PHVA Academy & Jobs

← Why REVAs Win or Fail — the Operating System

AI-orchestration foundations

Five lessons. AI before real estate. By Day 7, the trainee runs Claude like a colleague.

Why Week 1 is AI, not real estate: Because a 2026 REVA who learns real estate first and AI second is being trained for 2018 jobs. Reverse the order. Teach the operating layer first; the domain knowledge layers on top of it in Week 2.

Format for every Week 1 lesson:

Lesson 1.1 — The 5 Outcomes

Time: 90 minutes (45 read + 45 live) Format: Pre-read + Socratic discussion + group exercise + individual decision exercise Why this lesson is first: Because every action a REVA takes for the next 4 years should map to one of these five outcomes. If the trainee leaves Lesson 1.1 unable to recite all five and articulate which one each task they did this week mapped to, they fail the lesson.

Pre-read (assigned 24 hours before)

The Cineminn Operating System, Section 2 (full read). Plus this case:

Case 1.1: The 47 Tasks It's a Monday morning in early March. Maria Chen, a top-10 KW agent in Austin, Texas, hands you (her newly-hired Cineminn REVA) a list of 47 things she needs done this week. The list is unsorted — some are time-sensitive, some are "would be nice," some are existential to her business, some are vanity projects. Excerpt from the list: 1. Schedule next week's photographer for the Beaumont listing 2. Update CRM with leads from this weekend's open house 3. Write the holiday card for past clients (December? Already?) 4. Research what color logo works best for "luxury" brokerages 5. Send pre-listing prep email to the Hayes seller (closing in 8 weeks) 6. Verify wire instructions for the Garcia closing on Wednesday 7. Update the team's Instagram bio 8. Call the inspector to confirm Thursday's inspection at the Park property 9. Cancel the Costco membership we never use 10. Compile the YOY revenue report for our 1099 prep ... (37 more items) Your assignment: Sort these 47 items into the 5 Outcomes (O1, O2, O3, O4, O5) plus a sixth bucket: "Doesn't map to any outcome — should be killed or delegated." Bring your sort to Lesson 1.1.

Trainer demo (5 minutes)

Trainer pulls 5 items from the list at random and walks through the sort live, narrating the reasoning:

Item: 'Schedule next week's photographer for the Beaumont listing.' This is O2 (Listing Sold) — pre-launch prep is part of getting the listing to close. A photographer is a vendor in service of selling the home. O2. Item: 'Update the team's Instagram bio.' This is... look, it's tempting to call this O4 (Trust Earned) because it's brand-related. But updating the bio doesn't earn trust — content earns trust. The bio is foundational, not a trust-builder. I'd put this in the kill bucket unless it's actually broken. If the bio is hurting the team — say, it's outdated and confusing — then it's O4. If it just feels low-priority, it's busywork. KILL. Item: 'Verify wire instructions for the Garcia closing.' This is O5 (Operations Clean), and specifically Law 4 — zero-tolerance wire fraud. This goes to the top of the priority list, full stop.

The point of the demo: show that the sort isn't mechanical. It requires judgment. Some items have multiple outcomes. Some are traps (look like outcomes but are busywork). The 5 Outcomes is a thinking tool, not a checklist.

Live build (40 minutes)

Phase 1 (15 min): Trainees work alone. Sort all 47 items. Be ready to defend any contested classifications.

Phase 2 (15 min): Breakout groups of 5. Compare sorts. Each group nominates the 3 most-contested items for full-cohort discussion.

Phase 3 (10 min): Full cohort. Trainer leads Socratic discussion on the contested items.

"Group 2 — you put 'Research what color logo works best for luxury brokerages' as O4 (Trust Earned). Group 4 — you killed it. Defend your positions."

This is Harvard pedagogy. The instructor doesn't lecture. The cohort argues. The trainer adjudicates and surfaces the deeper principle.

The deeper principle here: research without action is busywork. Logo color won't be tested, won't ship, won't move a metric. KILL. If Maria wants better branding, that's an L4 conversation — let's redesign the brand. Logo color research in isolation is a vanity task that pretends to be strategic.

Peer review + AI grading (10 min)

Trainees post their final sort + reasoning in Slack. A pre-built Claude Project ("Outcome Sorter Grader") scores each submission:

The top scorer of the session shares their sort. Everyone takes it.

What the trainee owns after Lesson 1.1

Next-step assignment (next 24 hr)

Take your own task list from your last 5 working days. Sort each task into the 5 Outcomes. Identify the % of your time that mapped to outcome work vs. busywork. Bring the data to Lesson 1.2.

