results. They need structured data with provenance—an API that returns exactly what they ask for, with citations they can verify.

The companies that survived (and will survive) are the ones whose actual product is the data, not the interface to the data. Their value proposition shifts from “we built a nice UI for searching legal documents” to “we have 40 years of structured, verified, citable legal data that any system—human or AI—can query.”

What This Means for Planning

Planning technology is heading toward the same inflection point. Today, planning professionals use tools like council portals, GIS systems, and compliance databases through graphical interfaces designed for human navigation. These tools charge per user or per organisation.

As AI agents become standard in architectural practices, development consultancies, and council planning departments, the question becomes: is your planning data accessible to agents, or only to humans clicking through a portal?

Planning data providers face two possible futures:

The “Prompt Test”

Here's a simple test for any planning technology company: could someone write an AI prompt that replaces your core product?

If your product is a nice UI on top of publicly available data—council DA portals, NSW ePlanning API, ABS statistics—then yes, a well-crafted prompt could approximate it. That's the Thompson Reuters problem.

If your product is proprietary structured data that doesn't exist elsewhere—parsed and verified planning provisions with source citations, spatial relationships between properties and precinct-specific controls, heritage provision chains with cross-references mapped—then no prompt can replicate it. The data is your moat, and it compounds over time.

The Data Edge and the Accountability Edge

Two competitive advantages survived the crash unscathed:

The data edge: Decades of structured, verified, domain-specific data that no LLM has in its training set. An AI model can reason about planning law generally, but it cannot tell you the specific setback requirements for a heritage conservation area in a particular NSW council's DCP, Section 4.3.2(b), page 47. That requires structured extraction from the source document.

The accountability edge: In planning, someone needs to be responsible for the accuracy of the data that informs decisions affecting property rights. An AI plugin has no professional indemnity insurance. It has no ABN. You can't subpoena it. Councils and planning professionals need what the industry calls a “ringable neck”—a vendor who stands behind their data with contractual accountability.

The Hallucination Problem in Planning

The $285B crash happened in legal tech, but the implications are sharper in planning. A hallucinated legal clause in a contract review is caught by a lawyer before signing. A hallucinated planning provision could lead to:

• A development application based on controls that don't exist
• A compliance assessment that misses applicable heritage provisions
• A feasibility study that assumes incorrect building envelopes
• A council assessment report citing provisions from the wrong DCP version

The stakes are concrete: built structures, property values, community amenity. This makes the distinction between generative AI (which can hallucinate) and deterministic data retrieval (which cannot) a genuine safety question in planning technology.

What Comes Next

The agentic era is arriving in planning technology whether the industry is ready or not. Gartner projects that 40% of enterprise software will include AI agents by the end of 2026. The Model Context Protocol (MCP), which standardises how AI agents connect to data sources, already has implementations in US property data (ATTOM launched their MCP server in January 2026).

Australian planning technology companies have a window to prepare:

• Build API-first: Every feature should be callable by an agent, not just clickable by a human
• Price for usage, not seats: Per-query, per-project, or outcome-based pricing scales with agent adoption
• Structure data with provenance: AI agents need citable, verifiable data sources. “Trust me” doesn't work when agents are making decisions at scale
• Maintain the accountability edge: Australian company, Australian data, Australian law. Contractual liability. Professional indemnity. The things an offshore AI plugin cannot provide

The $285 billion wasn't destroyed. It shifted—from companies whose architecture assumed human users to companies whose architecture serves both human and AI users. The planning technology companies that understand this distinction will capture the value that the others lose.