# AI Meet Up recap: November 4, 2026, Provo, Utah

Matt Sumsion hosted the AI Meet Up at the Sumsion Real Estate office in Provo on the evening of Wednesday, November 4, 2026. Seven presenters showed real tools they use, then answered questions. It ran about 90 minutes, with a full room in person and more people online.

## How to use this file

Paste this whole file into your AI (ChatGPT, Claude, Gemini, or whatever you use) along with the prompt below.

> Below are detailed notes from a local AI meetup where seven people showed how they actually use AI in their work and life. First, ask me 5 short questions about my work, my biggest time drains, the tools I already use, and how technical I am. Then, using these notes, give me:
> 1. The 3 ideas from this meetup that would save me the most time or make me the most money, ranked, with the reason for each.
> 2. For the top idea, a step-by-step plan I can start this week, in plain English, with the exact tools and what each one costs.
> 3. One thing I should NOT copy yet, and why.
> Keep it practical. Assume I am busy and not a programmer unless I tell you otherwise.

## The big ideas from the night

1. **It's not the AI, it's what you give it.** Jared Fenn pointed out that everyone in the room had the same models (ChatGPT, Claude, Gemini, Grok). The people getting 10x results fed their AI a "brain": their calls, texts, emails, meeting recordings and notes, all in one place it could search.
2. **Hire an employee, don't order a gig.** Jared said most people use AI like Fiverr: hand over one task, get one result, start over. He treated his AI like full-time staff he trained once, that remembered everything and got better every week. His line: you train an AI like remote staff; you can't show it, you have to write it down.
3. **You don't need to know how to code.** Aaron Earnest, an English major, built a city permit data business by talking to AI in VS Code. Jeremy Dye, a structural engineer by training, built his firm's whole project management system. As someone in the room put it, "you just need to know how it should function."
4. **Build small personal apps for your own problems.** Perry Sessions and Alex Sisk showed a gym log, a family budget board, and a recipe book that ordered groceries. Free hosting (Vercel, Cloudflare) plus a free database (Supabase) made them cost nothing.
5. **Skills and memory files beat long chats.** Chad Mckinney and Jared saved what the AI learned into small files (memory) and turned repeat jobs into written instructions (skills). They started fresh sessions often. It was faster, more accurate, and used fewer tokens.
6. **Give the AI a structure to fill in.** Epic Engineering found that when 20 to 30 project managers each used their own AI, every proposal looked different. A shared template (letterhead, fixed sections) made every output consistent.
7. **Have one AI manage the others, and have another check the work.** Jared and Alex both used a "manager" or "chief of staff" agent that handed work to specialists, with a separate reviewer that checked the result before it reached them.
8. **Start with the boring, expensive thing.** Jeff Davis turned state licensing textbooks into online courses and saved his company about $100,000 in one year. Exam pass rates went from about 50% to about 90%.

## The presenters

### 1. Aaron Earnest (joined online): Agenda Insider
- **The problem he described:** builders, brokers and title companies want to know what's getting built before the permit is pulled. That information was buried in thousands of pages of city council and planning commission PDFs, and paying people to read them was never practical.
- **What he showed:** agendainsider.com. AI reads every agenda and supporting PDF from city websites (CivicClerk, Granicus, PrimeGov, Legistar and custom sites), pulls out projects, applicants, parcels, unit counts and zoning requests with citations back to the source, and sends alerts. He said it was live in 24 regions. On screen: 728 recent projects in Utah County and 15,718 units and lots.
- **How he built it:** he said he is an English major with no coding background. He spent about a year, around 50 hours a week, "managing a swarm of AIs" in VS Code connected to the latest models. He started in November 2025, and two bigger competitors have since entered the space.
- **Who uses it:** he uses it as an investor, and a title company friend uses it to contact every ADU and one-off project. His example: a friend passed on land by the new Springville I-15 off-ramp years ago, and it is now worth about 20x. Knowing the off-ramp was coming would have changed that.
- **His takeaway:** "Any cool idea that you have, sit down and start talking to your AI model." He offered to walk anyone through getting set up.

