Skip to content
GuidesReal EstateMarketingFamilyLive EventseSports

Production sheetFree classRun time 22:15

All free classes

Everyday AI workflows for real estate agents: How do I set up AI workflows for my real estate business?

Pick one real job on your desk, like a listing launch, and write down its steps: what starts it, the facts that go in, what the AI drafts, the check it has to pass and the finished piece. Then tell the AI the outcome, give it only the facts you trust, and review every draft before it leaves your hands.

Sheet 01

Storyboard: the 14 slides.

The class deck, in order and as it was taught. Pick a shot or use the arrow keys.

Slide 01 / 14What You're Getting Today
What You're Getting Today. Structured learning first. Lab work second.
Read all 14 slides as text

  1. 01 What You're Getting Today

    Structured learning first. Lab work second.Structured learning: Communication as the meta-skill: human-human, human-AI, AI-AI, and chains of all three.The operating idea: turn messy intent into usable input: outcome, context, constraints, evidence, owner, and next action.The lab portion: Everyday AI Workflows: work through your personal inbox, listing, CMA, and daily briefing workflows.The take-home blocks: Communication Diamond, Signal Check, Human-AI Spec, validation loop, quick context sheet, and workflow recipes.The teaching block gives you the structure. The lab block makes it real against your work.

  2. 02 Communication Is The Future Operating Skill

    AI AccelerationReal EstateMETA SKILLS · CLASS 02Human to human. Human to AI. AI to AI. Chains of all three. This is how work moves now.Presented by Build Things That Build Things

  3. 03 The New Communication Map

    Same meta-skill. Different receivers.Human to human — client calls, showing recaps, team asks, lender check-ins, partner work.Human to AI — prompts, quick context, specs, guardrails, audience, intent, and destination.AI to AI — one model or agent creates an artifact, validates it, then passes it to the next step.Chains — human -> AI -> AI -> human -> tool. Useful work is becoming a chain, not a single chat.Handoffs are the stress test. The larger skill is designing communication so meaning survives every receiver.

  4. 04 THE COMMUNICATION DIAMOND

    Human Communication For AI EntryCapture these fields and AI can turn the conversation into tasks, drafts, reminders, workflow inputs, and missing-info checks.WHO'S RESPONSIBLE?OwnerWHATOutcomeWHENDue dateHOWPathBLOCKERConstraintMOTIVATIONReward loopHOVER A DIAMOND POINTMotivationASK THISWhy will this move?Motivation is the reason the doer, client, or team member will actually care enough to act.Ask: what does this do for me, for them, for the client, or for the project?AI ENTRYShapes tone, urgency, persuasion, client framing, and follow-up language.

  5. 05 Your Format Does Not Have To Be Mine

    "The format is not sacred. Repeatability is the point."My transcript nuance matters: this is the best format I have found for myself. Every operator needs a repeatable capture format that fits how they actually work.

  6. 06 The Human-AI Communication Spec

    What the transcript says people still miss.Perfect output — what the finished thing should look like in format, quality, and destination.What not to do — guardrails, banned phrases, compliance limits, style misses, and no-go list.Prior attempts — what you already tried, what failed, and what should not be repeated.Business context — brand, ICP, offer, market, tools, workflow, and operating constraints.Audience + intent — who reads it, whose voice is speaking, and what reaction it should create.Missing-info question — ask the AI what else would help before it writes.A strong prompt is not just an ask. It is outcome, context, constraints, intent, destination, and a loop for missing information.

  7. 07 Context Helps. It Is Not The Class.

    Cards are accelerators. The communication spec is the skill.If you have cards: load only the one or two that matter for this workflow.If you do not: write quick context: who you are, the situation, facts, audience, constraints, and desired output.Do not paste the whole folder: more context is not automatically better communication.The class skill: know what information the receiver needs and what would make the work fail.The bridge: cards make repeated work faster, but the Diamond and Spec make today usable for everyone.Context Cards are useful LEGO blocks. They are not the ticket to enter the room.

  8. 08 AI-AI Validation Loop

    Before one AI passes work to the next, it checks the human outcome.CopyBefore passing this output to the next step, validate it against the operator criteria.Outcome:- What was this step supposed to produce?Handoff document:- What format does the next AI, tool, or human expect?- What required fields must be present?- What evidence or source material must travel with it?- What should be forbidden or marked [review]?Validation:- Does it match the requested output?- Does it include the required sections or fields?- Did it violate a guardrail?- Is anything missing that the next receiver expects?If any check fails:- Name the failure.- Resubmit the job with the missing requirement.- Do not pass the artifact forward until it passes.

