Happy Friday the 13th CRE Agents Founders Members! Sending along good vibes and an update on another busy week. I want to start with something that happened Monday morning that I think captures where we are right now. I sat down to write a single skill, Yield Maintenance Calculation, figuring it'd take an hour. Four days later, we've crafted nearly 20 skills covering things like Yield Maintenance Calculation, Leasing Cost Reserve, Loan Sizing, Stabilized Pro Formas, Chart of Accounts templates for the five major property types plus a general one, etc. Each skill started as "while I'm in here, I should also build..." and I couldn't stop. That's the feeling when you've hit your stride. Why does this matter to you? Every skill we write expands the universe of tasks an AI coworker can handle. Today, skills power the atomic tasks you're running on the platform. Down the road, they're what make truly autonomous work possible: an AI coworker that doesn't just do what prompted, but knows how to think about yield maintenance or lease cost reserves the way we do. We're not there yet, but every skill closes the gap. On the task side, we hit our goal of shipping one new task per day this week. Five new tasks went live, such as Map Income Statement to Standard Chart of Accounts (v2), Compare T12 vs. Broker Pro Forma, and a Development Multifamily Rent Study. If you haven't tried the T12 comparison task yet, it's one that should keep brokers honest. You can review all current skills, tasks, tools, AI coworkers, and datasets at creagents.com/capabilities . The most consequential add this week, though, was our CRE Agents Employment Data and Insights dataset. This one I'm genuinely proud of. We built a county-level employment intelligence API built for CRE: 1.3 million monthly employment records going back to 1990, 10 million quarterly industry records across every NAICS sector, covering all 3,225 U.S. counties. But the raw data isn't the point. What makes it useful is what we built on top of it: a Property Type Demand Index that scores every county for office, industrial, retail, multifamily, and hospitality demand by tracking the specific industries that drive each property type. When Professional Services and Finance firms are expanding in a county, the office demand score rises. When warehousing and logistics employment surges, the industrial score reflects it. There's also an Employment Momentum Score (is growth accelerating or decelerating?) and a Market Resilience Index (how did jobs hold up during the Great Recession and COVID?). Every metric comes with a national percentile rank, so cross-market comparison is instant. We've already wired it into the Deep Location Analysis task, and there's a standalone task for it too. This is dataset number 10, and I expect to ship another one next week. If you have any ideas for tasks where you'd like to use it, just shoot me a note. Here's an important point about this technology I think worth mentioning: an "AI coworker" is only as good as the data it has access to (as well as the tools it can use, the skills/knowledge it is given, and the instructions it follows). Prompt engineering matters, skills matter, but without proprietary, real estate-specific data, the output hits a ceiling. That's why we're investing so heavily here. Ten datasets and counting. The more we build, the more powerful every task on the platform becomes, for all of you. I should also mention: we brought in this week's new Founders Members, and the sign-up process broke for about 10 minutes. All seats were claimed within the first five minutes of sending out invite codes (which is incredible and a little terrifying), so the timing was rough. On top of that, our chief engineer has been delayed four days trying to get home due to the conflict in the Middle East, which made the scramble more intense than it needed to be. We got it fixed, but to our newest Founders: I'm sorry for the bumpy welcome. You're in good hands now. Looking ahead, next week we're targeting five more tasks, another new dataset, and 10 additional skills. We're also bringing on a new task builder next week, with a third joining us April 1. That means we're scaling our capacity to train AI on the work you do. So please keep telling us which tasks you want built. Every request shapes the roadmap. Bigger AI Picture: Voice AI took a leap forward this week. Hume AI open-sourced TADA, a new approach to text-to-speech that generates voice 5x faster than comparable systems, with zero hallucinations in testing, and it's lightweight enough to run on a phone. Why should you care? Because the path from "type a prompt and read a report" to "talk to your AI coworker on your commute and get a briefing read back to you" just got shorter. We're not integrating voice today, but I'm watching this space closely. ( hume.ai/blog/opensource-tada ) On a personal note: I'm having more fun than I've had in years. Teaching AI to do the tedious work that used to eat