Craft, elevated by intelligence.
This is a new 12-month arc that builds on our first three sessions. Month 1 compresses everything we have covered so far into one durable framework and launches the SOP sprint; every month after adds one advanced capability and one shipped artifact. Nobody watches; everybody builds. Laptops open, Sustainable 9 Design System loaded, real projects on screen.
Two people share one AI win and one flop since last session. Rotating order, no repeats. Flops teach more than wins.
Chad teaches one new capability with a live demo on a real S9 project. Twenty minutes, hard stop.
Everyone runs the same lab prompts on their own laptop, on their own project. Chad and fast finishers float and coach.
Break into role groups. Each group adapts the concept to their work using the role prompts below.
One artifact leaves the room finished: an SOP, a template, a saved prompt, a Project. Homework assigned.
Cohort: Josh, Nick & Danny (project management) · Brittany, Liz, Maggie & Amanda (interiors) · Danielle (procurement) · Nataliya (marketing & events) · Colin (warranty & service) · Paul (architecture & engineering) · Vince (construction, estimating & budgets) · Ryan (finance) · Chad (facilitator).
These eight rules apply to every tool, every role, every month. They are the quality checks in every SOP we write this year.
Nothing AI-drafted goes to a client, lender, inspector, or vendor without a human read. You own everything you send.
Prices, allowances, and budget figures shown to clients must come from a verified budget document. If Ryan or Vince has not verified it, it is a placeholder, and it stays internal.
Client financials, contracts, and personal details go only into our approved company accounts, never personal accounts or unapproved tools.
Purposeful, innovative, optimistic, collaborative. We are design+build artisans, never contractors. Aspirational language: achieve, artisanal, craft, elevate.
AI summarizes and drafts; it does not become the record. Schedules, budgets, and selections live in their systems, and AI output points back to them.
Research answers must cite where they came from. Code, product specs, and lead times get checked against the source before anyone acts on them.
The first draft is raw material. Critique it, tighten it, make it ours. If it reads like a robot wrote it, it is not done.
A prompt that saves you an hour saves the team twelve. Winning prompts go into the shared library the same week.
Quarter 1 locks the foundation, drafts the SOPs, and stands up every role's workspace and client visuals. Quarter 2 builds our document, meeting, and workflow systems. Quarter 3 goes deep on research, visuals, and numbers. Quarter 4 hardens quality, turns SOPs into working tools, and everyone ships a complete role system.
The six-part framework with worked examples, the Context Block saved into Claude, and two SOPs shipped.
Everyone 02Claude Projects per role, full custom instructions, and the shared prompt library launch.
Everyone 03Timeline and status visuals that make our five-phase journey legible to clients.
PMs · Interiors · Nataliya 04Plan sets, spec sheets, selections, and budgets: extract, compare, reconcile.
PMs · Danielle · Vince 05Fathom transcripts into action items, decision logs, and client recaps. First team demos.
Everyone 06Gmail, Drive, and Calendar connected. Checklists become repeatable workflows.
Everyone 07Vendors, materials, products, and code questions with sources and comparison matrices.
Danielle · Paul · Vince 08Advanced Claude Design: mood boards, presentations, event and marketing assets.
Interiors · Nataliya 09Estimating support, variance narratives, draws, and reporting, with verification discipline.
Vince · Ryan · PMs 10Hunting hallucinations, red-teaming drafts, and the S9 AI QA checklist.
Colin · Everyone 11Trackers, intake forms, and calculators: everyone builds a working tool for their role.
Everyone 12Every artisan demos a complete role system. We set the Year 2 roadmap together.
EveryoneEach plan is ready to run: agenda, lab prompts, role prompts, the artifact that ships, and homework. Tap a month to open it. Prompts are copy-ready; replace anything in [BRACKETS].
PrepHave ready: one real weekly-update email plus the raw notes behind it, your own Context Block drafted, and the Website Lead Intake SOP loaded on screen. Test the exact click path on your own account so the settings walk-through is smooth: initials (lower left) → Settings → General → Instructions for Claude. Know the older label too (a Profile field asking what preferences Claude should consider) in case someone's account shows it.
Talk trackOpen with the frame: this is the reset session; after today, fundamentals are assumed. Teach the six parts by contrast, not lecture: show the weak example, ask the room what is missing, then reveal the strong one. The storage moment is the day's biggest unlock: save your own block into Instructions for Claude live (Settings → General), narrate the three layers, and say explicitly that pasting context is over. For the SOP half, walk the lead-intake example top to bottom and linger on Quality Checks and Handoffs: those are the two sections every homemade SOP forgets. Demo the Expert Interview for only 3 questions, then cut to the lab before the room gets passive.
Watch forContext blocks ballooning into essays: cap Layer 1 and 2 at 150 words each. Interview rabbit holes: enforce the 14-question stop. Fast finishers go to the Gap Hunter in pairs. If anyone protests the pace, remind them the site holds everything for later.
Land itClose on the count: two SOPs exist tonight that did not exist this morning, and nobody pastes their context ever again. Backlogs are assigned. Brews.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | Kickoff round: everyone shares the single most useful thing AI did for them in the last 90 days, one sentence each. |
| 0:10–0:35 | Concept 1 | The S9 Prompt Framework, six parts with weak-vs-strong examples (below), plus where the Context Block actually lives in Claude: profile, Project, or paste. |
| 0:35–1:00 | Live Lab A | Build the Context Block and save it into Claude (Lab A1), rebuild your most-used prompt (Lab A2), run the self-critique loop (Lab A3). |
| 1:00–1:15 | Concept 2 | The S9 SOP format, taught by walking the finished example below: Website Lead Intake & First Response, top to bottom. |
| 1:15–1:45 | Live Lab B | The Expert Interview (Lab B1) on a task that lives only in your head, then the Checklist Extractor (Lab B2). Fast finishers: the Gap Hunter drill in pairs. |
| 1:45–2:00 | Ship It | Both SOPs filed in standard format. Quarter 1 SOP backlogs assigned per role. Homework set. |
Six parts. You will not use all six every time, but when an output disappoints, the fix is almost always one of these. Each example below is real S9 work.
Weak: "Write an email to my client."
Strong: "You are a senior project manager at a luxury design+build firm, writing to a detail-oriented client who reads every word and worries when updates are vague."
Role sets vocabulary, judgment, and what the model treats as obvious. A "senior PM" explains schedule logic; a generic assistant pads with pleasantries.
Weak: "The project is going fine."
Strong: "The Meyer project is in the Deliver phase, week 3 of framing. Windows arrive June 12. Last week we told them the fireplace surround decision could wait; it now cannot."
Context is the part you get tired of retyping. That is exactly why we save it into Claude itself today instead of pasting it forever (see below).
Weak: "Help me with the schedule situation."
Strong: "Draft a 6-line email asking the Meyers to confirm their fireplace surround selection by Friday, June 6, so we hold the framing sequence and the June 12 window install."
Verb + object + audience + why-it-matters. If you cannot name the verb, the model cannot either.
Weak: "Summarize this meeting."
Strong: "Output a table with columns Owner | Action | Due, then a 3-bullet summary I can paste into Slack, then one sentence flagging anything ambiguous."
Tables, word counts, reading level, section order, subject lines. Format is free to specify and expensive to fix afterward.
Weak: (none, and the draft invents a move-in date)
Strong: "No budget figures. Any date I did not give you is 'targeting.' Do not promise anything about the pool. 8th-grade reading level for technical items."
Constraints are where our Ground Rules live inside every prompt: no unverified numbers, no invented commitments, client-safe tone.
Weak: Accept draft one, lightly edit, send.
Strong: "Tighten to 120 words." → "Warmer opener; less salesy." → "Score it 1–5 against my five criteria, then revise anything under 4."
Direct the revision like a designer: one variable at a time. Regenerating from scratch is a coin flip; iteration is a craft.