  9. 09 Keep Meaning Intact Through The Chain

    COMMUNICATION CHAIN -> EXECUTION CHAINThe future skill is not one perfect prompt. It is keeping the work coherent as it moves across people, models, tools, agents, and back to you.HumanconversationAIspec + contextAItransform + validateHumanreview + judgmentToolCRM, Gmail, NotionCHAIN OF COMMUNICATIONEvery receiver gets the same operating truth.Diamond, spec, context kit, bullshit detector, validation loop.CHAIN OF EXECUTIONmeaning intactcontext intactevidence intactowner intactdone criteria intact

  10. 10 Run Pipelines, Not Prompts

    AI AccelerationReal EstateINNOVATION LAB · EVERYDAY AI WORKFLOWSA useful workflow starts with real work, filters the context, produces a business artifact, and gets reviewed before it leaves your hands.Presented by Build Things That Build Things

  11. 11 Sample Workflow Menu

    Pick the pipeline that matches the work actually sitting in front of you.Listing launch — seller intake or listing notes -> complete launch kit.Virtual staging — listing photos and room notes -> staging brief, image review, disclosure copy.Buyer consultation — lead inquiry -> consult plan, search criteria, follow-up, CRM note.CMA / pricing narrative — subject property and comps -> seller-facing pricing memo.Showing feedback / seller update — feedback and activity metrics -> pattern summary and next recommendation.Post-inspection negotiation — inspection report -> options, client explanation, repair request or response.Sphere reactivation — CRM export -> segments, outreach plan, replies, next-action queue.Daily deal desk — inbox, calendar, and CRM notes -> triage, risk, owner, next move.The value is the chain from messy input to reviewed output. The AI step is only one station in the chain.

  12. 12 Live Build — Listing Launch Pipeline

    Client sends listing notes. The pipeline turns them into a reviewed launch kit.Trigger: seller intake email, listing appointment notes, MLS draft, or phone transcript arrives.Context filter: extract verified facts, missing facts, strongest buyer angles, seller constraints, and no-go claims.Production: build or update the HOME/listing context, then produce MLS remarks, social assets, email blast, open-house copy, and seller update.Validation: check every factual claim, Fair Housing risk, MLS/brokerage rules, voice drift, disclosure needs, and unsupported hype.Human approval: agent edits, fills missing facts, rejects weak sections, and only then moves assets to MLS, email, social, print, or CRM.Final deliverable: one listing launch folder with source notes, context, approved copy, channel-specific assets, and the seller communication record.This is the workflow we run live because it shows the whole business path: inbox to context to production to approval to launch.

  13. 13 Workflow Failure Review

    Bad output does not automatically mean missing context.Start here: was the source material present, accurate, specific, and allowed to be used?Source failure: facts were weak, stale, unverifiable, private, or too vague.Context failure: cards or notes conflicted, overrode each other, or pointed at the wrong audience.Spec failure: the output format, destination, quality bar, or done definition was unclear.Sequence failure: the workflow skipped extraction, review, segmentation, or approval.Tool failure: the wrong model or app was used for the job.Review failure: there was no pass/fail checklist before the output left the chat.Execution failure: the AI had enough input and still produced bad work; reject it, split the task, or change tools.The fix might be better source material, better context, a sharper spec, a different tool, a human review gate, or a redesigned workflow.

  14. 14 7-Day Pipeline Plan

    TODAYChoose one real pipeline from the menu and write its trigger, source material, review gate, and final deliverable.TOMORROWRun the Listing Launch or Daily Deal Desk pipeline on live work, not a made-up example.DAY 3Add a review checklist to the pipeline before any output gets sent, posted, printed, or filed.DAY 5Save the full runbook: source input, context filter, AI output, edits, approval notes, and final artifact.NEXT LABBring the pipeline break: source, context, spec, sequence, tool, review, or execution.

Watch the run.

22 minutes. The communication toolkit first, then the pipelines that put it to work.

RecordingEveryday AI Workflows for Real Estate Agents22:15Watch on YouTube

Read and practice on this page. Everything on this page is free: the recording, the slides, the prompts, and the workbook. No account needed. The free portal account is for downloading the files.

Sheet 02

Set up your first workflow in 6 steps.

The order the class teaches it, one real job at a time.

  1. Pick one job on your desk

    Don't try to automate your whole business. Pick one job that's in front of you this week, like a listing launch, a buyer consult or your daily inbox.

  2. Write down the steps you already take

    A lot of agents know what they do when a new listing comes in, but it isn't written down anywhere. Write it the way you'd train a new assistant: what starts the job, what goes in and what comes out.

  3. Say the outcome in one sentence

    “Follow up with the seller” is too vague. “The seller picks a pricing plan before we prep the listing Friday” tells the AI what done looks like. Add what's blocking it and who owns it.

  4. Supply current, permitted facts

    Use a fictional example or only task-relevant facts allowed in the selected account. Mark uncertain facts [VERIFY] and provide the one or two approved context excerpts needed. Ask what evidence is missing before drafting.

  5. Check the draft and bound retries

    Ask AI to flag unsupported claims and possible conflicts with your rules, then inspect the actual draft yourself. Name one bounded repair and a retry limit. If the result still fails, stop and return the draft with the unresolved issue instead of looping indefinitely.

  6. Keep an appropriate review record

    Review facts, permissions and destination before any authorized client use. Save the reusable instructions, permitted source references and result notes in approved storage. Do not retain an entire private conversation simply to reuse the workflow.

Use the class workbook · Get the supporting files

Read and work through the lesson here. Downloadable files use your free student account.

Sheet 03

What it looked like in class.

Ryan's own words from the recording. Each time stamp opens the class at that moment.

  • Watch from 5:10

    “Hey, the outcome I want is the seller to choose a pricing plan.”

    Ryan turns a vague note about a nervous seller into one clear outcome, then adds how, when, motivation, constraint and owner.

