What I’m thinking about: How fast consumer expectations for AI tools are shifting (the bar is rising weekly), and how the cost (time and money) to keep pace keeps collapsing alongside it.
Where SLC Chat stands
Quick catch-up if you missed last month’s piece: we’ve been building our internal underwriting brain in chatbot form, trained on thousands of anonymized deal reviews, every piece of content we’ve put out, and the proprietary pricing math we’ve refined over years (that has led to millions in realized revenue and industry-leading operating margins).
The internal version was working. Voice match was where I needed it. Accuracy on detailed comping questions was sitting at the ~95% threshold I’d set as the no-ship-below line (our reputation runs through every answer this thing gives).
Time to push it live.
(Original plan was to ship as a custom GPT with a paywall through a third-party provider. Scrapped that plan because we wanted Anthropic’s models on the back end, and we didn’t want a third party taking a margin chunk we could keep in-house…driven by our realization that modern tools have made the DIY option viable for non-engineers in a way it wasn’t even a year ago.)
Over 3-4 days, with Claude Cowork handling the product planning and Replit (a vibe-coding deploy platform) handling the actual web app build…we had a working product, and I was the ONLY human in the loop. Email-based authentication, Stripe paywall, our SLC Chat custom instructions and knowledge base fully converted over to the Replit backend, conversation history, pricing tiers, landing page, and feedback mechanism.
Total out-of-pocket on this build: ~$160 in subscriptions and API tokens. Half a week of my time (future builds likely would take less time, now that the learning curves for Replit, Claude API, and Stripe webhooks are mostly behind us).
Now compare that to where I was on Land Pricer in early 2025 (a window that genuinely feels like ancient history now). Same category of build (our previous AI underwriting tool, now in the hands of a new operating team targeting appraisal firms), and the cost for the web app looked like:
- ~$25-30k out of pocket
- 4-5 different engineers in rotation, with a senior product manager
- Several weeks of iteration cycles where bugs continued to pile up as soon as we could squash them
Same kind of build (more complex in some ways, but not by orders of magnitude), roughly a 200x cost compression and ~10x speed compression on the same category of work…and the curve is still bending toward more cost and speed savings on a weekly basis. 2026 will be recognized as the start of an entirely new era from a technological progress standpoint.
(And that’s just from just year ago, I recall setting up payment processing on an e-commerce website 10 years, and it took about a week with a bunch of documentation and comm’s req’s. Noting how secure online financial processing should be, while I wouldn’t say setting up a Stripe webhook is dead simple…it IS remarkably easier to accept legal tender nowadays.)
=====
Once the limited beta was ready, I quickly got product feedback from Seth Williams (RETipster founder, one of the sharpest tech-product minds in the land investing space).
His core observation, paraphrased: SLC Chat as positioned today is a sharp framework that can be applied to deal data the user brings. But the bar has moved, even compared to a few months ago. With agentic AI tools like Cowork or Perplexity Computer, users of those programs have started expecting AI to do the work (e.g. finding the comps), not give them homework…and that’s the floor expectation now.
Seth’s right. The way the web app sits today, a user would still toggle between SLC Chat and their existing comping tools (e.g. Redfin, Zillow, Land Portal, Land Insights). Two windows, more friction.
(As a side note, Seth also uncovered a few other UI bugs, like the convo history not loading properly. A year ago, that would have required back and forth with an engineer, GitHub pull requests, and QA testing…now it’s fixed in 10 min after informing Replit about the issue, and the Replit AI agent QA’s itself. Just stunning progress.)
As AI-forward as I am, my first instinct was to pivot SLC Chat: scrap the Replit web app, rebuild as a Cowork plugin that inherits agentic capability natively (enabling targeted comp search), and change the offer to a white-glove install with a recurring sub for product updates.
Talked this through with Cowork (to combat sycophantic tendencies, I program my LLMs to challenge my thought processes relentlessly, enabling my own product strategy convictions to be pressure-tested).
