CarbonAI's $12M: AI agents that work inside a lender's existing model

CarbonAI is not trying to become a lender. The New York company, co-founded by CEO Jeff Saul and COO Josh Kaufman, is building agents that prepare the analytical work around a commercial real estate loan while the lender keeps its own models, reporting conventions and authority over decisions.

The company announced $12M in funding on October 15, 2026, led by Cowboy Ventures, with Timber Grove Ventures, Bling Capital and Max Ventures participating. The announcement identifies the amount and the investors without assigning the financing a named Series or publishing a valuation.

Saul and Kaufman are returning to commercial real estate software after selling Nativ to Built. Their team page describes a path through Starwood, Morgan Stanley, Argentic and Guggenheim before Nativ and Built — a history that matters because the product asks customers to hand over work shaped by their own lending standards. Cowboy Ventures partner Caroline Duffy pointed to the founders' industry experience and existing lender relationships as reasons for backing the company, an investor's account of its conviction rather than an independent product benchmark.

The implementation begins with the customer's documents and templates. Rent rolls, leases, operating statements and loan documents supply the inputs; the software extracts and reconciles the information, maps the lender's existing models and prepares an underwriting workbook with an accompanying investment memo. Work requiring judgment is routed to the lending team. Keeping the workbook matters because a model is also a record of institutional choices — the questions a lender asks, the assumptions it accepts, how a reviewer expects to inspect a deal.

CarbonAI also describes recorded agent activity, source-linked figures and rule-based reconciliation, giving a credit professional a way to trace where an input came from, what the software did with it and whether the arithmetic agrees. The company states its platform is SOC 2 Type II compliant, hosted in the U.S. and designed to keep customer data out of model training — company representations an institution would need to examine itself before deployment.

The market context is a calendar of maturities. The Mortgage Bankers Association's February 2026 survey identified $875B of commercial mortgages scheduled to mature during 2026, or 17% of outstanding balances in the surveyed market, down 9% from the amount scheduled for 2025. CarbonAI's scope extends past origination into quarterly reporting and asset surveillance, so the same institutional standards can matter repeatedly across the life of a loan rather than in a single document-processing task.

CarbonAI reports that early customers saved as much as 90% of the time on selected underwriting tasks, and that contractual revenue reached seven figures within six months of its early-2025 product launch. Both claims carry scope limits: a task-level saving does not transfer automatically to an entire loan process, and contractual revenue is not recognized sales or annual recurring revenue. Greystone appears as a named customer through a testimonial on CarbonAI's website, emphasizing the team's willingness to understand due diligence collection and build around the lender's actual process. The company has not disclosed a detailed allocation of the funding.