  • Watch from 10:33

    “I never want to see AI use the word phrase dream home. You can put guard rails in there”

    A banned phrase becomes a check the AI runs on its own draft, and it redoes the draft when it finds the phrase.

  • Watch from 12:15

    “human in the loop, which means the human is reviewing and quality controlling every single um, output that you get”

    The agent reviews every output in the chain before it goes anywhere.

  • Watch from 16:57

    “lay out all the steps that the AI is going to have to follow just like you would if you were onboarding an administrative assistant”

    You set up a workflow by writing the job down as steps, the way you'd train a new hire.

  • Watch from 18:51

    “a prompt is write me a listing description and the output depends on what you remember to uh paste what the AI guesses”

    A one-line prompt leaves the result to chance. A pipeline sets the trigger, source, check and finished piece first.

Sheet 04

Where AI workflows go wrong.

Starting with a one-line ask
“Write me a listing description” leaves the AI to guess. What you get depends on what you remembered to paste and whether you catch the mistakes.
Leaving the outcome fuzzy
“Talked to the seller, need to follow up” doesn't tell anyone what done looks like. Say the outcome first.
Pasting your whole folder
More context isn't automatically better. Load only the one or two cards that matter for this job.
Letting AI steps pass work along unchecked
Work can go through several AI steps before you see it, and by then it can be way off. Have each step check its work before it hands off.
Blaming every bad draft on missing context
Sometimes the source notes were weak, a step got skipped or it was the wrong tool. Check each one before you decide what to fix.
Sending a draft you haven't read
AI drafts aren't sent blind. You're the licensed agent, and you own what goes out.

Sheet 05

Your next question.

Job 01 · Listing launchFictional example

  1. 01Input

    Forms A · B

    Seller intake email: “renovated kitchen.” No year.

  2. 02Spec

    Forms C · D · G

    MLS remarks, 700 characters. No claim the facts can’t back.

  3. 03Owner

    Form F

    You, the licensed agent. AI drafts, you approve.

  4. 04Check

    Form E

    No year, so no “new kitchen.” Marked [verify].Checked

  5. 05Output

    Sign-off line

    One launch folder: source notes, approved copy, seller record.

You leave with one repeatable shape for capturing work, a clear spec for AI, a check at every handoff, and a run from trigger to a reviewed deliverable.

Taught live by Ryan Wanner

Venue Innovation Lab North, Compass Lakeside, Hendersonville

Each class was taught live by Ryan Wanner to working real estate agents.

Sheet 06

The forms.

Seven forms, each feeding one column of the sheet. The format isn't sacred. Repeatability is the point: change the labels to fit how you talk, but give AI the same shape every time.

  1. Form AFeeds Input

    Communication Diamond

    human communication for AI entry

    Turn a messy conversation into predictable input for AI, a task system, or a teammate.

    # Communication Diamond

Use this after a client call, team meeting, showing recap, lender check-in, or any conversation that needs to become AI-usable work.

## Outcome / clarity of outcome
What needs to happen? Write it in one sentence the receiver could repeat back.

## How / training needed
How do we get the outcome? What instruction, example, process, or training does the doer need?

## When / due date
When is it needed by? What depends on that timing?

## Motivation / reward loop
Why does this matter? What does it do for me, for them, for the client, or for the project? What keeps the doer moving?

## Constraint / missing resource
What is blocking completion? Is a resource, decision, access, fact, or approval missing?

## Responsible / accountable owner
Who is doing the work? Who ultimately owns successful completion?

## AI entry instruction
Turn this diamond into the right next artifact: CRM note, task, calendar item, client follow-up, project update, workflow input, or missing-information checklist.

  2. Form BFeeds Input

    The signal check

    three checks

    A filter for vague language before it becomes a task, prompt, message, or workflow step.

    # Signal Check

Use this before a claim, phrase, assumption, or recommendation becomes a task, prompt, client message, slide, workflow step, or decision.

## Input being tested
What is being said?

## 1. Logic
What does this mean? Translate the language into plain operational meaning.

## 2. Evidence
How do you know? What fact, behavior, result, source, or example proves or contradicts it?

## 3. Utility
So what? Who cares? What does this help a real person do, decide, avoid, or improve?

## Signal verdict
Signal, thin signal, or noise?

## Bundle check
If the message uses words like clarity, focus, alignment, professional, urgent, or done, unbundle the word into observable behavior, evidence, and utility.

  3. Form CFeeds Spec

    Unbundling Language

    vague word to executable spec

    Turn vague business language into behavior, constraints, evidence, ownership, output, and validation.

    # Unbundling Language

Use this when someone says clarity, focus, alignment, professional, done, urgent, polished, strategic, or any other bundled word.

## Bundled word or phrase

## Situation
Client, deal, listing, task, workflow, or decision context.

## Meaning
What does this word actually mean here?

## Behavior
What should a person or AI actually do?

## Constraint
What should not happen? What rule, boundary, or no-go list matters?

## Evidence
How do we know this is true, complete, useful, or ready?

## Owner
Who is responsible for the next move?

## Output
What artifact, message, field, file, or task should exist?

## Validation
What check proves it is good enough to pass forward?

## Action change
What changes next because we unbundled this?

  4. Form DFeeds Spec

    Human-AI Spec

    before you ask the model

    The missing context checklist from the Communication meta-skill.