### 2. Perry Sessions: personal apps, Codex vs. Claude, and Chrome extensions
- Perry introduced himself as co-founder of a small online advertising agency and a self-taught coder.
- **Gym Buddy:** he imported a year of gym notes from his iPhone Notes app (no other app could import them). The app charted his lifts, body weight and nutrition, and had a built-in AI chat that knew his goals (using Google's free Gemini API).
- **Money Board:** it imported bank statements, categorized spending, and answered questions about his finances. He noted ChatGPT can now connect to bank accounts directly (through Plaid) and do much of this.
- **The free stack he recommended:** Vercel or Cloudflare for hosting, Supabase for the database and logins, GitHub to store the code, and the Gemini API for a free AI brain inside your app.
- **Websites:** he builds and maintains client sites as plain HTML. To change anything ("add a sale banner," "swap the logo") he told the AI, which edited the code in GitHub.
- **Codex vs. Claude:** he preferred OpenAI's Codex, mainly because ChatGPT's image generation was much better at editing product photos and labels accurately. He said Claude was great for code but didn't generate images.
- **Chrome extensions:** Instagram doesn't allow automation, so he had AI build a Chrome extension that scrolled posts, opened profiles, and filtered creators by follower count and bio to find brand ambassadors. He reviewed each one with Keep or Skip, and AI drafted messages he sent himself.
- **Dictation:** he said ChatGPT's built-in dictation handled jargon (ROAS, lifetime value) far better than the Mac's built-in speech-to-text.

### 3. Jared Fenn: give your AI a brain
- Jared has been a home inspector for 20 years; before that he managed Nu Skin's first web team. He started building agents in January. His slides were branded Evolve Customers.
- **His core idea:** "Same AI, different results. It's not the AI. It's what you give it. Give your AI a brain." If your AI depends only on what's in your head, it stops when you stop.
- **Atlas, his chief of staff agent:** "I didn't automate a job. I hired an employee." Atlas runs mostly on Claude, on a VPS (a rented cloud server, about $200 a year from Hostinger), so it works when his laptop is closed. He said he talks to it through Telegram and a custom Android app, often while driving between inspections.
- **What feeds the brain:** recorded meetings (a Plaud pendant he wears all day), calls and texts (Quo, a business phone system that transcribes everything), and email. Every night Atlas files everything by person, connects people to tasks, and catches what he promised and what others promised him.
- **His real example:** he had Atlas pull about 8,000 call transcripts from Quo, match them against who booked and who didn't, find the patterns, and rewrite his quote page, all while he was in other meetings.
- **"Now my staff hires staff":** he showed an org chart with teams of agents under Atlas (Engineering: Builder, Tester, Gatekeeper; Night Shift: Librarian, Professor, Linker; Specialists: health coach, business coach, research analysts; Family & Legacy; Business Brains), plus 32 skills he called job descriptions any agent can pick up (draft in my voice, meeting follow-ups, CRM logging, contract review, check facts before claiming, a council for big calls).
- **The brain, visualized:** a 3D map of about 18,000 notes and 27,000 links, color-coded by business, family, health, faith, money, self and legacy.
- **YouTube hub:** Atlas watches the creators he follows, transcribes new videos, and tells him each morning what's worth knowing.
- **Own your brain:** the memory and logic live on his server, not inside one AI company. If a provider cuts usage or bans the account, he plugs in another model. He said he pays $200 a month for Claude and often gets close to the limit.
- **His recommendation:** look up Garry Tan's open-source "second brain" work; every skill is like a job description.

### 4. Jeremy Dye: Epic Engineering's AI-run project hub
- Epic Engineering, a full-service engineering firm, set a goal a year ago to become 20% more efficient. Jeremy Dye, a structural engineer by training, built the system in about four to five months at roughly 30 hours a week. They said outside quotes had been 18 months.
- **What it did:** each person's dashboard of today's actions, task management, employee scorecards, AI-assisted timesheets, a rebuilt CRM with "who to call next" and points for business development, proposals, contracts, project delivery, quality control on drawings, meeting recording with assignments, and checks that the design matched what the client asked for.
- **Proposals went from 4 hours to 15 minutes.** It read emails and photos, compared the job to past projects, knew each person's billing rate and speed, filled in the proposal, then created the project, the PDF contract, and the folders automatically.
- **Field reports:** Jeremy walked a site and narrated a video. The hub pulled still photos and wrote a nine-page report on letterhead, with captions, from a one-paragraph prompt.
- **Consistency:** everyone's personal AI had produced proposals that looked different. A shared structure (letterhead; sections for assumptions, limitations, exclusions) made all output match.
- **Connected to everything:** it kept a copy of their finance system's data (Vantage Point) and added fields it lacked, like contingency. Through an MCP connection, anyone could tell their own AI agent "I just got off the phone with so-and-so, add this lead," and the hub updated. Employees submitted feature requests, and AI triaged and built many of them.
- It was hosted on a VPS, with Microsoft login so only employees could get in.
- **From the Q&A:** a builder asked how to get this without learning to code himself. The room's answer: the barrier is low if you know how it should work, or you can hire someone (even a sharp 20-year-old) to build what's in your head.