Step 1: draft the block with the prompt below. Step 2: click your initials (lower left) → Settings → General → paste the "everywhere" layer into the "Instructions for Claude" box (older accounts: the Profile field asking what preferences Claude should consider) and save. Step 3: create a Project named "[Your Name] · [Role]" and paste the role layer into its instructions as a stub (Month 2 builds it out properly). Step 4: open a brand-new chat and ask for a quick draft; confirm it already knows you.
Help me write my S9 Context Block, then split it into the two places it will live inside Claude. About the company (keep this wording): Sustainable 9 Design+Build is a Twin Cities luxury custom home builder focused on green materials and artisanal design across Design, Build, Interiors, and Real Estate. Tone: purposeful, innovative, optimistic, collaborative. We are "design+build artisans," never "contractors." Aspirational language where fitting: achieve, artisanal, craft, elevate. Never state prices, allowances, or budget figures in client-facing text unless I paste them from a verified budget document; otherwise use [BUDGET REF] placeholders. About me: - Name: [NAME]. Role: [ROLE + one line on what you own]. - Current projects: [PROJECT NAMES + phase]. - My recurring outputs: [e.g., weekly client updates, selections schedules, draw requests, warranty responses]. - Quirks worth knowing: [e.g., "I write short and direct," "I present to clients on Tuesdays"]. Produce: 1. LAYER 1 — INSTRUCTIONS FOR CLAUDE (under 150 words): the always-true parts, for the account-wide instructions box (Settings → General) — identity, voice, standards, hard rules. Written as instructions to Claude. 2. LAYER 2 — PROJECT INSTRUCTIONS STUB (under 150 words): my role layer — recurring outputs with default formats, current projects, who I hand work to. 3. LAYER 3 — PORTABLE PASTE VERSION: the full block as one text file for use outside Claude. Keep every layer tight. If I gave you fluff, cut it.
Take the prompt you run most often (for most of you: the weekly client email) and rebuild it with all six parts. Note there is no pasted context block anymore; your profile and Project now carry it.
Draft this week's client update for [PROJECT / CLIENT NAMES]. Raw notes from the field (synthesize, do not quote): [PASTE BULLET NOTES: what happened, what's next, deliveries, inspections, decisions needed, photos being attached] Format: - Subject line + 5–8 sentence email. - Order: warm one-line opener tied to visible progress → what happened this week → what happens next week → dates to hold → ONE clear ask with a deadline, bolded. - Close with our standard sign-off. Constraints: - No prices or budget figures. - No promises on dates I did not give you; soft dates say "targeting." - 8th-grade reading level for anything technical. After the draft, list anything from my notes you left out and why.
Critique your draft against these five criteria before I read it: 1. Would a busy client understand every sentence on the first pass? 2. Does it sound like Sustainable 9 (purposeful, optimistic, artisanal) or like generic AI? 3. Is the single ask unmissable and dated? 4. Did anything creep in that I did not tell you (dates, numbers, promises)? 5. What is the weakest sentence? Score each 1–5, then give me a revised version fixing anything under 4.
Every SOP we write this year uses one format: Purpose · Owner · Trigger · Frequency · Steps · Quality Checks · Handoffs · Tools · Last Revised, plus a companion checklist short enough to actually use. Here is what finished looks like. Walk it top to bottom before Lab B; notice that Quality Checks and Handoffs are where most homemade SOPs fall apart.
Every lead that reaches us through sustainable9.com gets a warm, personal, correctly-routed response within one business day, so no future client's first impression of S9 is silence.
Chad (sales). Backup when traveling: Janelle.
A new submission arrives from the website quiz or contact form.
Per lead; inbox checked at 9am and 2pm every business day.
Qualified custom/remodel → discovery call on Chad's calendar. Interiors-only inquiries → the interior design studio. Land to sell or home search → Harker & Woods. Pre-designed-home fit → the Kvikk conversation. Any press or partnership inquiry → Chad directly, flagged.
Website quiz/contact form · Pipeline · Sales Project in Claude (templates per lane) · Calendar.
[Today's date] · Review quarterly as the website funnel evolves.
Now build one of your own. Pick the task you brought: the one that lives only in your head.
You are a process documentation specialist embedded at Sustainable 9.
Interview me to document a process I currently do from memory:
"[NAME THE TASK, e.g., 'how I run the pre-drywall client walkthrough']"
Rules for the interview:
- ONE question at a time. Wait for my answer.
- Cover, in roughly this order: what triggers the task, what I gather
first, the steps in sequence, tools/systems touched, who I hand things
to and when, what "done" looks like, common failure points, and the
judgment calls (things that vary by project).
- Push back when my answer is vague ("walk me through the last real time
you did this").
- 10–14 questions max, then stop and say "READY TO DRAFT."
When I confirm, produce the SOP in the S9 standard format:
Purpose · Owner · Trigger · Frequency · Steps (numbered, one action each) ·
Quality Checks · Handoffs · Tools · Last Revised [today's date].
Mark anything I was unsure about with ⚠️ so my team can verify it.Using the SOP we just drafted, extract a companion checklist for actual
use in the field or at my desk.
Rules:
- 15 items max. Each item is a checkable action, not a paragraph.
- Group under 3–4 phase headers (e.g., Before / During / After).
- Include the quality checks as checklist items, phrased as verifications
("Client confirmed X in writing").
- Format so it reads clean on a phone screen.
- Add a one-line header block: Project ____ · Date ____ · Completed by ____.Here is a draft SOP written by a teammate: [PASTE THEIR SOP] You are a skeptical new hire doing this task for the first time using ONLY this document. Identify: 1. Every step where you would get stuck or need to ask a question. 2. Missing triggers or handoffs (who tells me to start? who do I tell when done?) 3. Undefined terms a new hire would not know. 4. Steps that assume access to a tool or file the SOP never names. List the gaps as questions for the author. Do not rewrite the SOP.
Weekly client update · Site meeting agenda & minutes · Pre-drywall walkthrough · Sub scheduling & confirmation · Punch list management.
Selections process (per phase) · Design presentation prep · FF&E spec sheet standard · Client revision request handling · Install day run-of-show.
PO lifecycle · Long-lead tracking & escalation · Delivery receiving & damage claims · Vendor onboarding.
Project photography (shoot to publish) · Social calendar · Event run-of-show · Public-safe imagery approval (the gatekeeper SOP).
Warranty intake & triage · 11-month walkthrough · Seasonal maintenance comms · Emergency service response.
Sales-support site assessment · HVAC design handoff · Sustainability consult intake · Landscape/permaculture concept brief.
Estimate build & assumptions log · Sub vetting scorecard · Client budget meeting run-of-show · Complaint intake & resolution.
Draw request package · Monthly close checklist · Lender reporting calendar · Client billing cycle.
PrepBuild a throwaway PM Project before the session so the live build has no dead air. Have a lazy request ready ('update for hendersons') and know what good output looks like.
Talk trackWhiteboard the three-layer memory model first: Instructions for Claude travel everywhere, Project instructions hold the role, chat holds today. Then live-build the Project: instructions from the generator prompt, attach two SOPs, run the lazy-request stress test and let the room see the instructions doing the work. The library launch is a ritual, not an afterthought: everyone files three prompts with the naming convention before they leave the room.
Watch forKnowledge-file dumping: no client contracts or financials in Projects, restate Ground Rule 3. Instructions creeping past 350 words: they become unmaintainable. People skipping the 'use when' line in library entries: that line is what makes the library usable.