    # Human-AI Spec

## Perfect output
What should the finished output look like?

## Format
Bullets, email draft, CRM note, Notion page, Salesforce field, Gmail draft, table, memo, or checklist?

## What not to do
Banned phrases, compliance limits, tone no-go list, and anything that would make this unusable.

## Prior attempts
What have I already tried? What failed?

## Business context
Brand, offer, market, operating rules, audience, and tool constraints.

## Context available
Quick Context Sheet:
Optional existing cards loaded:
Source notes or facts pasted:

## Audience and intent
Who reads this? Whose voice is speaking? What should the reader feel, understand, or do?

## Desired destination
Where does the output go next?

## Missing-info question
Before writing, ask: what else would help you solve this problem?

  5. Form EFeeds Check

    AI-AI Handoff Contract

    The contract for one AI step passing work to the next AI, tool, or human.

    # AI-AI Handoff Contract

## Step name
What did this AI just do?

## Receiving step
Who or what receives this next?

## Input shape
Format, required fields, forbidden fields, and source requirements.

## Output shape
What must the next step receive?

## Done definition
What conditions prove this step is finished?

## Validation checklist
- Does it match the requested output?
- Does it include required sections or fields?
- Does it include evidence or source notes?
- Did it violate a guardrail?
- Is anything missing that the next step expects?

## Resubmit rule
If any validation check fails, name the failure and resubmit the job before passing it forward.

  6. Form FFeeds Owner

    Chain Map

    communication to execution

    Map the humans, AI models, tools, and handoffs in one workflow.

    # Chain Map

## Workflow name

## Final outcome

## Leg 1
Sender:
Payload:
Receiver:
Type: human-human / human-AI / AI-AI / AI-tool / human-tool
Done definition:
Validation:

## Leg 2
Sender:
Payload:
Receiver:
Type:
Done definition:
Validation:

## Leg 3
Sender:
Payload:
Receiver:
Type:
Done definition:
Validation:

## Weakest leg
Where does meaning, context, evidence, or ownership disappear?

## Fix
Which source, context, spec, sequence, tool, validation rule, or human approval gate needs to be improved?

  7. Form GFeeds Spec

    Quick Context Sheet

    no cards required

    The minimum context needed to run today's workflows if you do not already have cards.

    # Quick Context Sheet

Use this when you do not have Context Cards yet. Paste it above one workflow pipeline step.

## I am
Role, market, voice, and how direct or polished the output should feel.

## Situation
What workflow are we running and why does it matter today?

## Known facts
Paste only facts you trust. Mark uncertain details as [verify].

## Audience and intent
Who receives the output? What should they do, feel, understand, or decide?

## Constraints
Compliance, privacy, banned phrases, deadlines, missing facts, and no-go zones.

## Output shape
Email draft, CRM note, memo, checklist, briefing, table, or next-action list.

## Optional cards loaded
If you already have SELF, GUARDRAILS, HOME, CLIENT, PERSONA, or CHANNEL, name only the ones loaded for this workflow.

Sheet 07

The runs.

Four pipelines from the workbook. Each starts with real work, loads the forms it needs and says exactly what the output must look like. Pick the one on your desk this week.

  1. Run 01

    Daily Deal Desk Pipeline

    Load Quick Context Sheet (G), inbox slice, optional SELF/CLIENT/GUARDRAILS, Human-AI Spec (D)

    Note Use a small slice first. Do not paste private data into a tool your brokerage forbids.

    Prompt to AI

    Use this inbox/calendar/CRM source packet to create a daily deal desk brief. Return urgent items, this-week items, FYI, draft replies only where enough context exists, and one missing-info question where it does not.

    Format spec

    • Buckets: URGENT, THIS WEEK, FYI, DRAFT REPLIES, MISSING CONTEXT.
    • No invented client facts.
    • Drafts must match the pasted voice context.
    • Each actionable item ends with the smallest next step.

  2. Run 02

    Listing Launch Pipeline

    Load Quick Context Sheet (G), property facts, audience, destination, optional HOME/CHANNEL/SELF/GUARDRAILS

    Note Cards help if you have them. Without cards, paste facts, format, voice, and safety constraints directly.

    Prompt to AI

    Using the pasted listing intake, property facts, audience, destination, and constraints, extract the listing context first, then create MLS-safe copy, channel-specific launch assets, missing facts, and a compliance check.

    Format spec

    • 1-sentence property angle.
    • 700-character listing description.
    • 5-bullet content brief for the selected channel.
    • Missing facts list.
    • Compliance check.

  3. Run 03

    CMA / Pricing Narrative Pipeline

    Load subject property notes, pulled comps, constraints, optional HOME/GUARDRAILS, The signal check (B)

    Note This is a reasoning aid, not a replacement for MLS judgment or agent responsibility.

    Prompt to AI

    Build a client-ready pricing narrative from the pasted subject property, constraints, and comps. Flag weak evidence instead of pretending the comps support a conclusion.

    Format spec

    • Subject property summary.
    • Comparable rationale.
    • Price range with reasoning.
    • 2-sentence market context maximum.
    • Recommended next step.
    • Verify-before-sending list.

