Your AI Knows Your Documents. It Doesn't Know Your Project.The missing layer between document search and real project intelligence - and why it matters more than which AI you're using.

Earlier this year, Klarna - the fintech company - reported that its AI customer service agent had handled 2.3 million conversations, cut resolution times from 11 minutes to two, and saved the company $60 million.

Then customers started leaving.

The AI was extraordinarily good at resolving tickets fast. The problem was that resolving tickets fast was the wrong goal. Klarna's real objective was building lasting customer relationships in a competitive market. Those are profoundly different things, and the human agents who understood the difference - the ones who knew when to bend a policy, when to spend three extra minutes because a customer's tone said they were about to walk - had already been let go.

The institutional knowledge that mattered most had never been written down. The humans just knew.

If you work in commercial real estate - whether you're underwriting acquisitions, managing construction, running asset operations, or navigating entitlements - that story should sound uncomfortably familiar. Not because you're deploying customer service bots, but because the same structural problem exists at the center of how our industry handles project intelligence. And it's about to get a lot more expensive to ignore.

The Three Layers of AI Capability

The tech industry has been moving through a progression that matters for anyone managing complex, document-heavy work in CRE.

The first wave was prompt engineering - learning to talk to AI tools effectively. This is where most of our industry still sits: drafting RFP responses with ChatGPT, summarizing meeting notes, asking an AI to review a lease abstract. Individual productivity gains, usually around 30%. Useful, but limited.

The second wave, happening right now, is context engineering - giving AI systems access to your actual project data. This is the retrieval-augmented generation (RAG) approach: connect your document library so the AI can search across your contracts, reports, and correspondence instead of relying on its general training data. Significantly more useful than a blank chat window.

But here's where it gets interesting. A growing number of enterprise AI leaders are realizing that context alone isn't enough. Giving an AI access to your documents doesn't mean it understands your project.

The emerging third wave is being called intent engineering - the practice of encoding organizational purpose, decision boundaries, and tradeoff hierarchies into AI systems so they don't just know what to search, but understand what the organization is trying to achieve and what matters most when priorities compete.

What This Looks Like Across Commercial Real Estate

The intent gap shows up differently depending on where you sit in the CRE lifecycle, but the pattern is the same: the knowledge that drives good decisions lives in people's heads, not in systems.

In acquisition and due diligence, a team might upload hundreds of documents to an AI-powered data room - leases, title reports, environmental assessments, rent rolls, service contracts, zoning analyses. A good search tool can find every paragraph that mentions an environmental concern. But the question the acquisitions lead actually needs answered is closer to: "Across the Phase I, the estoppels, and the lease amendments, are there any environmental liabilities that aren't captured in the seller's representations - and do any of them affect our underwriting assumptions on Parcels 3 and 7?" That question requires understanding how documents relate to each other, which version of a lease is current, and how a finding in one report changes the risk profile in another. No keyword search gets you there.

In construction management, the challenge is tracking dependencies across hundreds of moving parts. A GC's pay application references a schedule of values, which ties to subcontractor contracts, which tie to approved change orders, which tie to the owner's budget line items and lender draw requirements. When a project manager asks, "Can we approve this draw?" the answer depends on tracing a chain across six or seven document types and confirming that nothing has changed since the last cycle. Today, that chain lives in the PM's head and a patchwork of spreadsheets.

In asset management, the knowledge that matters most is often the least documented. Which tenants have renewal options approaching? Do any of the HVAC service contracts conflict with the terms of the new green lease rider? Is the property tax assessment consistent with what was assumed in the hold proforma, or has the basis shifted since the last reassessment? The answers are scattered across lease files, vendor contracts, tax records, and operating budgets - and the person who knows how all of those connect is usually exactly one person on the team.

In entitlements and development, the complexity compounds further. A single large project can generate 400+ Conditions of Approval, each with its own timing trigger, responsible party, and verification requirement - plus a separate set of mitigation measures from the environmental review, plus additional obligations from any litigation settlements. The compliance chain from an environmental finding to a vendor's contracted scope to a permit milestone is something only an experienced PM can trace, and it's never captured in a single document.

Building the Intent Layer

This is precisely what we're building with Docuity at Project-Path.

Docuity isn't another document search tool. It's a knowledge architecture designed to encode the dependency chains, document hierarchies, compliance requirements, and decision logic that experienced professionals carry in their heads - and make them persistent, structured, and queryable across the full CRE project lifecycle.

When Docuity processes a project's document set, it doesn't just index text. It builds a knowledge graph that maps the relationships between entities: which lease provisions affect which operating assumptions, which contract terms tie to which budget line items, which regulatory conditions have which milestone deadlines, and where the gaps are between what's required and what's actually covered.

It understands document hierarchy - that a lease amendment supersedes the original terms, that the most recent agency correspondence on a topic is the one that governs, that a change order modifies the contract value downstream.

And it tracks temporal changes, because CRE projects are living processes that span years. When a lease is amended, a contract is modified, or a regulatory requirement changes, the system preserves the full history while surfacing the current state. The question "What changed since we closed?" or "What's different from what we underwrote?" is a real question that professionals ask constantly - and it requires a system that understands time, not just text.

One of the sharpest observations from the emerging conversation around intent engineering is what researchers are calling the "two cultures problem." The people who understand organizational strategy are not the people building AI systems. And the people building AI systems are not the people who understand how organizations actually make decisions.

In CRE, this gap is enormous. Generic enterprise AI vendors can build powerful search tools, but they don't understand that a tenant's estoppel certificate and the underlying lease can tell two different stories - and knowing which one governs requires judgment, not just text retrieval. They don't know why a GC's pencil draw and the lender's draw requirements aren't the same thing, or how a single missed Condition of Approval can hold up a certificate of occupancy for months.

That domain knowledge isn't something you can hire a prompt engineer for. It comes from decades of managing acquisitions, construction projects, entitlement processes, and asset operations - and then translating that expertise into system architecture.

This is why Project-Path exists at the intersection of these two worlds. Twenty years of CRE development experience isn't just informing the product - it's the product's foundation. Every entity relationship, every document hierarchy rule, every query pattern in Docuity reflects how projects actually work across the full lifecycle, not how a software team imagines they might.

The Bottom Line

The industry conversation about AI in commercial real estate is stuck on the wrong question. It's not about which model to use or whether AI can "replace" project managers, construction administrators, or acquisitions analysts. It's about whether your project's institutional knowledge - the dependency chains, the document hierarchies, the cross-references between contracts and compliance requirements and budgets - is accessible only to the people who happen to be in the room, or whether it's encoded in infrastructure that makes every team member, and every AI tool, smarter.

The companies that figure this out won't just be more efficient. They'll be structurally better at managing complex, multiyear projects - because the knowledge that drives good decisions will persist beyond any single person's tenure, any single phase of the lifecycle, or any single market cycle.

Where does your project's most valuable knowledge actually live - and what happens when the person carrying it moves on?