After logical pushback from Cowork, we landed on a hybrid approach instead:
Track 1, the SLC Chat web app, for breadth. Self-serve, lower price tier, addresses the operators who want SLC’s institutional-grade underwriting framework applied to data they bring. File upload (PDFs, screenshots, DD spreadsheets, or comp reports from Land Portal) is the highest-leverage near-term addition (will be ready by today). My subconscious was already heading in this speed/ease-of-use direction even before receiving Seth’s feedback. I woke up in the middle of the night realizing we needed to add file upload capabilities ASAP.
Track 2, the Cowork plugin, for depth. The agentic version Seth was pointing at. Would actually pull comps, natively connect to external APIs, running the full underwriting chain on the user’s behalf. Fortunately everything we’ve built so far will seamlessly translate into developing this track too.
Why a full pivot would have been wrong: Seth and I represent tech early adopters, as far along the AI curve as anyone. The web app version catches operators who are hesitant to (or may never) install Cowork, which requires a higher tier Claude plan, and has its own learning curve to deal with.
The plugin track catches operators already living inside Cowork who want the agentic experience, and also serves as an upgrade for those who want a higher level experience after testing Track 1. Both populations are real, and serving them well requires different products.
In sum, we’re shipping the version we have, listening HARD to the feedback the moment it lands (e.g. ‘Approach vs. Avoid’ psychology a la Dr. Ben Hardy), and routing the next iteration toward what the bar actually requires (…not what we assumed it would).
=====
The recurring takeaway from the Land-Pricer-to-SLC-Chat compression: the only durable AI moat right now is the underlying DATA and METHODOLOGY behind the build.
The build itself is increasingly commoditized (don’t take my word for it…look at how many indistinguishable wrapper apps are launching weekly). The nuanced judgment the AI applies, the actual reps logged inside your business (which enables you to catch when the AI build is incorrect, often subtle errors a non-expert would never notice), the documentation of how your team thinks about problems…that’s the part that can’t be cloned for $60 over a long weekend.
If I’d tried to build SLC Chat from scratch with no data moat or underlying expertise, the build would be technically possible and substantively useless (which, frankly, is a fair description of a lot of AI products on the market right now…caveat emptor).
And without the many years worth of reps, off-the-shelf LLMs won’t save you either. I’m seeing a growing number of funding deal submissions that are glorified ChatGPT summaries, down to the formatting. Utter garbage, laughable in how wildly incorrect they are with pricing assumptions. If there’s one thing to remember, AI is only as good as the context you give it.
See SLC Chat Live
I’m hosting a free 60-minute live masterclass tomorrow, Tuesday, May 19 at 3pm Central, covering:
- Live demos on actual deal-comping (bring your deals, I’ll run them through the bot in real time) with side-by-side comparisons: SLC Chat vs. existing AI solutions on the same underwriting questions (this is the ultimate quality test, and I won’t offer pre-set land deals to review. Either watch SLC Chat outperform the status quo…or fall on its face, we’ll find out live.)
- A preview of what the Cowork plugin version would do that the web app can’t (the agentic build you’re hearing about here for the first time)
- A walkthrough of SLC Chat’s underlying knowledge base architecture, the voice calibration process, and what it took to get answer accuracy to our institutional standards
- Open Q&A
- An exclusive BONUS for the Cowork plugin build, for anyone who attends live
Sign up for the masterclass here. Even if you can’t make the live session, anyone who registers will receive the replay within 48 hours.
If you’ve been on the fence about whether AI is actually useful for land underwriting (or you’ve been burned by a tool that gave you confident-sounding garbage on a real deal)…this is the no-cost way to see what a methodology-trained build looks like compared to what’s currently available.
P.S. Quick reminder that we’re still actively funding land deals, and the methodology behind SLC Chat is one piece of the wider underwriting infrastructure we bring to every review. We write checks from $50K+, close 100% of deals we commit to, and bring national land underwriting experience built across thousands of deals. Submit your deal on our website or hit reply with the details.
P.P.S. Two big updates I’ll write about more over the coming weeks once everything’s locked in: we’re closing in on the largest single disposition in SLC’s history (massive sigh of relief in the midst of this tough market nationally), and we’re days from pulling the trigger on the biggest entitlement project we’ve taken on to date, a deal that’s been months in the making. The operators making it through this cycle are the ones with patience and discipline at the entry and relentless willpower to achieve favorable exits, and we continue to invest hard in both.