  4. Run 04

    Buyer Consultation Pipeline

    Load Quick Context Sheet (G), buyer inquiry, CRM history, search data, optional SELF/PERSONA, Chain Map (F)

    Note Do not assume financing, motivation, family status, protected-class details, or seriousness from thin data.

    Prompt to AI

    Turn this buyer inquiry and available CRM/search context into a consult prep packet, missing-question list, search criteria draft, follow-up email, and CRM-ready note.

    Format spec

    • Buyer criteria and urgency.
    • Missing questions.
    • Consult agenda.
    • Search criteria confirmation.
    • Follow-up email.
    • CRM-ready note.

Sheet 08

The stations.

The class ran on three tools. Start with the one you already open every day, then run its first five minutes.

Station 01

ChatGPT

Fast daily AI surface for inbox triage, briefing, and first-pass drafts.

Best for You want the lowest-friction place to run the workbook pipeline steps during class.

Set up

  1. Open ChatGPT Free works for the class. Paid tiers give more context and better models.
  2. Create one Project for this class Name it Compass Lab Communication + Workflows. Add your Quick Context Sheet. If you already have Context Cards, add only the ones you actually use.

First five minutes

  1. Paste the Communication Diamond template into a new chat.
  2. Paste one recent client or team situation.
  3. Ask the model to identify the missing outcome, how/training, due date, motivation, constraint, and responsible owner.
  4. Save the cleaned result into your notes.
  5. Use that cleaned input as the source material for one workflow pipeline.

Station 02

Gemini

Strong Google Workspace companion when your work lives in Gmail, Docs, and Drive.

Best for You want to turn inbox, calendar, and document context into daily operating briefings.

Set up

  1. Open Gemini Use the same Google account you use for business if your workspace policy allows it.
  2. Create a Gem for your cards Paste the Human-AI Spec and Quick Context Sheet into the Gem instructions. Add existing Context Cards only when they are relevant.

First five minutes

  1. Paste your Quick Context Sheet or one existing card if you have it.
  2. Paste today's calendar or a small inbox slice.
  3. Ask for a daily briefing under 250 words.
  4. Check whether it invented anything.
  5. Add the missing rule or fact back into your quick context.

Station 03

Perplexity

Search-first AI for sourced facts before they become client-facing claims.

Best for You need citations for market context, neighborhood facts, or CMA-supporting research.

Set up

  1. Open Perplexity Use it for facts, not voice matching.

First five minutes

  1. Ask for sourced market facts about your subject area.
  2. Open every citation you plan to use.
  3. Paste verified facts into your workflow notes or HOME card if you already use one.
  4. Mark anything uncertain as [verify].
  5. Run the CMA / pricing narrative pipeline only after facts are sourced.

Sheet 09

The sign-off line.

Two checks before anything leaves your hands. First the AI checks its own work against the handoff. Then you sign off.

The validation loop

The deck's own prompt. Paste it at the end of any AI step, before the work moves to the next AI, tool or person. It pairs with Form E, the AI-AI Handoff Contract.

Before passing this output to the next step, validate it against the operator criteria.

Outcome:
- What was this step supposed to produce?

Handoff document:
- What format does the next AI, tool, or human expect?
- What required fields must be present?
- What evidence or source material must travel with it?
- What should be forbidden or marked [review]?

Validation:
- Does it match the requested output?
- Does it include the required sections or fields?
- Did it violate a guardrail?
- Is anything missing that the next receiver expects?

If any check fails:
- Name the failure.
- Make one revision with the missing requirement, then check again.
- If it still fails, stop and return the failed check and missing input to me.
- Do not pass the artifact forward until I have reviewed it.

Your sign-off

Tick each box before a draft goes to a client, the MLS, or social.

Sheet 10

This week's production schedule.

Five dates, one pipeline. Start with fictional inputs, then try permitted work after checking the chain.

  1. TODAY

    Save the Communication Toolkit and Quick Context Sheet in one folder

    Communication Diamond, Signal Check, Human-AI Spec, AI-AI Handoff Contract, Chain Map, and Quick Context Sheet. Add existing cards only if you already have them.

  2. TOMORROW

    Run the Daily Deal Desk before opening email

    Use calendar, inbox snippets, CRM notes, and active deal risks to produce a short execution brief.

  3. DAY 3

    Run one pipeline on a small real source packet

    Use a buyer inquiry, listing intake, showing feedback set, inspection report, or 10 to 20 inbox messages.

  4. DAY 5

    Run Listing Launch or CMA / Pricing Narrative on one live deal

    Start with source material, extract facts and gaps, produce the artifact, then use the review gate before client-facing use.

  5. DAY 7

    Run a Workflow Failure Review

    Check source, context, spec, sequence, tool, review gate, and AI execution before deciding what to fix.

Sheet 11

The binder: class files.

Create a free student portal account to download this class's files.

  1. Three-Spec Worksheet

    Fillable worksheet — Communication Diamond, AI Spec, Handoff Contract, plus an OODA pass on the leakiest leg. One per workflow.