Land itYour Project is your workbench now. If you re-explain something to Claude this month, that is a bug: fix the instruction, not the prompt.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | Focus: an SOP moment. Did anyone use their new checklist for real? |
| 0:10–0:30 | The Concept | Projects vs. chats: instructions, knowledge files, and why "attach once, use forever" changes daily work. Live build of a PM Project. |
| 0:30–1:15 | Live Lab | Everyone builds their role Project: instructions, files, first saved prompts. |
| 1:15–1:45 | Role Studio | Groups test-drive each other's Projects with a real task and tune the instructions. |
| 1:45–2:00 | Ship It | Prompt Library launch: everyone files their 3 best prompts with a title and "when to use" line. |
Help me write custom instructions for a Claude Project dedicated to my role. Here is my Context Block: [PASTE] Here are the recurring outputs this Project handles: [LIST 4–6, e.g., weekly updates, walkthrough prep, sub confirmations, punch lists] Here are the files I will attach: [LIST: SOPs, checklists, templates] Draft instructions that: 1. Set role, company voice, and standards (fold in my Context Block). 2. Tell the model which attached file to consult for which task. 3. Set default formats for my top 3 outputs so I never re-specify them. 4. State my constraints as hard rules (no unverified numbers, no auto-send language, client-safe tone). 5. End with: "When a request is ambiguous, ask up to 2 clarifying questions before drafting." Keep it under 350 words so it stays maintainable.
I just set up this Project. Test yourself: here is a deliberately lazy request I might send on a busy day: "[LAZY REQUEST, e.g., 'update for hendersons']" Show me what you would produce. Then tell me which instruction or attached file you relied on for each choice you made. If you had to guess at anything, list what instruction I should add so you never guess there again.
Here is a prompt I use that works well: [PASTE PROMPT] Package it for our shared team library: 1. Title: [Role] · [Task] · v1 2. "Use when" line (one sentence). 3. The prompt, generalized: replace my project specifics with [BRACKETED PLACEHOLDERS] and add a one-line note for each placeholder. 4. One example of a filled-in placeholder set. 5. A "watch out" line: the one way this prompt goes wrong if misused.
One shared "PM Playbook" Project template, then personal copies. Attach: PM SOPs, walkthrough checklists, weekly update format, S9 process phase names.
A studio Project with selections SOPs, spec sheet template, presentation outline, and the brand one-pager. Design work still runs through Claude Design with the S9 Design System.
Procurement Project: PO SOP, receiving checklist, vendor onboarding form, standard PO email templates.
Marketing Project: brand voice examples, tagline bank, photography SOP, event checklist. Public-safe imagery rule written into instructions as a hard stop.
Warranty Project: intake SOP, triage categories, response templates by severity, seasonal maintenance calendar.
One Project with "hats" defined in instructions: sales support, HVAC, sustainability, staircase, landscape. Each hat lists its default output format.
Estimating & budgets Project: assumptions log template, sub scorecard, budget meeting run-of-show. Instructions state: all numbers are working drafts until verified.
Finance Project: close checklist, draw package SOP, reporting calendar. Instructions forbid client PII and account numbers in any chat.
PrepS9 Design System attached in Claude Design; pick one live project with real notes for the status one-pager demo; know the five phase names cold: Discover, Design, Define, Develop, Deliver, with Deliver being construction through keys.
Talk trackSell the why before the how: clients do not churn on delays, they churn on uncertainty; a visual that answers 'where are we and what do you need from me' in 10 seconds is a retention tool. Build the journey infographic live in two rounds and narrate your art direction: one variable per round. Introduce the status-language discipline (On track / Watching / Needs decision, 'targeting' for soft dates) as non-negotiable vocabulary.
Watch forFont and color freelancing: everything starts from the Design System, no exceptions. Numbers creeping onto visuals: budget health is a status word unless Vince or Ryan verified a figure. Over-decoration: run the 10-second test by grabbing whoever walks past the room.
Land itPMs leave with a real visual scheduled into next week's client update. The kit only matters if it ships monthly.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | Focus: Projects in daily use. What instruction did you add? |
| 0:10–0:30 | The Concept | Why clients churn on uncertainty, not delays. Anatomy of a great status visual: where we are, what changed, what is next, what we need from you. Live build of the journey infographic. |
| 0:30–1:15 | Live Lab | Everyone builds the journey infographic, then PMs build live-project status one-pagers while other roles build their variants. |
| 1:15–1:45 | Role Studio | Critique pairs: does a stranger understand the visual in 10 seconds? Revise. |
| 1:45–2:00 | Ship It | Kit filed as templates. PMs schedule their first real status visual into next week's client update. |
Using the Sustainable 9 Design System, create a horizontal client-facing infographic titled "THE SUSTAINABLE 9 JOURNEY" with the Playfair tagline "For You + the Earth." Content — our phases, in order, each with a 6–10 word plain-language description of what the client experiences: 1. Discover — we listen: your vision, your site, your budget scope 2. Design — concepts take shape; you see your home for the first time 3. Define — selections, engineering, and a budget you can trust 4. Develop — permits, final drawings, and the build plan 5. Deliver — your home rises: construction, weekly updates, milestone walkthroughs, orientation, and keys Design direction: - One elegant horizontal spine with numbered phase markers; Terra accent for the current-phase marker, Driftwood for completed, Quartz for upcoming. - Uppercase Trade Gothic phase names, generous letter-spacing. - A "typical duration" band under each phase using ranges I will provide, shown as "months," never exact dates. - Footer: S9 wordmark + "Design | Build | Interiors | Real Estate." - Format: landscape for email/print, then a stacked vertical variant for mobile. Ask me for the duration ranges before rendering.
Using the Sustainable 9 Design System, create a one-page project status visual for [CLIENT FAMILY NAME] — [PROJECT NAME], dated [MONTH YEAR]. Inputs: - Current phase: [PHASE] (of our 5: Discover, Design, Define, Develop, Deliver) - This month's completed milestones: [3–5 BULLETS] - Next month's focus: [2–4 BULLETS] - Dates to hold: [ANY WALKTHROUGHS/DELIVERIES — mark soft dates "targeting"] - Decisions we need: [ITEM + NEEDED-BY DATE] - Overall status: [On track / Watching / Needs decision] + one-line reason Layout: - Top band: project name, month, and the 5-phase journey spine with a Terra "you are here" marker. - Left column: "This Month" and "Next Month" lists with simple icons. - Right column: "Hold These Dates" and a Terra-bordered "We Need From You" box — this box is the visual hero. - Status chip top-right using our status words, not colors alone. - No budget figures anywhere on this visual. - Warm one-line closer in Playfair italic, e.g., "Your home, thoughtfully crafted." Make it beautiful enough to print and pin to their fridge.
Using the Sustainable 9 Design System, create a celebratory single graphic announcing that [CLIENT NAME]'s project has moved from [OLD PHASE] to [NEW PHASE]. - Hero: the phase name in huge Trade Gothic with a Terra underline bar (echo our "9" mark). - One Playfair line about what this phase means for them personally. - A compact 3-item list: "What happens in this phase." - A small journey spine at the bottom showing phases completed vs. remaining. - Square format for email embed; also produce a 9:16 variant Nataliya can use if the client agrees to a social share.
Build the status one-pager for one live project each. Real notes, real phase. This becomes a monthly ritual: notes in, visual out, attached to the first update of each month.
Build the Selections Roadmap variant: every selections category as a lane (plumbing, tile, lighting, millwork, paint), each with a decision deadline and status chip. The visual that ends "wait, we needed to pick that already?"
Build the Long-Lead Tracker visual: items on order, lead-time bars, "order by" markers for upcoming decisions. Internal first; client-facing version pairs with the selections roadmap.
Build the public journey infographic for sustainable9.com and the CraftBuilt.ai quiz result pages. Same spine, marketing voice, taglines from the brand bank.
Build the Warranty Year timeline: what to expect at 30 days, seasonal touchpoints, the 11-month walkthrough, how to submit a request.
Prototype the internal Draw Timeline visual (phases → draw points) for lender and client education. Numbers stay placeholder until Ryan verifies; this one graduates in Month 9.
PrepSanitize two vendor quotes for the faceoff demo and pre-run the extraction so you know where the model stumbles: a live catch is the best teaching moment of the day.
