AI agents are arriving in property transactions. Not chatbots, but autonomous systems that complete tasks across every participant in the chain. In the first of a weekly series, Ed Molyneux explains why the technology’s ‘jagged frontier’ of capability means the real challenge isn’t intelligence – it’s architecture.
A few days ago, a clip went viral. Someone asked GPT-5.4, OpenAI’s latest and most powerful model: “The car wash is 100 metres away and I need to get my car cleaned. Should I walk or drive?”
It wrote a confident, well-structured essay recommending walking. Health benefits. Environmental impact. The irony of driving a dirty car just to clean it.
Completely wrong. You need the car at the car wash.
If you’ve been uneasy about clients asking ChatGPT for a second opinion on your advice, I understand. But the car wash problem should focus your attention on something more specific than general anxiety about AI. These systems are superhuman at some tasks and bizarrely incompetent at others, in ways that don’t match our intuition about what’s hard.
The jagged frontier in property
Researcher Ethan Mollick calls this the “jagged frontier” of AI capability. The same model that fails the car wash question can outperform doctors at differential diagnosis. It can synthesise 12 medical meta-reviews in two days, work that normally takes 12 person-years. The frontier is growing fast, but it’s growing unevenly. Reading, reasoning, and knowledge are improving rapidly. Common sense, memory, and real-world context remain unreliable.
For conveyancers, this matters in very specific ways. AI can read a title register and extract key information faster than any human. It can summarise a lease pack in seconds. But can it reliably spot that a poorly-scanned 1960s restrictive covenant actually prohibits the extension your client’s buyer is planning? Can it distinguish between a covenant modified by a subsequent deed and one that hasn’t been? Can it know when the answer isn’t in the document at all, and that the right response is ‘insufficient evidence’ rather than a confident guess?
Superhuman speed, but unpredictable reliability on the judgments that actually matter.
The agent era has arrived
And while we’re still debating whether AI is reliable enough, the agent era has already arrived.
An AI agent isn’t a chatbot. It’s an autonomous system that takes a goal, breaks it into steps, coordinates with other systems, and executes without waiting for human instruction. It doesn’t answer questions. It completes tasks.
In the last few weeks alone: OpenClaw, the open-source agent framework, hit 247,000 GitHub stars, letting anyone run persistent AI agents on their own infrastructure around the clock. Anthropic launched Claude Cowork with enterprise connectors for Google Workspace and DocuSign, a system the financial press is calling the ‘SAASpocalypse’ because it replaces functionality entire software companies were built to provide. Perplexity launched Computer, coordinating 19 AI models as autonomous sub-agents running complex workflows for weeks at a time.
And it won’t just be conveyancers using them. Estate agents will deploy agents that generate listing copy and match buyers to properties. Lenders will run agents that verify property data against handbook requirements in seconds rather than weeks. Mortgage brokers, surveyors, insurers: every participant in the transaction will have AI agents working on their behalf.
The architecture question
When every participant has an agent, those agents need to communicate. They need to read from the same data. They need to trust what they’re reading. And right now, property data in England and Wales isn’t ready for that.
So the challenge isn’t whether AI is clever enough. It is. The challenge is architecture. How do you combine the speed and breadth of AI with the reliability that professional services demand? How do you ensure that when an agent tells your client’s buyer there are no restrictive covenants, someone can prove where that conclusion came from? ‘The AI said so’ won’t satisfy your professional indemnity insurer.
The answer lies in what we call agentic diligence: deterministic rules for things you must get right every time, intelligent analysis where AI reasoning genuinely helps, and clear provenance showing where every conclusion came from and what evidence supports it.
Over the coming weeks, I’ll lay this out in detail. What AI can and can’t do with your case files. Why provenance matters more than ever. Why property needs a shared system of record. And what agentic diligence means for your practice in concrete terms.
Let’s start.
About the author
Ed Molyneux is co-founder and CTO of Moverly, the property intelligence platform working with LMS and Connells Group to bring structured, verified data to property transactions. He is the architect of the Property Data Trust Framework (PDTF), the open standard for machine-readable property data now being adopted across the industry. Ed writes about AI, property data infrastructure, and the future of conveyancing.
