The room where Austin businesses go Agentic.

He turned 15 years of messages into a clearer way to run his business

David Gonzalez · built with Christopher 'CT' Schenk

1.4 million texts read · 7,368 contacts sorted · ran in 6 minutes

The problemDavid had never had a CRM. He hated that CRMs made him work for them — logging calls, entering contacts after every meeting. He wanted something that worked for him.

CT agreed to help and spent three hours building a private system while literally packing for a flight to China. The system reads 15 years of iMessages — 1.4 million texts — and infers the strength, direction, and context of every relationship without David having to feed it manually. Six minutes after connecting it to his iCloud, it told him his wife texts him 25% more than he reciprocates, that he had 37 unanswered messages from direct reports in the last two months, and that the person he has texted most outside family is the founder of Baby Bathwater. He calls it Triple IE: Intelligence, Inference, Insights. The deeper discovery came when he asked it what would surprise him about himself. It said he had been miscast for 17 years as a connector. He is actually an architect. His connector identity had been taking credit for the work his architect self was doing all along. CT's method: spend three days planning and brainstorming before writing a single line of code. Build a PRD. Know exactly what you want it to do. The system is built locally in Cursor using Claude Opus, runs entirely on his laptop, and never touches the cloud. The whole thing looks like a simple CLI, but it monitors his iMessage database in real time.

Takeaway The questions you ask up front matter more than the tool you build. CT's rule: spend three full days on planning and EBI loops before writing any code. Direction first, build second.

He grew a home-services brand 15x for the cost of a phone bill

Oscar Hidalgo

800 → 15,000 visitors/week · ~$900/mo all-in vs $10K/mo agency

The problemAn HVAC company doing $3–4M a year in commercial work decided to launch a residential brand. No marketing agency. No budget. Just Oscar, managing everything himself.

The first move was rebuilding the website from scratch in Claude Code — no Framer, no website builder — and deploying it on GitHub and Vercel so his AI agents could read and modify it directly from the command line. He then used the FireCrawl MCP to scrape every competitor in his target zip codes and identify content gaps: topics they were not ranking for, questions they were not answering. That research became a 200-topic content plan built around the seasonal rhythms of HVAC work — air conditioning in summer, heat pumps in winter, efficiency rebates year-round. From there he built an OpenClaw automation agent with three layers of configuration: a humanizer skill so it always sounds like the brand, full access to the competitor research so it stays on-topic, and hard rules like no dashes in copy. The agent fires every 30 minutes, pulls the next unwritten topic from the list, uses the Perplexity MCP to check current accuracy, and drafts a blog post in GitHub for Oscar to approve before it publishes live. Weekly, the Ahrefs MCP runs a full technical audit and tracks keyword rankings by zip code. One tip that came up in discussion: language models tend to recommend businesses that playfully call out what their industry gets wrong. Transparency and a little irreverence toward competitors builds trust with AI overviews. Total monthly cost is $800–1,000, versus $10,000 a month for an agency — and he is already getting calls from people who found him on Google.

Takeaway An agent can run your entire content operation for a fraction of the cost of an agency — as long as a human still approves every piece before it goes live. Build the machine, stay in the loop.

He built one dashboard that tells him the next right move

Justin Day

150+ data sources · one action hub · built to sell with the business

The problemJustin ran a 30–40 person SEO agency and realized he had been building half-finished tools for months. Nothing talked to each other. Revenue stayed flat. He was working in the business, not on it.

He started by pulling every data source he had into a single system: Stripe revenue, GoHighLevel CRM, Instagram, Facebook Ads, email, Slack, and even his Apple Watch wearables. He is now connecting his Meta Glasses so he can speak to the business in real time without touching a keyboard. From there he deconstructed Alex Hormozi's three books — Offers, Leads, and Sales — inside Claude and built processing engines modeled on Hormozi's frameworks. Each data source feeds into the processing engine that matches its stage of the business. Those then feed into growth hubs that interpret what is actually working at every level of the funnel, which feed into an action hub that outputs one thing: the three highest-leverage moves to make today, based on the real numbers. The system is entirely local, built through Claude Code in the terminal. Everything is visualized as an HTML dashboard so he is not reading walls of text — he even has an Obsidian-style three-dimensional graph of all his data connections. Every new module is required to prove itself by showing exactly where its data comes from, what processing it runs, and what it would look like if it improved. He is not letting his team access it yet because of how much business data it holds. The long-term play: this system adds multiple multiples to his business valuation when he eventually exits.

Takeaway Do not build more tools. Build one brain that knows your business deeply, tells you what to do next, and gets smarter with every data source you add to it.