Why Commercial Real Estate Is Ditching Packaged SaaS for Custom AI Agents

NEW YORK, NY – September 22, 2026 – For decades, the commercial real estate industry has been trapped in a digital paradox. While the skylines they build are bespoke and highly customized, the software they use to manage the billions of dollars flowing through them has been anything but. Asset managers and operations teams have long been forced to cram complex, nuanced financial realities into the rigid, unyielding templates of packaged enterprise software.

But a quiet revolution is taking hold in the back offices of major property operators. The era of waiting on a vendor's multi-year product roadmap is ending. Instead of buying off-the-shelf software to manage their reporting, commercial real estate firms are now building their own autonomous AI agents.

The vanguard of this shift was illuminated today as Friedman Real Estate, a powerhouse managing over 20 million square feet of commercial property and 19,000 multifamily units nationwide, announced a sweeping deployment of enterprise AI infrastructure provided by New York-based Leni. Rather than purchasing another static reporting tool, Friedman utilized Leni's underlying architecture to assemble bespoke, automated workflows that plug directly into their existing enterprise resource planning (ERP) systems.

This partnership represents far more than a routine software upgrade. It signals a fundamental paradigm shift in enterprise technology: the value is moving away from the application user interface and down to the contextual data and verification layer.

The Custom Reporting Bottleneck

To understand why this shift is necessary, one must look at the operational nightmare that is institutional commercial real estate reporting. As a firm like Friedman scales across 29 states, its operations teams absorb an exponentially growing volume of custom reporting requests.

Different lenders, joint venture partners, and institutional investors all demand bespoke reporting packages. They each possess proprietary definitions for critical metrics like Net Operating Income (NOI), Common Area Maintenance (CAM) reconciliations, and economic versus physical occupancy.

Historically, legacy ERP walled gardens—platforms like Yardi, RealPage, and MRI Software—have operated as closed ecosystems. Extracting data from these systems to satisfy a specific lender's unique format usually required an army of financial analysts manually exporting CSV files, reformatting spreadsheets, and cross-checking formulas. When an off-the-shelf Software-as-a-Service (SaaS) product encountered a non-standard metric calculation, it simply failed, forcing highly paid professionals back into Microsoft Excel.

Friedman realized that a packaged reporting tool is fixed on the day it ships, but the way a firm reports is constantly evolving.

"We did not want another reporting product that we would immediately need to change. We wanted to build our own workflows, and Leni's infrastructure is what made that possible," noted Jared Friedman, Co-CEO of Friedman Real Estate. "Our team defined the metrics and the review steps, and the agents now produce the reports and dashboards on their own, in our format, traceable back to the source. When a lender or an owner asks for something new, we build it ourselves instead of filing a request and waiting."

From Geoffrey Hinton's Lab to $80 Billion Portfolios

The technological engine powering this autonomy is Leni, an AI infrastructure company that supports over $80 billion in North American commercial real estate assets. Spun out of the Vector Institute Labs—co-founded by AI pioneer and Nobel laureate Geoffrey Hinton—Leni brings a distinctly different architectural approach to enterprise AI.

In heavily regulated, numbers-critical industries like commercial real estate, standard generative AI hallucination rates of 2 to 15 percent are catastrophic. A hallucinated debt covenant or a fabricated ledger variance can derail a multi-million-dollar transaction.

Instead of relying on compute-heavy, generalist frontier models (like the latest iterations from OpenAI or Anthropic) to process entire financial workbooks, Leni orchestrates ensembles of task-specific Small Language Models (SLMs). These models are governed by deterministic verification loops. Every generated calculation, cell link, or extracted metric is programmatically verified against the source record in the ERP database before it ever reaches a human reviewer.

This architecture has yielded staggering results. In independent AI evaluations, including the rigorous GAIA (General AI Assistants) benchmark developed by Meta and Hugging Face, Leni took the number one ranking across all difficulty levels for complex, multi-step workflows. On the SpreadsheetBench evaluation for complex corporate financial modeling, Leni achieved a 91.25 percent accuracy rate, placing it in the top tier globally.

In production, Leni reports a task accuracy above 99.6 percent while operating at roughly one-third the compute cost of frontier models. For Friedman, this means reports are generated with mathematical certainty, and every output carries a transparent, bidirectional audit trail tracing back to the original ledger entry.

The Context Layer as the Ultimate Moat

The true disruption here lies in who owns the "context." Leni's patent-pending Universal Data Model normalizes transactional property data across 85 percent of the commercial real estate software market without requiring Friedman to alter its underlying ERP configurations.

But it is Leni's organizational memory system—a routed-knowledge graph—that changes the game. It captures the firm's internal definitions, calculation logic, and approval chains as people work. Rather than prompt-engineering public models where proprietary institutional knowledge could leak, this context layer remains the intellectual property of Friedman Real Estate.

"We provide the right infrastructure so that firms can experience the true promise of AI," explained Arunabh Dastidar, Co-founder and Chief Executive Officer of Leni. "Friedman's data was locked in closed systems, and the knowledge of how Friedman reads that data lived with its people. We connect the systems and capture that context, then give teams specialized models and verification they can assemble into their own agents. Friedman built reporting agents first because reporting hurt the most. The point is that the context layer belongs to Friedman, and every workflow it builds next runs faster and cheaper on it."

The operational impact is already palpable. One managing director at the real estate firm noted that the platform's automated pipelines and clear data visualizations have effectively cut company meetings in half, liberating asset managers to focus on strategic capital deployment rather than arguing over conflicting spreadsheet numbers.

Redefining Enterprise Agility

Reporting is merely the beachhead. With the data connected and the context securely in place, Friedman is already looking to expand this agentic infrastructure into annual budgeting, rolling variance analysis, investment committee deal memos, and acquisition due diligence.

The broader takeaway for the business world is clear: the era of conforming company operations to fit the limitations of vendor software is drawing to a close. Firms no longer need to construct massive, multi-million-dollar data warehouses in-house just to build with AI. By leveraging specialized infrastructure that supplies the data architecture and the guardrails, companies can supply their own conventions and build directly on top.

In the high-stakes world of commercial real estate, agility and accuracy are the ultimate currencies. By transforming their proprietary operational knowledge into repeatable, autonomous workflows, firms are not just streamlining their back offices—they are turning their institutional memory into an unassailable competitive advantage.