Talk trackOne mantra: name the columns before you upload. Demo schema-first extraction, then show the VERIFY section catching something real. The two-quote faceoff lands hardest with Vince and Danielle: the differences-only table is the payoff slide. Close the concept block with the rule that extraction is never done until three fields are spot-checked against the source.
Watch forPeople trusting a clean-looking table: make someone verify three fields out loud. Giant PDFs choking: split plan sets into sheets. 'NOT STATED' discipline: the model must never infer a blank field.
Land itEveryone saves one extraction prompt tuned to a document they touch monthly. That prompt pays rent forever.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | Status visual reactions from clients. What landed? |
| 0:10–0:30 | The Concept | Documents as raw material: upload, extract to a defined schema, verify against the source, then use. Live demo on a plan set page. |
| 0:30–1:15 | Live Lab | Everyone runs extraction on the messy document they brought, then the two-quote comparison drill. |
| 1:15–1:45 | Role Studio | Role-specific extraction workflows below. |
| 1:45–2:00 | Ship It | Each person saves one extraction prompt tuned to a document type they touch monthly. |
[ATTACH DOCUMENT] Extract from this [DOCUMENT TYPE, e.g., "vendor quote" / "plan set sheet" / "spec sheet"] into a table with exactly these columns: [DEFINE YOUR SCHEMA, e.g., Item | Manufacturer | Model | Finish | Qty | Lead time | Notes] Rules: - One row per item; do not merge or summarize rows. - If a field is not stated in the document, write "NOT STATED" — never infer. - Preserve exact model numbers and units as written. - After the table, add a section titled "VERIFY" listing: anything ambiguous, any handwriting or low-quality areas you may have misread, and the 3 fields I should spot-check first.
[ATTACH QUOTE A and QUOTE B] Compare these two [bids/quotes/spec sheets] for [SCOPE, e.g., "cabinetry package, Linden Hills"]. 1. Alignment table: line items that appear in both, side by side. 2. DIFFERENCES ONLY table: scope included in one but not the other — this is the table I care about most. 3. Exclusions & allowances: what each explicitly excludes or carries as an allowance. 4. Questions to send each vendor to make this apples-to-apples (draft them as a short email I can adapt). Do not tell me which to choose. Do not total or restate prices in any client-facing language; this analysis is internal.
Below are my raw notes/emails/texts about [TOPIC, e.g., "outstanding selections for the Weber project"]. Turn them into a clean schedule. [PASTE THE MESS] Output a table: Item | Room | Decision owner | Status | Needed by | Blocking what. Then: - Flag conflicts (two different dates or answers for the same item). - List items mentioned but missing a date or owner. - Draft a 4-line summary I can paste into my weekly update, no dates I did not give you.
Plan revision diff: upload rev A and rev B of a sheet, ask for a changes-only list by discipline, then verify on the drawings. Also: turn inspection reports into punch tables with owners.
Spec sheet standardization: extract any manufacturer cut sheet into the S9 FF&E schema. Build the master selections schedule from a messy email thread.
Order acknowledgment vs. PO: attach both, ask for a mismatch-only table (qty, model, finish, ship date). This becomes your receiving pre-check SOP step.
Sub bid leveling: three bids into one leveling sheet with scope gaps highlighted and clarification questions drafted per sub. Award logic stays yours.
Budget export hygiene: upload a cost-report export, ask for anomalies (duplicates, miscodes, blank cost codes) as a review list, not corrections. Verify in the source system.
Paul: extract mechanical schedules from plan sets into equipment lists. Colin: warranty claim photos + description into structured intake records. Nataliya: pull project fact sheets from closeout docs for marketing one-pagers.
PrepPick a juicy transcript for the demo (a real one with a garbled number is gold). Print ballots or set up the vote link. A visible timer for demos: 4:00 hard.
Talk trackTeach 'transcripts are drafts of the truth': names mishear, numbers garble, and 'discussed' is not 'decided'; that distinction is the day's teach. Run the three-artifact ritual fast on one transcript: action table, decision log, client recap: ten minutes total, then hand the room over. The showcase is the point: you emcee, protect the clock, and let the team teach each other.
Watch forShy presenters: pre-assign the order and go first yourself with something imperfect. Demos drifting into slideware: require the live run. The decision-log habit dying after today: assign every active project a log owner before the room empties.
Land itWinners go into the library this week, named and credited. Five months in, the team is teaching itself: say that out loud.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | Extraction time-savings tallied on the whiteboard. |
| 0:10–0:25 | The Concept | The meeting stack: Fathom captures → Claude structures → you verify → systems of record get updated. Live demo on a real transcript. |
| 0:25–1:00 | Live Lab | Everyone processes the transcript they brought through Labs 1–3. |
| 1:00–1:55 | Showcase | 13 demos, 4 minutes each: one workflow, live, on real work. Team votes: most time saved, best client-facing, most creative. |
| 1:55–2:00 | Ship It | Winning workflows earmarked for the shared library. Month 6 preview. |
[PASTE FATHOM TRANSCRIPT or ATTACH] From this meeting transcript for [PROJECT], extract an action table: Owner | Action | Due (only if stated) | Depends on | Source quote (short). Rules: - Only actions someone actually committed to. Ideas and maybes go in a separate "Discussed, not committed" list. - If the owner is unclear, put "UNASSIGNED" — do not guess. - Flag any numbers or dates in the transcript that sound garbled. - Attendee names: [LIST CORRECT SPELLINGS] — correct the transcript's versions to these.
Same transcript. Extract only DECISIONS — choices that close a question. Format each as one log entry: [DATE] · [Decision] · Decided by [who] · Context (one line) · Supersedes (if it reverses an earlier decision, note that explicitly) Then list "Open questions raised but not decided" so nothing silently becomes a decision that wasn't one. I will paste entries into the project decision log after verifying.
[PASTE YOUR CONTEXT BLOCK if outside your Project] Same meeting. Draft the client recap email: - Warm 1-line opener acknowledging their time. - "What we decided together" — 3–6 bullets in plain language. - "What we're doing next" — our commitments with targets. - "What we need from you" — their items, each with a needed-by date, bolded. - Offer to walk through anything. Constraints: no internal discussion, no sub names, no pricing talk unless I paste verified figures, no tentative options presented as decisions. After drafting, list anything from the meeting you deliberately left out.
PrepConfirm everyone's Google connection works before the session (a 10-minute IT check saves 30 live). Make sure your own inbox is demo-safe or use a sandbox account.
Talk trackLead with the discipline before any demo: draft, review, send: Claude never sends, and that sentence goes into every Project's instructions today. Then the wow: morning triage on a real inbox, followed by the day-prep brief pulling calendar and Drive together. Teach tight scoping ('unread client emails, last 24 hours, these projects') as the difference between useful and noise.
Watch forPrivacy flinches: company accounts only, and nothing happens without review. People asking for auto-send: the answer is no, and the reason is Month 10's stakes ladder. Vague scope requests producing garbage: coach the fix live.
Land itEveryone calendars morning triage as a daily habit before leaving. Habits beat demos.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | Meeting ritual reports: what did the decision log catch? |
| 0:10–0:30 | The Concept | Connectors demoed live: morning triage, meeting prep from Calendar + Drive, drafted replies for review. The draft-review-send discipline. |
| 0:30–1:15 | Live Lab | Everyone connects tools and runs Labs 1–3 on their own accounts. |
| 1:15–1:45 | Role Studio | Convert one SOP into a guided workflow prompt (Lab 4 pattern). |
| 1:45–2:00 | Ship It | Each person schedules their morning triage as a daily habit and saves their workflow prompt. |
Review my unread emails from the last 24 hours. 1. Sort into: CLIENT (anything from or about [ACTIVE PROJECT NAMES]) / VENDOR-SUB / INTERNAL / NOISE. 2. For CLIENT and VENDOR-SUB: one-line summary each + what they need from me + suggested urgency (today / this week / FYI). 3. Draft replies for the top 3 that need responses today, one at a time, for my review. Match my usual tone from my sent replies to that person. 4. Do NOT send anything. Do not archive or label anything. End with: anything time-sensitive I might miss if I only skim your summary.
