What I’m thinking about: How Claude Cowork has collapsed the distance between “I have an idea” and “it’s shipping”…and what that’s letting us build inside SLC right now.
Quick behind-the-scenes on something the team and I have been heads down on: SLC Chat (SLC = Serious Land Capital).
We are compiling every podcast, every newsletter, every internal underwriting doc, and the anonymized notes from thousands of deals we’ve reviewed as funders, and feeding all of it into a custom GPT that answers land investing questions in my voice.
Internal tool first (so our team has every learning, procedure, and underwriting nuance we’ve developed at their fingertips). Soon to be a low-ticket paid product for the broader land investing community.
=====
If you’ve been reading Serious News for a while, you know I spent well over a year (and a lot of personal capital) on Land Pricer…the AI underwriting software that would live-pull comps, read aerials, and spit out conservative offer prices (rest assured, this sounds simpler than the actual mechanics at play).
FYI Land Pricer is now in the hands of a separate operating team that’s continuing to push it forward as an enterprise tool for appraisal firms (design partners engaged, daily progress).
The technology when we started simply wasn’t where it needed to be, and I’m honestly still a little shocked at how much I paid for product dev techniques that are now inferior to what Cowork can do out of the box, at a far faster pace.
SLC Chat is a piece of that original vision: translating my actual underwriting voice and nuance into something my team (and eventually other land investors) can interact with directly.
Smaller scope, lives inside SLC (instead of a separate company), dev pathway is straightforward.
To note, just a few months ago we seriously considered pulling the trigger on a third-party AI service that was pitching basically a watered-down version of what SLC Chat is becoming.
Discounted price was $6K per year. Inferior product, WAY longer iteration turnaround, and the voice match was nowhere close to what we needed. While that product was “cutting edge” at the time, it seemed like it would be more hassle than it was worth, given the difficult tuning process.
A few months later (representative of how FAST progress is with AI), Cowork lets us build a better version ourselves, on a faster timeline, for pennies on the dollar.
=====
None of this would have come together as fast as it has without Callan Faulkner and the Uncommon Business team.
Their CalChat+ custom GPT is the model for SLC Chat. Callan’s team spent the better part of a year stress-testing knowledge base formats, file structures, custom instruction patterns, and voice calibration protocols to get CalChat+ working at the level her internal team and mastermind community expects (and, most importantly, actually uses).
That year of refinement is what we’re standing on. We didn’t have to test 15 different file formats to find the one the LLM reads cleanly. We didn’t have to invent a voice grading rubric from scratch. Callan reverse-engineered her own CalChat+ process into a generalized approach that ANY business with the underlying data can use to build their own version (which is exactly what we’re doing).
Our learning curve dropped by an order of magnitude because of her team’s ground work (if it isn’t abundantly clear, we wouldn’t be nearly as far ahead in the AI game if we hadn’t worked with her).
=====
I’ve said this before, but it keeps hitting harder every week I use Cowork…
The single biggest gift of this current AI wave is that it lowers the inertia on starting the thing you’ve been avoiding.
I’ve never met an entrepreneur who doesn’t occasionally procrastinate on the hairiest, highest-leverage problem on their list (myself absolutely included).
Cowork collapses the ‘action potential’ needed to get started:
You describe what you want (or even if you DON’T know what you want, just start a stream of consciousness on your initial thoughts), Cowork helps clarify the direction, and then it produces an initial pass. You review the output like a manager instead of grinding through it as a solo contributor…and the ball just starts rolling faster than you ever thought possible.
That’s how we’ve been able to cut overhead and dramatically raise productivity at the same time over the past several weeks, a critical combo in the midst of maximum macro uncertainty.
=====
After significant back-and-forth aligning the product direction, Cowork helped break the SLC Chat build into four phases inside our Notion task manager (Cowork is connected directly to Notion, so it scheduled subtasks, assigned realistic delivery dates, and generated a full voice-calibration grading rubric without me touching any of it).
Phase 1: Knowledge processing. Boiling down collectively thousands of podcast transcripts, newsletters, emails, CRM notes, and internal docs into a correctly formatted knowledge base an LLM can actually read cleanly. Doing this manually would’ve taken HUNDREDS of hours…Cowork completed this in less than half a day. (Cowork even caught a file-type inefficiency in the format we inherited from CalChat+ that was leaving roughly 15% of bandwidth on the table. Already fixed.)
Phase 2: Custom instructions and the GPT build itself, plus running a voice calibration protocol against 15 stress-test questions with a grading rubric. Deep in this phase right now.
Phase 3: Internal team testing and a small group of beta testers from this audience.
Phase 4: External deployment. We already have the paid wrapper provider lined up (though it’s likely we will convert to building our own paid wrapper through Replit, a popular ‘vibe-coding’ platform, so we can control the user experience, and default to Anthropic’s Opus 4.6 model…largely considered the best publicly available AI model.)
We blew through Phase 1 and we are almost two weeks ahead of schedule. Cowork handled the manual drudgery (e.g. sorting through 300+ podcast episodes and surfacing the 48 most relevant ones for the first version) without me selecting a single thing by hand.
=====
Here’s the piece I want to really drive home…
Cowork is only this powerful for us because we have thousands of detailed deal breakdowns within our CRM, we’ve spent over a year producing value-add content, we have detailed internal voice, business ops, and industry-leading underwriting documentation…and we spent weeks organizing our underlying data before we ever tried to build SLC Chat.
If you tried to build something like SLC Chat (or CalChat+) from scratch with no moat of existing data, the build itself would take ages…and likely result in an inferior user experience.
This is the quiet lesson of this whole project: the data moat you’ve been building without realizing it is the most valuable asset you own right now (one of the few moats that’ll be worth ANYTHING over the coming years…or maybe months) That’s the fuel for everything AI lets you build next.
Our company North Star, “Underwriting over everything,” still applies here. We are not going live until SLC Chat is answering correctly ~85-90% of the time minimum. Our reputation is built on nuance and accuracy. If SLC Chat can’t protect that, we don’t ship it. Period.
(Even within initial testing, less than a week into this project, I’ve already gotten responses from SLC Chat that required NO changes at all…there’s certainly promise here).
=====
Zoom out for a second on what this actually means for anyone sending us deals. Every internal AI tool we build (SLC Chat is one of several in the queue) compounds the underwriting edge we already bring to the table. Faster pattern recognition across thousands of historical deals. Tighter feedback loops on dispo strategies. More of my actual reasoning available to my team in real time, on every deal we evaluate.
If you’re an experienced operator looking for a capital partner whose underwriting moat is widening, not shrinking, we should be talking. We write checks from $50K+. We close 100% of deals we commit to. And we bring national underwriting experience (now amplified by the tools we’re building internally) to every transaction.