    Markdown5 KBFree account

    Get the file: Three-Spec Worksheet

  2. Failure Modes — Quick Reference

    One-page diagnosis tree for the 3 ways chains break (unspecced, wrong-spec, frozen-spec) plus repairs.

    Markdown6 KBFree account

    Get the file: Failure Modes — Quick Reference

  3. Chain Anatomy Cheat Sheet

    Sender → Payload → Receiver labels, OODA-on-every-leg reminder, two real chains mapped end-to-end, and which spec goes on which leg.

    Markdown7 KBFree account

    Get the file: Chain Anatomy Cheat Sheet

  4. Class skill card

    The class method on one page, written so you can hand it to your AI as instructions.

    Markdown14 KBFree account

    Get the file: Class skill card

Next on the sheet AI Automations for Real Estate Agents. Once one pipeline runs clean by hand, that class shows you how to put a recurring job on a schedule while you keep the send button.

Your pipeline, specified

Write the spec for the workflow your business needs most.

Clear specs and checked handoffs are how pipelines hold up. The Agent Constraint Accelerator writes the detailed implementation plan for your one constraint and supports you through it with async help from Ryan Wanner's team, with a reply within one business day.

Sheet 12

Short answers, from the class material.

What is an AI workflow for a real estate agent?

A workflow is a chain, not a prompt. It starts with a real trigger like a seller intake email, filters the facts, has AI draft the output, runs a review, and ends with an approved deliverable such as a listing launch kit or a CRM note. The AI step is one station in the chain.

Why does my AI output break even when my prompt looks good?

Often the prompt isn't the problem. Run a failure review: was the source material weak, did the context conflict, was the spec unclear, did the workflow skip a step, was it the wrong tool, was there no review, or did the AI just fail? Only one of those is missing context.

Do I need context cards before I can use these workflows?

No. Context cards speed up repeated work, but they aren't the ticket in. Without them, write quick context: who you are, the situation, the facts you trust, the audience, the constraints, and the output you want. Turn it into a reusable card later if the workflow keeps repeating.

How do I stop one AI step from passing bad work to the next?

Add a validation step before every handoff. Tell the model what the step was supposed to produce, what the next receiver expects, and what's forbidden, such as a phrase you never use. If any check fails, it names the failure and redoes the job before passing anything forward.

Which workflow should I build first?

Choose a familiar job, such as a listing launch, a daily deal desk, a buyer consultation plan, or a pricing narrative. Practice with fictional inputs first. Use permitted real work only after checking the chain, and save the inputs, output and review together.

Sources

All free classes

Sheet 13

Read the class.

Automatic captions from the recording, lightly corrected. Stretches with personal details, unsourced numbers, dated news or advice the class has since changed are left out. That's 12 stretches in all.

Read the transcript, 10 chapters

0:00 Break work into small blocks

I've been in here um give you a small structured learning and then we're just going to build stuff that's useful for you guys um is the plan. So that's kind of my plan anytime there's a small group is the entire curriculum is going to change to actually building real stuff that solves your problems. So we're just going to have kind of a small group of like actually building things that are you'll be able to deploy within your business. So um the last lab we did we started thinking about context in the form of Lego blocks. So meaning my can break things into small problems, you have a much um higher likelihood of finding the next actionable step you can take to actually solve that problem, right? So it's not sell a house, it's prepare a listing description, right?

Like you break it down into each of the phases of selling a house. And that's when you have those phases laid out, we can then add context to those phases, right? So to sell a house, you typically need a deal, the home, you need a client, buyer, seller, and you need the information on yourself, right? And so those Lego blocks allow you to piece that context together. So, if you have one client that likes antique um old gothic style furniture and you have another client that likes modern, sleek, new style um interior, what you can do is you can take the client card and say, "Hey, I've got this deal and here's the client card." Decorate this home in their style and it will identify from the client card what the style is. So, if you have that this client um needs a

two uh two door garage, he needs um lots of windows, he needs uh whatever, right? Whatever that client style is, the virtual staging and imaging will move it to that style. Um the same with the emails you write, all that kind of stuff. So, if you have context for, and when I say context, I mean information that is not generally publicly put it in and it will do a much better job personalizing the results to that person. Um, and so other things that is one of the other reasons that we want to have context cards. If you are only using ChatGPT and that's that's all you use and let's say Claude releases a new model that's incredible like the smartest thing ever. You want to be able to switch to that as quickly as

possible. And if you have contact cards you own it. You're not renting it. Whereas if it's just in ChatGPT's memory, ChatGPT has you locked down. If it's just in Claude's memory, Claude has you locked down. But if you have these things external to those entities, you can move your context anywhere. And one of the things that I've really seen with realtors is the speed to execution is one of the simplest ways you guys can grab an advantage.

Right? If you just move a little bit quicker than everybody else to get to the most intelligent model or to get to the best product, you'll be ahead of guys can just maintain a little bit of a speed advantage, it it'll it'll look like it's a much bigger advantage.

3:54 Three kinds of communication

Um, so today we're going to go over communication. I'm going to focus on three types of communication because I think that's kind of the way the future is going. And that is human to human, which I'm not going to talk too much about because I can almost guarantee you that every single person in this room is better at human to human communication than I am. Um, we're going to do human to AI and we're going to do AI to AI. Um, and then we're going to do chains of each because I do believe that this is kind of the way that the future of work is going to be. It's going to be humans and AI agents kind of intermingling to accomplish tasks.

