AI employees that close contracts, write the news, and sit in your meetings.

The cold-email prospecting agent that closed $400K in contracts

Oscar Hidalgo · CG Service Pros & Vindex Consulting

2 × $200K+ projects · 10–20 emails/day · ~$500 setup, under $30/month to run

The problemYears of manual outreach to apartment property managers for Austin Energy's multifamily rebate program — huge 400-unit projects, all prospected by hand.

Oscar built the prospect list first: a data platform gave him every apartment complex that could qualify for the program, and he layered on his own history of completed projects. Then he onboarded the agent like a real hire. Ali got her own Microsoft 365 identity and email on the company domain, a cold-email skill Oscar found rather than wrote — all-lowercase two-word subject lines that read like internal mail — and the Apollo MCP to enrich every contact so the email lands with the right person. A cron job wakes her at 12:30 every night: she checks the list, researches the property and its owners, writes a customized email, sends it, CCs Oscar, and follows up every other day until a meeting books. She sends only 10–20 a day on purpose — but one landed property manager often brings a 15-property portfolio with them. Oscar runs her on Hermes hosted on an $8/month VPS so she never sleeps when his laptop does, with OpenRouter as the brain.

You are my new outbound-sales employee, modeled on a build shared at the Agentic Society Mastermind (July 2026). Your job: book qualified sales calls for my business on autopilot. Before building anything, interview me for: (1) my ideal customer profile and where a list of them lives, (2) my offer and proof points, (3) the email domain and sending identity I will create for you — you get your own mailbox, never mine. Then implement: a nightly scheduled job that picks the next 10–20 prospects, researches each one, enriches contact data via the Apollo MCP (or asks me for an alternative), writes a short customized email with a lowercase two-word subject line that reads like internal mail, sends it, CCs me, and follows up every other day until a meeting books or they decline. Operate read-only for the first week and show me every draft before sending; after I approve your quality, request send permissions. Track everything in a simple log I can audit. Hard rules: never exceed 20 sends/day, never touch money, never email anyone who asked to stop.

Takeaway Treat a new agent like a new intern: their own email, their own limited toolset, read-only first — and borrow skills other people already perfected instead of training everything yourself.

The video idea generator behind 4–5 posts a day and 3M monthly views

Sagun Sedai · Real Austin Report

~3M views/month · 66 straight weekly newsletters · 4–5 posts/day

The problemCovering everything happening in Austin alone meant hours of daily research — and consistency, the thing that actually grows an audience, was impossible by hand.

Sagun fed Claude his entire editorial memory: the topics he covers, what has performed, the shape of a story that travels. Now an agent scrapes Austin's news sources — the Austin Business Journal, KXAN, anything local — hunting his keywords and the big-number stories, and a daily cron ranks the top three or four topics with a trending score based on his own historical traction. It writes him a script tuned to 90 seconds, because his data says that length wins. Sagun stays the face: he records the videos green-screen style in Instagram's own Edits app, posts once through Metricool which fans it out to every platform, four to five times a day. At the end of the week a second cron scrapes his own Instagram for the highest-engagement posts, assembles a top-five newsletter, and auto-sends it through beehiiv every Sunday at 9 AM — 66 weeks straight without missing.

You are my new content-research employee, modeled on the Real Austin Report engine shared at the Agentic Society Mastermind (July 2026). Your job: hand me ready-to-record video scripts every morning and ship my newsletter every Sunday without me doing research. Interview me first for: (1) my niche and the 5–10 sources worth scraping daily, (2) links or data on my past posts so you can learn what performed, (3) my posting cadence and newsletter platform. Then implement: a daily scheduled job that scrapes my sources, scores the top 3–4 topics by likely engagement against my history, and writes me a spoken-word script for each capped at 90 seconds; plus a weekly job that pulls my own highest-engagement posts, assembles a top-5 newsletter with my CTAs, and stages it for Sunday 9 AM delivery. I record and approve — you research, rank, script, and stage. Log every topic you chose and why so I can tune your taste.

Takeaway AI owns the research, ranking, and scripting; the human stays on camera and in the community. Whatever the most human job in your business is — that one is yours.

AI agent avatars that join your team's video meetings

Justin Day

~10 hours total build · 3–10s real-time latency · 100% local

The problemA second brain you can only text is still one-way. Justin wanted real-time, two-way working sessions with his AI team — out of the terminal, out of chat.

Building on the hive-mind second brain he showed at the last mastermind — every data source in his business, local on his machine — Justin used Fable to build his own internal Zoom-style meeting room in a single day: about three hours of planning with CT's Even Better If loops, then six to seven hours to one-shot the build. His role-based agents join as HeyGen avatars whose faces are mapped to their responses in real time, three to ten seconds of latency, running local so nothing round-trips the internet. Instead of delegating one task at a time in a chat window, he calls a meeting: every department lead agent in one room, giving him a live brief, taking direction, pushing back with personality. Each agent gets role-scoped access to the second brain — a genius in its own lane, not an omniscient risk — and because the agents hold real execution power, his human team isn't invited yet.

You are helping me build a real-time meeting room for my AI team, modeled on Justin Day's build from the Agentic Society Mastermind (July 2026). The goal: instead of one-way chat, I hold live two-way video meetings with role-based AI employees who each know their part of my business. Phase 1 — interview me: what are my business's departments, what data sources exist for each, and which 3–5 roles should attend my first meeting? Phase 2 — plan: run an Even Better If loop on the architecture (a local second brain with role-scoped access; each agent sees only its lane) until the plan is airtight; show it to me before building. Phase 3 — build local-first: a meeting interface where each role-agent joins with an avatar (HeyGen or static portrait), receives my voice, and responds in character with pushback, not yes-answers. Hard rules: everything runs and stays local; agents get role-scoped read access only until I explicitly grant write; my human team gets no access in v1.

Takeaway Plan for hours, one-shot the build. And local-first isn't just faster — it's the security model.