We quietly shipped the largest dataset on the platform this week. And it's one I've wanted to build for a long time. Behind the Glass The LIHTC Data Hub is live. It indexes 3.7 million affordable housing units across 54,000+ projects going back to 1987, pulling from HUD, state housing finance agencies, and nonprofit databases into a single, address-driven lookup. Rent limits, income limits, subsidy layers, preservation risk, QCT and DDA designations, inspection trends. All of it normalized and de-duplicated at the property level. Why this matters: if you touch affordable housing, whether as a developer, lender, syndicator, or acquisitions shop, you know how fragmented this data is. Piecing together a single property's story means bouncing between three or four federal sources and a state HFA portal that looks like it was built in 2004. The Data Hub collapses all of that into one call. It powers two new tasks on the platform: the LIHTC Property Intelligence Report (deep underwriting-grade output on a specific property) and the LIHTC Property or Feasibility Report (market-level feasibility for a given site or submarket). Beyond LIHTC, we shipped seven more tasks this week. The Stabilized Pro Forma task builds a hold-period cash flow from your inputs, and the T12 Mapping to Chart of Accounts task takes a raw trailing-twelve and normalizes it to a standardized chart of accounts by property type. Both of those came directly from Founder requests. We also added a Demographics Report, an Employment Report, Retail Spending Data and Insights, a Location Research Report, and a task that sets up the A.CRE Apartment Development Model from your deal documents. On the skills side, we wrote eight new ones this week. A few worth calling out: the Excel Model Build Guide (a reference our AI coworkers use when working inside Excel-based financial models), a set of four financial modeling skills covering return metrics, debt and operating risk metrics, unlevered cash flows, and levered cash flows, and a Retail Spending Data Methodology skill that documents how we built the composite metrics behind our retail dataset. All of this continues to be shaped by Founder conversations. If there's a task you want or a workflow that's taking too long, reply to this letter. We're building for you first. Bigger Picture Andrej Karpathy (OpenAI co-founder) published a job exposure map this week that went viral. He scored 342 U.S. occupations for AI exposure and the pattern is stark: jobs paying over $100K averaged 6.7 out of 10, versus 3.4 for jobs under $35K. His logic is simple: if the work product is digital and can be done from a laptop, exposure is high. ( karpathy.ai/jobs ) Here's what caught my eye as a CRE person: nobody has overlaid this data with submarkets yet. Office markets where the dominant tenants are in high-exposure occupations face a different demand story than markets anchored by low-exposure industries. Someone should build that analysis. (Maybe we will.) Talk soon. Spencer