So, so basically I again it's this is going to be I think one of the meta skills that really allows people to move forward with not as much effort as some of the other skills. So, if you can communicate really well, if you can get clarity of your outcomes, if you can convey that to all of the parties involved, you'll be able to move uh in a

4:53 Start with a clear outcome

much more qualitative way. So, as an example, this is what would be considered a messy capture or a fuzzy capture. So, I talked to the seller, I need to follow up, they seem nervous, I probably need a new plan, right? Like, that's that's pretty vague, right? Like that's that's not going to really get an outcome. Whereas if you say, "Hey, the outcome I want is the seller to choose a pricing plan." How are they going to do that? They're going to compare the three pricing scenarios that I've come up with. When are they going to do that?

Before we prep the listing on Friday. What's their motivation to do that? It's going to reduce the days on market risk. What is the constraint? We don't have comps uh that are updated. Who owns this project? I do until I pass it to them. Right? Like if you have all of those things laid out in a humanto human communication structure, you can give this to the AI and it can help you a lot more. But if you leave the human conversation and it's vague like this, you now have to go to the AI, you have to brainstorm, you have to get the specifics. But if you can answer the questions of this, I have tried to get this to look decent for probably two hours. So it it looks like trash. Um, so but the goal is the what the what.

Um, so what is the actual thing that you want? And I say that because a lot of people get focused on process, right? Process is it's one of those things that especially like me being like semi-engineered, it's very easy for me to be like, I love processes. However, the only good process is a process that is supporting an outcome. There are so many processes that are in place that I've like seen in companies or business or whatever that I'm like, hey, what's this process do? Oh, it does this. Okay, so what outcome does it to support?

Nothing. Okay, so we're just doing that process to get information that we don't need. Got it. Perfect. Right. So, we need to have clear outcomes on the what. We need to understand the how. We need to understand when the blockers that are preventing this from happening, what is the motivation for the person to do it? and who's responsible for ownership. So this is again AI is very intelligent. So I have a bunch of formats and shortcuts and things that I use. That doesn't mean that it's going to be the best um formula for you. You can try out my formula, see if it works. If you like it, great. If not, adjust it to how your communication style is. Everything with AI, one of the cool things is that it can go in so many different directions.

One of the problems is it can go in so many different directions, right? So, you have to kind of figure out what of the processes that I'm putting out here are work for you and then how can you make them work for you best. So, you may have a CRM that you use like Salesforce, go high level, HubSpot, whatever. Um, that is going to have different tools than people who just use Google Sheets or whatever the case is. So, um, the the big goal is to figure out a format that works for you, um, and so that you can create predictable inputs, which create predictable outputs. The better the input that goes into the AI, the better the output you're going to get out, which is why it's so important that you have clarity of exactly what you're

hoping to do. And if you don't have clarity, you can

8:24 Ask how do you know

someone brings you a piece of information the the the series of questions and this is for AI as well the series of questions is what does this information mean um what do I do or how do you know so what is the evidence that you're promoting so this is one of the biggest things with AI the how do you know part so you want to ask for citations you want to ask for um links evidence scientific papers if that's kind of what you're searching for. So this is how do you know? Because a lot of times people have to understand how like how did they get this information? Right? And then utility. So what? Who cares? What do I do with it? Does this information even matter? Right? And so these are kind of the filters that I look at when I get my

AI output back. what basically what can I do with that information that moves my business forward. Um so see so context helps um but it's not everything. So like the context we built context cards in the last class and we may do that today but the context really helps kind of move things faster. So you'll be able to move all your pieces along faster when you have the context built out. And I will show you kind of how to do that. And then the other thing

9:57 Check work between AI steps

that you want, and this is uh big when we're talking AI to AI communication because this is where it can go very very sideways very very quick is AI to AI because you can have four AI to AI interactions before you ever see the output again. And by the time you get the output back, it could be completely wrong. the AIs could have taken it in a very different direction. Right? So you need to understand like what does a good output look like? What does a bad output look like? And how can I tell the AI to either validate or quality control the output that it's getting, right? So if you're like, I hate the word the phrase dream home. I never want to see AI use the word phrase dream home. You can put

guard rails in there and it'll say in the validation um search this output to see if the any of my restricted phrases are in the output, right? And then before the AI gives you the output, hopefully it searches through and then um it'll flag like instances. So it'll say, "Oh, I found three instances where I use the word dream home. I need to resend this through and create another output that doesn't use that phrase." Right? And so that's AI to AI validation is big. It's it's all about trying to keep the AI as accurate as possible. And then