### 5. Alex Sisk: a manager-led team of Claude sessions
- Alex grew up with Matt and is in BYU's MBA program. He said he applies AI to his own life while doing AI consulting on the side.
- **"I don't write most of the code. I run the team that does."** His Claude Code setup worked like a small company: one Manager session started a Lead for each area, each Lead started Workers (one per project), and a fresh Reviewer checked the live result before anything counted as done. **His rule: every project keeps a written goal and a dated log, so a new session can pick up where the last one stopped.**
- **Why:** he said he tends to chase new ideas mid-build, and the manager structure kept things moving in one direction.
- **Claude mods:** he showed a Build Board (his next five tasks, every project with a progress bar, and an idea shelf) and a button bar. **His favorite was an "Enhance" button.** He dictated a messy idea, clicked Enhance, and it rewrote it into a clear prompt with what I want, constraints, "done when," and "ask me before starting if." He offered to share the code on his GitHub.
- **100 simulated readers:** for LinkedIn posts, a Lead spun up 100 sub-agents that each played a LinkedIn user and reacted to the draft. The post had to pass the reviewer before it reached him.
- **Personal apps:** like a family recipe book that built a grocery list and ordered it from Walmart.
- **His tips:** search "Claude mods" for ideas; turn off skills you aren't using if your chat keeps running out of room; you can open two Claude windows side by side. He said he wants a home server or Mac mini to run a local model for small jobs, because he expects AI prices to rise.

### 6. Chad Mckinney: skills, memory files, and scoping integrations
- Chad said he manages about 30 software engineers and, on the side, builds integrations between business software, often for clients referred by friends at BambooHR. With a master's in cybersecurity management, he said he is careful about letting AI run free.
- **Memory:** he constantly wrote memory files, nested by client and project, as plain text (markdown) files on his computer. He never stayed in one session long: save, start fresh. He said it used very few tokens.
- **Skills:** his "integration viability" skill read all public API documentation for two systems (his example was BambooHR and Paylocity), ranked each source by trust (official docs 10 out of 10, a forum post much lower), and produced a scorecard: authentication, data model, how often the API changes, sandbox, custom fields, bulk vs. webhooks, rate limits, coverage of each HR task, overall complexity, risks, and questions to ask the client before taking the job. The verdict on screen: viable with conditions, high complexity.
- **From meeting to scope:** he recorded client calls with Granola, connected it to Claude, and fed in the transcript with the skill. Then a statement-of-work handoff produced a polished SOW.
- **Keeping secrets out of the AI:** API keys lived in local config files, so the AI could query live data without ever seeing the keys.
- **AI to build, not to run:** the finished integrations ran without AI at all. They were self-hosted and encrypted, with logs and alerts, so clients had no ongoing AI cost or risk.

### 7. Jeff Davis: turning 400-page textbooks into courses
- Jeff handles compliance and licensing for Alta Pest Control, which operates in 13 states. He explained that every state has its own license requirements and training.
- **His process:** gather the training material (state PDFs, or physical books he unbound and had scanned), upload it to Claude, prompt it to build interactive modules, export a **SCORM** file (a standard course package that drops into almost any learning management system), import it into Paylocity's learning tool, then test and tweak.
- **Results:** about $100,000 saved this year in third-party training fees (about $200 per employee, with about 400 employees in Texas alone). Exam pass rates went from about 50% to about 90%, because the courses cut out the categories they don't work in.
- The courses read the text aloud, required time on each screen, and ended with quizzes (70% to pass). He called them "very 90s looking," but they met state requirements.
- **What's next for him:** packaging the courses on a website and selling them to other companies at half the price of the third parties.

## Open floor moments
- **AI assistants that talked to each other:** one attendee described connecting GrokBot, Claude and other assistants to a shared knowledge base in Google so they could hand work to each other, and talking to GrokBot from his Tesla to update spreadsheets and calendars.
- **What gets more valuable:** information is getting cheap. Several people said relationships are what AI can't replace, so build them now.
- **Custom beats one-size-fits-all:** instead of paying for big software with 40 menu items you don't use, people said you can now build exactly what your business needs and change it on the fly.
- **How often to meet:** the room wanted more. Ideas included twice a month, plus hands-on workshops where people build together.