Look at my calendar for today and tomorrow. For each meeting or site visit: - Who is attending and which project it belongs to. - Pull the most recent relevant thread or document from my email/Drive and summarize where things stand in 3 lines. - One suggested goal for the meeting and one question I should be ready to answer. - Flag conflicts, missing locations, or meetings with no agenda. Format as a single brief I can read on my phone in 2 minutes.
Search my Drive for the [DOCUMENT, e.g., "latest selections schedule"] for [PROJECT]. If multiple versions exist, list them by modified date, tell me which appears current and why, and summarize what changed between the last two versions if both are readable. Link, don't duplicate: I want the file, not a recreation of it.
Here is one of my SOPs: [PASTE SOP] Convert it into a guided workflow you can run with me: 1. Walk me through the steps in order, one at a time. 2. At each step, tell me what input you need from me (or which connected file/email to pull), then produce that step's output before moving on. 3. Steps a human must do physically or decide personally: present them as a checkpoint and wait for my "done." 4. At the end, produce the completion summary defined in the SOP's Quality Checks and a list of anything skipped. Save this as a reusable prompt titled "[Role] · [SOP name] · Guided Run."
Guided run: the weekly update SOP end-to-end (triage inbox for project news → pull schedule notes → draft update → draft status visual notes). Sub confirmation drafts from tomorrow's calendar.
Guided run: presentation prep SOP (pull latest selections, assemble decision list, draft agenda + client pre-read email).
Daily PO sweep: unread vendor emails → acknowledgment mismatches → escalation drafts. Long-lead date changes surfaced against the tracker.
Content pipeline: photo delivery emails → shot lists filed → caption drafts queued for approval. Event RSVPs summarized from inbox.
Warranty inbox triage with severity classification per the intake SOP, plus drafted first-response emails using the response templates.
Vince: budget meeting prep brief from calendar + latest cost docs. Ryan: draw-day guided run through the draw package checklist. Paul: consult prep briefs pulling site info from Drive.
PrepPre-run the comparison demo (thermally modified siding works well) so you know the sources. Have a code question ready where Minnesota amendments bite, so the protocol's caution lands as real, not theoretical.
Talk trackOpen with 'a claim without a link is a rumor.' Demo the matrix with sources inline, then the code protocol as a hard rail: AI orients, the adopted code and the AHJ decide, Paul or Vince signs off before anything code-flavored reaches a client. Give greenwashing its own two minutes: certifications beat adjectives, and our clients deserve the receipts.
Watch forAI-quoted prices treated as facts: cost tiers only. Stale info: make 'as of when?' a reflex. Blog-sourced claims: push to manufacturer pages and primary sources.
Land itMatrices filed on real, current decisions: research that changes a choice this month, not a demo that evaporates.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | Triage habit report: best catch of the fortnight. |
| 0:10–0:30 | The Concept | Search vs. deep research; why sources are non-negotiable; the code-question protocol (AI orients, the AHJ and the inspector decide). Live demo: triple-pane window supplier comparison. |
| 0:30–1:15 | Live Lab | Everyone researches the question they brought using the matrix pattern. |
| 1:15–1:45 | Role Studio | Role research patterns below; pairs sanity-check each other's sources. |
| 1:45–2:00 | Ship It | Matrices filed; the code-question protocol added to the S9 AI Standard. |
Research [DECISION, e.g., "thermally modified wood siding options that give a cedar look for a Minnesota lakeside build"]. Requirements to score against: [LIST 4–7: e.g., freeze-thaw durability, maintenance cycle, sustainability certifications, regional availability, installed cost tier (low/med/high only), warranty] Output: 1. A matrix: Option | each requirement | Source link per claim. 2. "As of" date for your information; flag anything that may be stale. 3. Three questions I should ask suppliers that your research cannot answer. 4. What the marketing pages gloss over for each option (be blunt). Cost tiers only — no dollar figures will be shown to clients from this research.
Find [TYPE OF VENDOR/SUB/SUPPLIER, e.g., "custom steel staircase fabricators"] serving the Minneapolis–St. Paul metro. For each candidate: company, location, evident specialty, project examples if public, certifications/licensing signals, and source links. Then: a shortlist of the 3 strongest fits for luxury custom residential work with one-line reasoning, and a draft intro email requesting portfolio and references (our voice: warm, specific, artisan-to-artisan). Note clearly what you could NOT verify (insurance, capacity, current licensing) — that is our vetting work, not yours.
I have a code question: [QUESTION, e.g., "guard height and opening requirements for an interior stair landing in a single-family home in Minnesota"]. 1. Identify which code and sections likely govern (name the code and the currently adopted edition in Minnesota, with source). 2. Summarize the general requirement in plain language, labeled "ORIENTATION ONLY — VERIFY WITH ADOPTED CODE AND AHJ." 3. List local-amendment risks: where Minnesota or the municipality commonly modifies the model code on this topic. 4. Give me the exact questions to ask the building official. Do not present this as a final answer.
Alternate-supplier drills: when a product is discontinued or delayed, run the matrix on substitutes with lead-time evidence. Build the "substitution brief" template the designers approve from.
Deep research on systems: HVAC equipment comparisons (HRV/ERV models by NFRC/AHRI data), permaculture plant palettes by zone, staircase code orientation. Everything sourced, everything verified before client use.
Sub discovery + market context for estimating assumptions (material trend directions, not prices-as-facts). Draft reference-check question sets per trade.
Product research with certification checks; build the "spec defense" habit: every spec sheet claim a client might question gets a source.
Competitive scan: how other luxury design+build firms present process and sustainability; gap list for our site and the CraftBuilt.ai quiz copy.
Colin: manufacturer warranty terms lookups with source links. Ryan: lender/market context reads, never as advice. PMs: means-and-methods research for unusual details, verified with Vince.
PrepSave a deliberately mediocre mood board beforehand to improve live in three rounds. Clear wall space or a shared screen rotation for the gallery walk.
Talk trackThe teach is art direction: 'make it better' is a coin flip, 'increase whitespace, drop the third accent, headline two sizes up' is a craft. Run your three-round improvement live, narrating the one variable each round. Frame consistency as the luxury signal: same phases, same chips, same spine everywhere a client looks. The rationale rail on mood boards is the client-value move: connect the palette to how they said they want to live.
Watch forRegenerate-roulette instead of direction. Skipping rationale because the visuals look good. Anyone freelancing off the Design System: redirect without ceremony.
Land itReal deliverables exported and scheduled for actual client moments, not pretty files in a folder.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | Research outcomes: what decision moved faster? |
| 0:10–0:30 | The Concept | Iterating visuals like a designer: art-direct in rounds, change one variable at a time, keep a rejects file. Live: a mood board goes from generic to S9 in three rounds. |
| 0:30–1:15 | Live Lab | Everyone builds their real deliverable through at least three critique rounds. |
| 1:15–1:45 | Role Studio | Gallery walk: work on screens, sticky-note critique, one more round. |
| 1:45–2:00 | Ship It | Deliverables exported and scheduled for real use. Best templates filed. |
Using the Sustainable 9 Design System, create a client mood board for [PROJECT / ROOM, e.g., "Weber primary suite"]. Design intent (from my notes): [PASTE: adjectives, materials, palette direction, 2–3 anchor pieces already selected] Layout: - Title band: project name + a Playfair intent phrase (draft 3 options for me to pick from). - 6–9 swatch/inspiration tiles in a broken grid with generous whitespace, single hairline borders per our photography style. - A "Why this works" rail: 3 short rationale lines connecting the palette to how the clients said they want to live (I'll paste their words: [CLIENT QUOTES]). - Materials strip along the bottom: named finishes, no pricing. Round 1 is composition only — placeholder tiles are fine. I will art-direct rounds 2 and 3.