11:18 Keep a human in the chain

this is what I mean when I say a chain of communication events. So you've got human to AI to AI to human to a tool, right? And so what's going to happen is the human has a conversation with the AI. the AI um has a specifications and context that it's using that it then sends that conversation to here where it's transforming and validating. So transformation means um let's say I have 200 views on a social media post. The transformation of that data could be the AI analyzing the social media post and saying, "Hey, your visual hook wasn't that good. your caption could have been better, X Y and Z. Um, and so it's transforming the input that you're giving it into something that is more useful for you. Whatever that something is, that's the transformation. Um, and

then the human, it's always important that humans are what we call human in the loop, which means the human is reviewing and quality controlling every single um, output that you get because the more that you do this stuff on the front end, the faster you can go. A lot of people see um it a lot of people get progress with AI a little bit backwards because progress with AI is a lot of planning and specifications and front-end work and it's kind of a bit of a turtle tortoise in the hair type of race. If you build your foundation strong, the ability to go as fast as you want on the back end becomes much easier. But if you rush through building the foundation, all the other stuff takes a little bit longer because you're like, "Oh, that's not

quite right. I need to make this edit. That's not quite right." And every time you're saying, "Oh, that's not quite right." Whereas, if you just take an hour, two hours, whatever it is to sit down and really build out context and then iterate it forward as you're using it, you'll have this really nice context that you can pull in from every different part of your business. And so with this, the things that we're trying to control for are each thing understanding exactly what output they're going for. So we're all trying to keep the meaning of the request intact and get to the output. So the meaning of the request is somewhere over here. The output is over here. All this stuff in the middle are the problems that are preventing you from getting there. Right? And so we've got human to

AI to AI to human. And then this is new.

13:54 Connect AI to your tools

And that's where the AI does a tool call. And so when the AI calls a tool, that's where we go from having the AI be in kind of a chat window to being into your systems that are living and breathing that you use, right? So um if you use notion, if you use Gmail, if you use Google calendar, um if you use Google Drive, like HubSpot, Salesforce, go high level, all those things, this can plug into it and add an intelligence layer to those tools. So that as you're operating your business, intelligence is now being applied to um let's just as an example, the CRM. Let's say a lead comes into your CRM with an intelligence layer in there. The intelligence can review the CRM to look for any um things that you already have

uh like contact stuff like that. But that's not uncommon. The CRM, however, case is, right? And so having the intelligence pull context in and build human connection is one of those things that we really should be striving for because like I said so much with AI is uh it kind of blunts human connection or it's like oh I can tell that that was written by an AI or minimally I can tell that that wasn't written by this person who I know right because it used all these words that they would never use it it you know they would never send me a bullet point list whatever the case is Right? And that's why the context is important because it's I want everything that comes out of your computer to look like it's come from you even if it

hasn't. Right? Because your clients while a lot of them will say, "Oh, I want that personal touch." At least from what I've seen, what they mean is, "I want you to get back with me in the way that I understand as quickly as possible." Right? And that that's the human touch. And so if we can make that kind of as easy as possible, that's that's the

16:21 Run pipelines, not prompts

goal. Um, and then this is where we get into more automations and like really chaining things together and building out useful tools. Um, so we're going to run pipelines, not prompts, right? And so we want to think in terms of steps. Um, we want to think in terms of engineering processes to get our outcomes. And so instead of just saying, "Hey AI, build this thing." We want to plan a little bit on the front end and we want to say, "Okay, AI, you need this information. You're going to need these tools. You're going to need to um take the information that I give you and do this with it. You're going to need to do this research." And you just lay out all the steps that the AI is going to have to follow just like you would if you

were onboarding an administrative assistant or anyone to your team, right? Like the better you understand your workflow and the way your work moves, the better you're going to be able to use AI to make that work move through that pipeline for you. Because a lot of people have work pipelines, they've just never thought about actually building them out, right? Like a lot of people have like, oh, when I get a new listing, this is what I do X, Y, and Z, right? But they don't have it written down anywhere. They don't have it where they can give it context. And so this is one of those things where if you build out some documentation for yourself, you'll be able to really use that documentation over and over and over again um to push um tasks down the pipeline. So

17:45 Workflows you can start with

So what we can do is um we're at the point of the movie where um we can just kind of sit down together and build some stuff. Um, these are things that I have built in the past or that have um been useful to some degree. So, there's a listing launch. So, basically, you can build your SOP for your listing. Uh, virtual staging, um, which is how to very easily take an empty room and tailor it exactly to a client's taste. um CMA pricing narrative, um showing feedback and seller updates, post inspection negotiation, um sphere reactivation. So that's if like someone is on your email list and they're just not responsive, stuff like that. How do you get them engaged in case they do want to purchase again? And then um there's also a daily deal desk, which is

like an update um an update automation. So

18:44 A prompt versus a pipeline

my thought um okay so this is um kind of the difference between a pipeline and a prompt. So a prompt is write me a listing description and the output depends on what you remember to uh paste what the AI guesses and whether it catches any mistakes and after it's what's the trigger what's the source what's the context filter the AI production the review gate and the final deliverable right and so that goes from we add the seller email the listing facts the home context the launch assets the compliance um facts and review and then it makes it publishable ready and so it'll create the actual ual assets a lot of different things with that AI allows you to create that you may not think you have the skill set to actually create, right? And so, you may not be

like, I'm not a web developer. I have no idea about, you know, building a landing so if we want to figure out kind of what's the most useful thing that we can build either for each of you for one thing, but I'd actually like to build something that you guys can deploy in your business. And I've got a couple um samples like a CMA listing thing, uh, a social media provider, stuff like that. So if you guys want to build something, we can do that. Um but yeah, I mean honestly that is kind of the structured presentation and then we can we can either build um