## Tools mentioned
| Tool | What it is | Mentioned by |
|---|---|---|
| ChatGPT / Codex | OpenAI's assistant and coding agent; strong image editing; good dictation | Perry |
| Claude / Claude Code | Anthropic's assistant and coding agent; skills, memory, sub-agents | Jared, Alex, Chad, Jeff, Jeremy |
| Gemini API | Google's AI, with a free tier to put AI inside your own app | Perry |
| GrokBot, Muse | Other AI assistants people in the room used | Jared, attendees |
| VS Code | Free code editor that connects to AI models | Aaron |
| Vercel, Cloudflare | Free website and app hosting | Perry |
| Supabase | Free database and logins | Perry |
| GitHub | Stores your code; AI edits it for you | Perry, Alex |
| Plaid | Securely connects bank accounts to apps | Perry |
| Plaud | Wearable recorder that records and transcribes your day | Jared |
| Quo | Business phone system that transcribes calls and texts | Jared |
| Granola | Meeting recorder and transcripts that connect to Claude | Chad |
| Obsidian | Notes app that links ideas into a graph | Jared |
| Telegram | Messaging app used to talk to your agent from your phone | Jared |
| Hostinger (VPS) | Rented cloud server so agents run 24/7 (about $200 a year) | Jared, Chad |
| MCP | A standard way to plug your own tools and data into AI agents | Jeremy, Chad |
| SCORM | Standard package format for online courses | Jeff |
| Agenda Insider | Aaron's city permit and agenda data service | Aaron |

## Things to look into or implement

**Start this week (free or close to it)**
- [ ] Use dictation instead of typing your prompts (ChatGPT or Claude app microphone).
- [ ] Write a one-page "about me and my business" file and give it to your AI at the start of every chat. This is the first piece of your AI's brain.
- [ ] Pick one task you do every week and write it down as a "skill": the steps, the rules, what good looks like. Have AI do it from that.
- [ ] Before a big request, have AI rewrite your messy idea into a clear prompt: what I want, constraints, done when, ask me first if. (Alex's Enhance button, done by hand.)
- [ ] Start a memory file. When AI learns something about you or a project, save it. Start fresh chats often instead of one endless chat.
- [ ] Ask your AI to review its own work, or ask a second AI to check it, before you use it.

**This month**
- [ ] Record your meetings and calls (Plaud, Granola, Quo, or your phone app) and feed the transcripts to your AI for follow-ups and "what did I promise?"
- [ ] Find your most expensive repeat document (proposal, quote, report, training) and build a template plus prompt that produces it every time. Measure the time before and after.
- [ ] Build one small personal app for a real annoyance (budget, workouts, recipes, a tracker) using Claude Code or Codex plus free hosting.
- [ ] Have AI build a Chrome extension for a repetitive clicking task in a website you use every day.
- [ ] Look at where expensive outside training, software, or data could be replaced by something AI builds from material you already have.
- [ ] Industry-specific: if you're in real estate, building or title, look at public city agendas (or a service like Agenda Insider) for deals before they hit the market.

**Bigger builds (when you're ready)**
- [ ] A "chief of staff" agent that reads your calls, texts, email and meetings every night and keeps a page and timeline per person.
- [ ] Run your agents on an always-on machine (a VPS or Mac mini) so work continues while you sleep.
- [ ] A manager / lead / worker / reviewer setup, where every project keeps a written goal and a dated log.
- [ ] A company hub with shared structure so every employee's AI produces consistent work, connected to your existing systems through APIs and MCP.
- [ ] Keep your brain portable: store memory and instructions in your own files so you can switch AI providers.
- [ ] Keep keys and passwords out of the AI (local config files), and let finished automations run without AI where they can.

## The presenters
- Aaron Earnest: Agenda Insider (agendainsider.com). He offered to help anyone get started with VS Code and AI.
- Perry Sessions: online advertising agency, personal app builds.
- Jared Fenn: home inspector, Evolve Customers.
- Jeremy Dye: Epic Engineering, structural engineer who built the firm's AI hub.
- Alex Sisk: BYU MBA, AI consulting. He offered to share his Enhance button on GitHub.
- Chad Mckinney: software engineering manager, business software integrations.
- Jeff Davis: compliance and licensing, Alta Pest Control.

The recap video, each presenter's full presentation, and the full transcript are at ai.meet.up.sumsionrealestate.com/recap.