Using the Sustainable 9 Design System, build a reusable client design presentation shell (10 pages): 1. Cover: project name, family name, date, Playfair tagline. 2. "Where we are" journey spine (our 5 phases, Terra marker). 3. Design story: one page, big statement, one hero image slot. 4–6. Concept pages: image-dominant, short captions, option A/B framing. 7. Selections summary table page (Item | Room | Status | Decision needed). 8. "Decisions today" page — the meeting's purpose, unmissable. 9. Next steps + dates to hold. 10. Closing page with our wordmark and service lines. Every page: uppercase Trade Gothic headers with wide tracking, image borders per brand, no clutter. Build it with placeholder content clearly marked [PLACEHOLDER] so any designer can duplicate and fill it.
We are telling one story: [STORY, e.g., "the Linden Hills project reaching
Deliver phase" or "our annual client appreciation event"].
From this story, create a coordinated set using the Sustainable 9 Design
System:
1. Instagram square announcement.
2. 9:16 story version with a bolder crop.
3. Email header banner.
4. One-page print piece (event) or blog header (project story).
Shared elements: same headline treatment, same image, same tagline from
our bank ("[PICK: For You + the Earth / Environmentally Cozy / High Impact
Design, Low Impact Development]"), Terra used exactly once per piece.
Deliver as a set so the campaign reads as one voice across channels.Each designer ships one real deliverable: mood board, presentation, or install-day reveal graphic. Amanda builds the spec-sheet visual template; Brittany/Liz/Maggie pressure-test the presentation shell on live projects.
Ship the next real campaign set + the event kit template (invite, signage, name cards, follow-up email header) from one prompt chain.
Milestone celebration graphics (foundation poured, roof on, drywall complete) as a reusable set; pairs with the phase-gate graphic from Month 3.
Paul: a sustainability-features diagram template (how your home works: HRV, envelope, solar-ready). Vince: budget meeting cover page + agenda visual. Colin: seasonal maintenance mailer. Ryan: internal report cover standard.
PrepCo-teach: brief Vince (assumption challenger) and Ryan (variance narrative, draw memo) on their segments beforehand. Sanitized exports ready; strip anything with account numbers.
Talk trackOpen with the sentence of the day: AI drafts narrative about numbers; systems compute numbers. Then the watermark ritual: everything numeric is DRAFT — UNVERIFIED until Vince or Ryan signs, and only then can a figure face a client. Let Vince run the assumption challenger on a live estimate: watching it surface silent assumptions is the session's wow. Ryan's two-voice variance narratives (internal blunt, client calm) show the same facts wearing two correct outfits.
Watch forThe model recomputing totals: forbid it in the prompt and in the culture. PII sneaking into prompts: loan numbers and SSNs never, anywhere. Anomaly lists being treated as corrections: the human fixes the source system.
Land itThe Numbers Rules card gets pasted into every Project's instructions before anyone leaves. That card is the month.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | Template adoption check. Which visual is now a habit? |
| 0:10–0:30 | The Concept | Where AI helps with money (structure, narrative, anomaly-spotting, assumption-challenging) and where it must not (being the calculator of record, inventing figures). Vince and Ryan co-teach this one with Chad. |
| 0:30–1:15 | Live Lab | Labs 1–3 on sanitized real data. |
| 1:15–1:45 | Role Studio | Role tracks below; PMs pair with Vince on change-order narratives. |
| 1:45–2:00 | Ship It | The Numbers Rules card finalized and added to every Project's instructions. |
[ATTACH SANITIZED ESTIMATE or PASTE ASSUMPTIONS LOG] You are a skeptical senior estimator reviewing this estimate for a luxury custom home in the Twin Cities. Do not recalculate anything. Instead: 1. List the assumptions this estimate silently makes (site conditions, allowances, escalation, schedule, scope boundaries). 2. Rank the 5 assumptions most likely to move the final number, with the scenario in which each one bites us. 3. Scope-gap scan: what is commonly missed on projects like this ([DESCRIBE: e.g., walkout lot, pool-ready, high glass area]) that I should confirm is included or explicitly excluded? 4. Draft the "Assumptions & Exclusions" section for the client budget document in plain, confident, non-defensive S9 language. Label all output DRAFT — UNVERIFIED.
[ATTACH SANITIZED COST REPORT: budget vs. actual by cost code] Using ONLY the figures in this export (do not compute new totals; if a total seems needed, tell me to add it to the export): 1. Identify the 5 largest variances by dollar and by percent. 2. For each, draft a one-paragraph narrative in two versions: (a) internal — direct, cause-focused; (b) client-facing — clear, calm, no jargon, framed around choices made and value delivered, per our voice. 3. Flag any lines that look like data problems rather than real variances (duplicates, sign errors, miscodes) in a separate review list. Label everything DRAFT — UNVERIFIED. I will verify against the accounting system before anything is shared.
Draft the cover memo for draw #[N] on [PROJECT] to [LENDER NAME]. Inputs (verified by me before this prompt): - Draw amount and period: [PASTE] - Work completed this period: [PM'S BULLETS] - % complete by major division: [PASTE] - Inspections passed: [LIST] - Next period preview: [BULLETS] Format: professional lender memo, one page. Confident, factual, zero fluff. Include a completeness checklist footer of attached exhibits (invoices, lien waivers, inspection reports, photos) I can check off. Flag anything in my inputs that a lender reviewer would question so I can pre-empt the call.
Assumption challenger on a live estimate; sub bid leveling refresher from Month 4; budget meeting narrative: turn the verified budget into the meeting story arc (what changed, why, choices ahead).
Variance narrative + draw memo on real (sanitized) packages; monthly close checklist as a guided workflow (Month 6 pattern); lender reporting calendar reminders drafted.
Change-order client narratives: cause, options presented, decision, schedule effect, in S9 voice, figures pasted only after Vince verifies. The infographic from Month 3 gets a "budget health" status word, never a number.
Interiors: allowance-tracking language for selections over/under allowance (status words client-facing, figures internal). Danielle: PO totals reconciliation review lists. Colin: warranty cost logging categories. Nataliya: zero client financial info in any marketing material, restated as a hard rule.
PrepBuild the planted-error document (8 errors: 2 invented facts, 2 wrong numbers, 1 misattributed decision, 1 stale claim, 1 misspelled name, 1 over-promise) and keep the answer key. Collect two 'almost sent it' stories in advance so the confession round starts warm.
Talk trackTeach the failure taxonomy with your own real flops: credibility comes from confession. Draw the stakes ladder on the whiteboard and name never-AI territory out loud: hiring and firing language, legal opinions, final warranty coverage calls. The hallucination hunt runs as a competition: points per catch, minus per false accusation: the minus is the teach. Colin's masterclass closes it: acknowledge first, never diagnose remotely, never promise coverage in writing.
Watch forThe cynicism swing ('so AI is useless?'): reframe: the QA pass is the price of moving fast, and it is cheap. Teams over-accusing in the hunt: precision matters as much as recall. QA steps written vaguely: each role's additions must be checkable.
Land itRatify the QA checklist v1 and embed it into every SOP's Quality Checks. Announce the capstone spec and tease Month 11: bring a tool idea.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | The "almost sent it" confessions. Best catch wins coffee. |
| 0:10–0:30 | The Concept | Failure taxonomy with real examples: hallucination, garbled numbers, misheard names, confident staleness, tone drift, over-promising. The stakes ladder: internal note → client email → contract-adjacent → never-AI territory. |
| 0:30–1:15 | Live Lab | Hallucination hunt on planted-error documents; red-team drill on each person's own saved output. |
| 1:15–1:45 | Role Studio | Colin's warranty masterclass + each role writes their QA additions. |
| 1:45–2:00 | Ship It | The S9 AI QA Checklist v1 ratified; SOPs updated with QA steps. |
Facilitator setup (Chad, before session): take a real (sanitized) project
summary and plant 8 errors: 2 invented facts, 2 subtly wrong numbers,
1 misattributed decision, 1 stale product claim, 1 wrong name spelling,
1 over-promise ("we guarantee...").