Lesson 1.2 — Your Personal AI Stack

Time: 90 minutes Why this lesson is second: Once the trainee can identify what produces value (Lesson 1.1), they need the operating layer that produces it (Lesson 1.2).

Pre-read

OS Section 4 (the AI Stack). Plus:

Case 1.2: The 6-Tool Trap A new Cineminn REVA, Mark, just finished onboarding. He has subscriptions to Claude, ChatGPT, Gemini, REI Reply, Follow Up Boss, Granola, Loom, Canva Pro, Descript, ElevenLabs, Midjourney, and Make. Total monthly cost: $247. Mark spends his first two weeks testing every tool. By week three, he's overwhelmed. He has 14 different prompt libraries scattered across 6 tools. He doesn't remember which tool he used for which task. He's been "using AI" but his output is no faster than a non-AI VA. What did Mark do wrong? What should an L1's AI stack actually look like?

Trainer demo

The demo for Lesson 1.2 is a setup demo. Trainer walks through their own stack, live:

  1. One Claude account. Show 3 active Projects: "Cineminn Voice Library," "MN Compliance Checker" (note: this is a STATE MODULE — only loaded when needed), "Daily Brief Generator."
  2. Demonstrate a Project setup. Open Claude. Create a new Project. Drop in: Cineminn brand voice doc, 5 sample emails the agent sent that worked, the team's vocabulary. Set custom instructions: "Write in this voice. Keep messages under 120 words. Always end with a clear next step."
  3. Test the Project. Prompt: "Draft a Just Listed email for the Beaumont property at $725K, going live Friday." Watch the output land in voice, on length, with a clear CTA.
  4. Show the rest of the stack. Granola is always on during agent meetings — auto-summarizes. Loom is the async tool — every internal handoff is a 2-min Loom, not a 5-min email. Canva uses the team's brand kit — never start from blank.

The lesson: stack discipline beats tool maximalism. L1 trainees should have one Claude account, one CRM, one social scheduler, one video tool. Period. Add tools only when an outcome demands it.

Live build (50 minutes)

Each trainee builds their personal stack:

Phase 1 (20 min): Each trainee sets up 3 Claude Projects.

Phase 2 (15 min): Each trainee tests their Projects with a real prompt and submits the output to Slack.

Phase 3 (15 min): Peer review. Cohort scores each submission for voice match, brevity, action orientation.

What the trainee owns after Lesson 1.2

Next-step assignment

Use one of your Projects for a real task this week. Document: (a) what you produced, (b) how long it took without AI, (c) how long it took with AI, (d) what the AI got wrong that you had to fix.

Lesson 1.3 — Prompt Engineering for Operators

Time: 90 minutes Why third: Stack without prompting skill produces mediocre output. This is where Lesson 1.2 gets sharpened.

Pre-read

Case 1.3: Two Drafts Same task: "Draft a price-reduction email to a seller whose home has been on the market for 28 days with no offers." Operator A's prompt: "Write a price reduction email to a seller." Output: A generic 3-paragraph email with vague language about market conditions. Operator B's prompt: "You are writing in the voice of [Agent Name], a top-10 KW agent in Austin TX. The seller is the Hayes family — both spouses are mid-50s professionals, time-sensitive (relocating to Phoenix in 90 days), and emotionally attached to their home (lived there 11 years). Their home at 1234 Beaumont has been on the market 28 days. Listed at $749K. Three showings, no offers. Comparable homes sold at $710-725K. We need to recommend a $30K reduction. Draft an email that: (1) acknowledges the emotional weight without being patronizing, (2) presents the market data without sounding like a textbook, (3) frames the reduction as a strategic decision (not a defeat), (4) requests a 15-minute Zoom this week to discuss. Length: under 150 words. Tone: empathetic but direct. End with two specific time options for the Zoom." Output: A specific, useful, ship-ready email. Both prompts produce output. Why is Operator B's worth $14/hr and Operator A's worth $5/hr?

Trainer demo

Trainer demos the CDFER prompt structure (Cineminn-original):

Trainer rewrites a bad prompt live, narrating each transformation.

Live build (45 min)

Phase 1 (15 min): Trainees pick one of their tasks from this week and rewrite it using CDFER. They run both the bad and good version through Claude.

Phase 2 (15 min): Cohort breaks into pairs. Each trainee runs their improved prompt for their pair partner; pair partner critiques.

Phase 3 (15 min): Cohort discussion. Trainer surfaces the patterns — what makes prompts that work vs. prompts that don't.