Team prompt during session:
"Here is a project summary: [PASTE]. Cross-examine it against the source
documents attached. Produce a claims audit: every factual claim | supported
by source? (yes/no/partial) | evidence line. Then list the claims you
CANNOT verify from the sources at all."
Score: teams get points per planted error caught, minus points per false
accusation. Debrief: which error types survived longest?Here is a draft I am about to send: [PASTE] Red-team it before it leaves: 1. CLAIMS: list every verifiable claim (dates, numbers, names, commitments, product facts). Mark each: source I gave you / assumed / unknown. 2. PROMISES: list every explicit or implied commitment. Which would we regret in a dispute? 3. TONE: anywhere it drifts from purposeful-optimistic-collaborative into defensive, vague, or robotic? 4. READER TEST: as [THE RECIPIENT, e.g., "an anxious client mid-build"], what is the worst reasonable misreading of this message? 5. Give me the revision that fixes 1–4, with changes listed.
[CONTEXT BLOCK + warranty intake SOP in Project] A homeowner reports: [PASTE REQUEST, e.g., "hairline drywall cracks above two doorways, second winter in the home; they are frustrated and mention 'shoddy work'"]. Draft my response: - Acknowledge the frustration genuinely in the first two lines; zero defensiveness. - Explain the likely cause in homeowner language (seasonal movement, normal material behavior) WITHOUT diagnosing remotely or promising a coverage determination. - Lay out exactly what happens next: inspection scheduled, what we will look at, when they will hear from us. - Close warm and confident, our voice. Constraints: no coverage decision, no cause stated as fact, no timeline promises beyond the inspection. Then red-team your own draft with the 5-point QA pass.
Each role group writes 2–3 role-specific QA checks and adds them to their SOPs' Quality Checks sections (e.g., PMs: "no date leaves without a schedule check"; Interiors: "every product claim has a spec sheet"; Ryan: "every figure ties to the export").
1. Every fact traced or flagged. 2. Every number verified or watermarked DRAFT. 3. Names and dates spot-checked. 4. Promises intentional. 5. Tone is ours. 6. Stakes-ladder level identified, review matched to it. 7. Would I own this sentence-by-sentence in front of the client?
PrepBuild one tool yourself beforehand (the selections tracker is a crowd-pleaser) so the 10-minute live build is a re-run, not a gamble. Print the role tool menu.
Talk trackFrame: documents tell, tools do. Teach the plain-language spec (Purpose · Users · Inputs · Rules · Screens) and insist on spec-before-build: it is the difference between minutes and hours. Do the live build, then iterate twice on audience suggestions: one change per round. Be equally loud about the boundary: prototypes are workbenches and client-education aids, not systems of record; real project data lives in real systems, and a tool that starts accumulating it graduates into a portal conversation with you.
Watch forScope monsters: v1 does one job. Pixel-polishing before the logic works. Client-facing tools skipping the QA pass: the stakes ladder does not care that it is interactive.
Land it60-second tool tours around the room, phones in hand. The best three go team-wide. Capstone reminder: two weeks of real use decides if the tool earns a demo slot.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:10 | Wins & Fails | Red-pen month report: the best catch the QA checklist made in the wild. |
| 0:10–0:30 | The Concept | Documents tell; tools do. The plain-language app spec (Purpose · Users · Inputs · Rules · Screens), then a live 10-minute build: a selections-deadline tracker from nothing. |
| 0:30–1:15 | Live Lab | Everyone turns one of their SOPs or checklists into a working tool (Labs 1–2), then runs the 60-second usability test (Lab 3). |
| 1:15–1:45 | Role Studio | Build from the role menu below; pairs swap tools and try to break them. |
| 1:45–2:00 | Ship It | 60-second tool tours around the room. Specs filed; keepers shared with the team. |
Here is one of my SOPs or checklists: [PASTE]
Help me turn the repetitive part of this into a small interactive tool.
1. Propose the ONE job the tool should do (resist scope; a v1 does one
thing well).
2. Draft the plain-language spec:
- Purpose: one sentence.
- Users: who touches it and in what situation (desk? phone on site?).
- Inputs: every field, with type (text/date/choice/number) and which
are required.
- Rules: what the tool computes, sorts, flags, or hides — including
what happens with missing or weird inputs.
- Screens: what I see first, and what changes as I use it.
3. Ask me up to 3 clarifying questions where my SOP leaves the spec
ambiguous, then finalize the spec.Build the tool from this spec as an interactive artifact I can use in my browser: [PASTE FINAL SPEC] Requirements: - S9 look: clean, uncluttered, our palette (Onyx, Forest, Driftwood, Quartz, Terra accents used sparingly), generous whitespace, uppercase tracked headers. - Phone-friendly: I may use this standing on a job site. - Obvious empty state: when nothing is entered yet, tell me what to do first. - A visible note in the footer: "Working tool — not a system of record." Build v1 now. I will iterate one change at a time after I try it.
Before I share this tool, audit it: 1. What will a first-time user try to do that the tool does not handle? 2. Which input will people most often enter wrong, and what happens when they do? 3. What is unclear in the first 10 seconds of looking at it? 4. If this tool quietly gave a wrong answer, where would it come from? Fix the top issue from each question, then summarize what changed. (Then the human test: hand it to a teammate with zero explanation and watch them for 60 seconds. Every hesitation is a finding.)
Pre-drywall walkthrough capture form (room, item, photo note, change flag) · Punch list tracker with owner and status · Weekly-update note collector that drafts the Thursday email from the week's entries.
Selections tracker: category lanes, decide-by dates, status chips, overdue-to-top sorting · Presentation decision-list builder · Install-day run-of-show timer.
Long-lead watchboard: item, order-by date, ship date, buffer countdown, escalation flag · Receiving day checklist with damage photo log.
Shot list runner for photography days · Content calendar board with consent-status gate: nothing marked ready without the PUBLIC-SAFE check.
Warranty intake form matching the triage SOP: severity auto-suggested by keywords, but the human confirms · Seasonal maintenance reminder builder.
Paul: site assessment scorecard (capture list → one-page Site Read draft). Vince: bid leveling input sheet with scope-gap flags, and an assumptions-log tool. Ryan: draw package completeness checker: every exhibit checked before submit.
PrepTally the year-in-numbers from the monthly logs (SOPs shipped, library size, hours saved). Ballots, timer, the Operating Standard draft printed, and the food and brews actually organized: this one is a celebration.
Talk trackYou emcee, you do not teach. Protect the 5:00 demo timer ruthlessly and require one live run per demo: no slideware. Call out specific before/afters yourself for the quiet ones. Walk the Operating Standard section by section, invite objections, resolve or park them, then adopt it and publish to the wiki. Run the roadmap brainstorm with the facilitation prompt on the big screen and dot-vote to five initiatives with named champions.
Watch forDemos overrunning: the timer is the kindness. The Standard sailing through without real assent: silence is not agreement, ask the skeptics directly. Roadmap wishlists with no owners: no champion, no initiative.
Land itChampions named, Year 2 dated, glasses raised. Bots & Brews is no longer a program; it is how Sustainable 9 works.
| Time | Block | What Happens |
|---|---|---|
| 0:00–0:05 | Kickoff | The year in numbers: SOPs shipped, prompts in the library, hours saved (tallied from the monthly logs). |
| 0:05–1:20 | Capstone Demos | 13 demos, 5 minutes each: the system, one live run, the before/after. Ballot categories: Most Transformed Role · Best Client Experience · Best SOP Set · The Artisan Award (best overall). |
| 1:20–1:40 | Ratify the Standard | Walkthrough of the S9 AI Operating Standard doc; objections handled; adopted into the CraftBuilt.ai wiki as policy. |
| 1:40–2:00 | Year 2 Roadmap | Structured brainstorm (prompt below), dot-vote the top 5 initiatives, assign champions. Then we celebrate — brews earned. |
You are facilitating Sustainable 9's AI roadmap session. Here is what we built this year: [PASTE: the year-in-numbers summary + capstone list] Here is everyone's raw wishlist from today: [PASTE STICKY-NOTE IDEAS] 1. Cluster the ideas into themes (automation depth, client experience, CraftBuilt.ai integration, new capabilities, training gaps). 2. For each theme, propose 1–2 concrete Year 2 initiatives with: what it is, who benefits, effort (S/M/L), and the risk if we skip it. 3. Sequence a suggested Q1–Q4 roadmap assuming one 2-hour session per month continues. 4. Tell us the one thing we built this year that we are underusing.
These are v0.1 starter drafts written to be wrong in useful ways: each role refines their own in Months 1–2 using the Expert Interview, then builds out their full backlog. Every SOP uses the S9 standard format. The universal refine prompt is below; run it on your starter first.
[PASTE YOUR CONTEXT BLOCK] Below is a starter SOP drafted for my role. It was written by someone who does not do my job. Interview me to fix it: - Ask me, one question at a time, where it is wrong, incomplete, or in the wrong order (max 8 questions). - Pay special attention to triggers, handoffs, and the judgment calls the draft glosses over. Then produce v0.2 in the S9 standard format (Purpose · Owner · Trigger · Frequency · Steps · Quality Checks · Handoffs · Tools · Last Revised) plus the companion checklist, phone-readable, 15 items max. STARTER SOP: [PASTE THE STARTER FROM THE LIBRARY]
Every active client hears from their PM weekly with progress, plan, and one clear ask, so uncertainty never accumulates.
Assigned PM · Every Thursday by 3pm · Weekly per active project.
Decisions affecting selections → tag the assigned designer. Budget-touching items → Vince before mention. Anything warrantable post-closing → Colin.
The client sees and approves everything inside their walls before it disappears: outlets, blocking, low-voltage, plumbing, HVAC runs.
Assigned PM · Rough inspections passed, before insulation is scheduled · Once per project.
Price-affecting changes → Vince. Selection-affecting changes → designer. Insulation release → field schedule.
Every selection is made, documented, and communicated before it can block procurement or the field, so beauty never causes a delay.
Assigned designer · Phase entry (Define) and rolling · Continuous per project.
Confirmed specs → Danielle. Field impacts → PM. Allowance variances → Vince before client discussion.
Every client presentation is decision-ready: curated options, clear rationale, and an unmissable "decisions today" list.
Presenting designer · 1 week before any design meeting · Per meeting.
Decisions → selections SOP. Pricing questions raised → Vince. Photography-worthy boards → Nataliya.
Every ordered item is tracked from spec to delivery with no surprises: acknowledged, scheduled, received, and reconciled.
Danielle · Confirmed spec sheet received from design · Per order.
Slips/subs → designer + PM. Invoice discrepancies → Ryan. Site delivery timing → PM.
Every completed project becomes a documented, rights-cleared, client-approved story asset, and nothing unapproved ever goes public.
Nataliya · Project reaches Deliver phase (or milestone worth shooting) · Per project + milestones.
Staging → designers. Access + timing → PM. Award submissions and press → Chad.
Every homeowner request gets a fast, warm, precise response and lands in the right lane: emergency, standard, seasonal, or education.
Colin · Any homeowner request received · Continuous.
Sub-caused patterns → Vince (scorecard input). Design/system issues → Paul. Warm stories worth telling → Nataliya (with consent).
Every prospective site gets a consistent, expert first-pass read that supports the sales conversation and flags cost drivers early.
Paul · Sales requests assessment on a prospect's lot · Per prospect site.
Estimating implications → Vince. Survey/soils/septic studies → PM once engaged. Lot marketing angle → Harker & Woods via Chad.
Every sub on an S9 project has been scored the same way, and the bench stays deep in every trade. Award decisions stay human.
Vince · New sub candidate or annual re-review · Per candidate + annual.
Trial-scope feedback → PMs. Warranty patterns → Colin's monthly report. Bench gaps by trade → recruiting research (Month 7 pattern).
Every draw is complete, defensible, and on time: the lender never has to ask twice, and cash flow never waits on paperwork.
Ryan (with PM inputs) · Draw milestone reached per the loan's schedule · Per draw per project.
Milestone confirmations → PMs. Percent-complete verification → Vince. Payment releases → bookkeeper per Ryan's approval.
The standing set for communicating timelines and status. All run in Claude Design with the Sustainable 9 Design System attached. Status words over colors, "targeting" for soft dates, and no figures unless verified. The three core prompts live in Month 3; these extend the kit.
Using the Sustainable 9 Design System, create the Selections Roadmap for
[CLIENT] — a horizontal lane chart of selection categories:
[Windows, Cabinetry, Appliances, Plumbing Fixtures, Tile & Stone,
Lighting, Millwork, Paint, Hardware, Flooring — edit to project]
Each lane: decision deadline ("decide by [MONTH WEEK]"), status chip
(Decided / In review / Needs decision), and a small lead-time bar showing
why the deadline exists. Terra highlights only the lanes needing decisions
in the next 3 weeks. Header: "Your Selections Journey" + Playfair line
"Every detail, thoughtfully chosen." Mobile vertical variant too.Using the Sustainable 9 Design System, create a 90-day look-ahead for [PROJECT]: three month columns, each with 3–5 milestone cards ([PASTE MILESTONES + "targeting" dates]), client touchpoints marked with a distinct icon (walkthroughs, meetings, decisions), and a bottom rail "How you can help this quarter" listing client actions with dates. Confident, calm, uncluttered. No budget content.
Using the Sustainable 9 Design System, create a "Decisions That Shape Your Home" map for clients entering the Design phase: a flow of the major decision clusters in order (Site & Orientation → Floor Plan → Exterior Character → Interior Palette → Systems & Sustainability → Selections), each with 2–3 example decisions and when it happens. Purpose: clients see that early decisions unlock later ones. Warm, editorial, one Terra thread connecting the clusters like a path.
Using the Sustainable 9 Design System, create "Your First Year Home" — a 12-month arc for new homeowners: move-in orientation (day 1), 30-day check-in, seasonal care moments (fall prep, winter humidity, spring drainage walk), the 11-month walkthrough (hero moment, Terra), and how to reach us anytime. Tone: care, not fine print. Playfair closer: "We're your partners long after the keys."
Using the Sustainable 9 Design System, create a single-page diagram of [CLIENT]'s home's sustainability systems: simple house section/elevation sketch with callouts for [EDIT: envelope & insulation strategy, triple-pane glazing, HRV/ERV fresh air, high-efficiency mechanicals, solar-ready roof, water strategy, native landscape]. Each callout: name + one line on what it does for comfort or cost, homeowner language. This doubles as a closing-gift artifact and a marketing piece (Nataliya gets a generic version).
Using the Sustainable 9 Design System, create "How Your Build Is Funded" for clients with construction loans: the 5-phase spine with typical draw points marked, a 4-step "what happens at each draw" strip (milestone verified → package to lender → inspection → funding), and a reassurance note on lien waivers protecting them. NO dollar figures, NO percentages unless I paste verified ones. Educational, calm, confidence-building.
Using the Sustainable 9 Design System, adapt [PROJECT]'s phase-gate graphic into a public teaser: no client names, no address, no interior identifying details — "A lakeside story reaches the Build phase." Journey spine, one PUBLIC-SAFE hero image, tagline from the bank. Square + 9:16. Nothing posts without the consent log checked.