What the trainee owns

Next-step assignment

Build 5 reusable prompts for your most common weekly tasks. Submit them to Slack. Cohort will vote on the best 5 across the cohort to be added to the Cineminn Prompt Library.

Lesson 1.4 — AI as Colleague: The Mental Shift

Time: 60 minutes (no live build — this is a discussion lesson) Why fourth: Skill without mindset stalls. The trainees who treat AI as a colleague — delegating, reviewing, building shared context — out-produce trainees who treat AI as a tool by 5–10×.

Pre-read

Case 1.4: The Two Mark McKenzies Two REVAs at Cineminn, both named Mark McKenzie (anonymized; not real). Same starting cohort, same training, same Claude account, same agent. Mark A: Treats Claude as a tool. Opens it 4–6 times a day for one-off prompts. Writes a prompt, gets output, copies into email, sends, closes. Doesn't track which prompts worked. Doesn't build Projects. After 6 months, Mark A has produced standard work at standard pace. Mark B: Treats Claude as a colleague. Has 12 active Projects covering recurring workflows. Spends 30 min every Monday "training" Claude on the prior week's wins (drops in the best emails, the best client interactions, the best cases). Treats every Claude output like a draft from a junior colleague: reviews, edits, files the good stuff back into the Project. After 6 months, Mark B has built a system that runs at 5× the pace of Mark A and produces visibly better work. Same starting point. Different mindsets. The mindset is the variable.

Discussion (45 min, full cohort)

This is a Socratic discussion. Trainer asks, doesn't tell.

The discussion forces the trainees to articulate the mindset. The trainer doesn't deliver a lecture; they extract the mindset from the cohort's own thinking.

What the trainee owns

Next-step assignment

Write a 200-word reflection: how will you specifically work with Claude differently after this lesson? Submit to Slack. Trainer reads every submission.

Lesson 1.5 — AI Ethics, Disclosure, and Liability

Time: 60 minutes Why fifth: Before the trainee uses AI on real client work in Week 2+, they need the ethical and legal frame. This is the guardrail lesson.

Pre-read

Case 1.5: The Hallucination That Cost $80K A REVA at another agency (not Cineminn) used ChatGPT to draft a market analysis for a seller. The AI cited a "$725,000 average sale price" for the seller's neighborhood, with no source. The REVA didn't verify. The agent presented the analysis to the seller. The seller listed at $735K based on the analysis. Three months later, the home was still active. Comps showed the actual neighborhood average was $640K. The seller had passed on a real $670K offer in week 2 because the analysis suggested they could get more. The seller eventually closed at $625K — $45K below the offer they'd refused, $80K below their initial expectation. The seller sued. The agent's E&O insurance covered most of it. The agent fired the REVA. The REVA's career in real estate ended. What happened? What rules would have prevented this?

Trainer-led content + discussion (50 min)

The 5 AI rules every Cineminn REVA follows:

Rule 1 — Verify every fact before client delivery

AI hallucinates. Especially numbers. Especially specifics. Every fact in client-facing output gets verified against a primary source. If you can't find the primary source, the fact gets cut.

Rule 2 — AI disclosure when appropriate

Client-facing analytical work (CMAs, market reports, comparative analysis) gets a disclosure: "This analysis was prepared with AI assistance and reviewed by [operator]." Not because the law requires it (yet, in most states). Because trust requires it.

Rule 3 — No AI for legal interpretation

AI does not interpret contracts. AI does not advise on contract clauses. AI does not predict litigation outcomes. Cineminn REVAs flag these for the agent or the agent's attorney. AI is the drafter, not the lawyer.

Rule 4 — No AI for personal client communication

AI drafts the email. The operator edits it. Pure-AI client communication — where the operator just hits send on Claude's output — is forbidden. The client can tell. Trust erodes.

Rule 5 — Document what AI produced, what you edited

A Cineminn REVA keeps a simple log: "Claude drafted this, I edited X and Y." This protects everyone if a client question arises later.

What the trainee owns

Next-step assignment (the Week 1 capstone)

Build the "Cineminn Prompt Library v0.1." Each trainee contributes 1 production prompt with: the task it solves, the CDFER breakdown, the output it produced, the editing it required. Cohort builds the library together. Top 20 prompts make the official Cineminn Prompt Library.

Week 1 — Trainee Output (what they have at end of Week 1)

By Day 7, the trainee is operating at L1+ — task operator with AI orchestration baked in. Week 2 layers real estate on top.

Cineminn REVA Academy · Week 1 Complete · v1.0 · May 2026

← The Operating System — 5 Outcomes, L1–L4, 7 Laws
Join free to track progress  Reference: the 5 Outcomes cheat card and career ladder →