HomeZero · Build plan

The 12-Day Build

End-to-end first, then retune. Fri Jun 19 → Tue Jun 30. AI speed. We build the whole spine thin and working — quick first passes on the hard stuff — then come back and sharpen. Milestones you can see work, not a tower that only stands when the last brick lands. The product is from The Three-Agent Ladder.

Date written 2026-06-18

The philosophy

  • Vertical slices, not perfect layers. The product works end-to-end early and gets sharper — never “perfect only when the last piece lands.”
  • Quick first passes on the hard stuff (cost-to-zero, the score, brief depth). A basic advisor skill, a basic mini-briefing, a basic double-click briefing — good enough to run end-to-end, not a day of back-and-forth. Get it working rough, then come back and add the real value where it counts.
  • The structure is milestones — each a capability you can watch work — not isolated parts.
  • We loop daily: check the “done when,” roll misses forward, spin up agents as needed. This conversation is the heartbeat.
  • The board is Linear for the sprint — a deliberate scaffold (polished, multiplayer, mobile, with a native agent model + webhooks), so all our build goes into the agents. It's the dogfood surface, not the consumer end-state; see “The board: Linear” below.

The five milestones (the spine)

  1. The advisor works — sign up spins up your advisor; zero a home → mini-brief → board → a rough Home Zero Score. The whole single-user loop, working.
  2. Market, anywhere — the double-click spins up / engages a market agent for any place, not just Delta. Born-anywhere.
  3. The scan — the market's cloud deep-dive (county records, water, permits, satellite) layers onto the card and sharpens the score.
  4. The walkthrough scan — the home agent captures the first in-person tour (notes split you / market / home). Premium gets built in here.
  5. The deep scan — the return visit: measurements, photos, a guided self-inspection — the deepest layer.

Then: round it out + retune

Once the spine runs end-to-end: invite the others in (multiplayer — one shared board + advisor, so you + Melana + Tony dogfood together), export your advisor, add email reach, and come back to tune cost-to-zero, the score, and the briefs.

What stays a first pass (tune later, on purpose)

The cost-to-zero math (the score runs on honest heuristics until we tune it), brief depth, and score calibration. The formally-published brain schema (the export ships; the spec matures behind it).

The board: Linear (the dogfood scaffold)

For the 12-day build, the board is Linear — not our custom kanban. A deliberate scaffold, and the sharpest expression of assemble, don't invent: Linear hands us a polished, multiplayer, mobile board with comments, notifications, a first-class agent model, and instant webhooks — so all twelve days go into the agents, which is the moat.

The honest boundary

Linear is the dogfood surface, not the consumer end-state — a home buyer in crisis will never onboard into Linear. Our own board (Board v2 / PR #31) is shelved, not deleted; it becomes the consumer surface once the agents are proven. And self-sovereignty holds: the advisor's brain still lives on its cell and exports whole — Linear is a view of the board, swappable. We scaffold the UI, not surrender the soul.

The mapping (throughout this doc, “card” = a Linear issue)

  • The journey → Projects. Each phase (Footing → Targeting → Sizing up → Making the move → The gauntlet → Keys) is a Linear Project; we build Targeting first (the design's §2a). The user only ever sees the phase they're in.
  • Home → issue, created by the advisor.
  • Lanes → workflow states (Inbox → Worth seeing → Scheduled → Seen → Shortlist → Passed). The advisor moves issues; a human move fires a workflow-state webhook.
  • Home Zero Score / cost-to-zero / verdict → written into the issue body (markdown, advisor-owned) — not Linear custom fields, which the API exposes only partially. A verdict label (good / fair / caution) gives the at-a-glance color.
  • “Go deep” → assign the advisor agent (a home graduating from Targeting into Sizing up) → wakes the advisor → engages the market.
  • The advisor's asks → comments; the human answers in a comment → webhook → the advisor responds.
  • Invite → Linear team members — multiplayer is native; per-person attribution is free (Linear knows who moved or commented).
  • The full Brief → a linked doc / attachment on the issue (R2-hosted HTML). We trade the inline rich card for a real document.

The wiring

  • The advisor cell holds a Linear API key and is registered as a first-class Linear agent — AgentSessions give it native “working / waiting / done” status (the Symphony pattern).
  • Writes go out via Linear's GraphQL API (create/update issues, comments, move states).
  • Human actions come back via webhooks — issue / comment / label / workflow-state events (and AgentSession prompts) POST to a Cloudflare Worker → host-bridge → the advisor cell, reusing the zero-created doorbell pattern we already run.
  • Convex stays the spine — the extension still zeroes a home → Convex → the advisor sees it → the advisor creates the Linear issue. Linear is the human-facing board; Convex is the data behind it.

What this changes: Day 1 becomes stand up Linear + wire the advisor to it (not build a board). Multiplayer nearly evaporates (invite to the team). PR #31 isn't needed for the sprint. The capacity we save pours straight into agent depth — better briefings, a sharper score, more scans tuned.

M1The advisor works
01
Friday
Jun 19
Thing
Stand up Linear + wire the advisor to it. Create the Targeting project (the first journey phase — see the design) with the six workflow states as lanes + the verdict labels; register the advisor as a first-class Linear agent (a teammate that posts working / waiting / done status — the Symphony pattern); stand up the webhook receiver (Cloudflare Worker → host-bridge → advisor cell, reusing the zero-doorbell). Then: a zero → the advisor creates a Linear issue in Targeting with a stub mini-brief + score in the body.
Goal
The board is Linear and the advisor owns it — a zero lands as an advisor-authored issue in Targeting, and a human action comes back to the advisor.
Done when
Zero a home → an advisor-authored issue appears in the Targeting board with a stub mini-brief + score; the advisor shows as a teammate with status; and a human comment or move fires a webhook the advisor receives.
02
Saturday
Jun 20
Thing
Fill the issue — the Zillow-only mini-brief skill writes the body, and a first-pass Home Zero Score lands in it (basic heuristic, good enough to read).
Goal
A real, readable issue with a real score — the single-user loop works end-to-end.
Done when
Five real homes get distinct mini-briefs and a floor-level score in their Linear issues. M1 done
M2Market, anywhere
03
Sunday
Jun 21
Thing
“Go deep” spins up / engages a market agent for any place (born-anywhere), not just Delta.
Goal
Going deep wakes a market for whatever geography the home is in.
Done when
Go-deep on a home in a brand-new market brings a market agent to life for that place. M2 done
M3The scan
04
Monday
Jun 22
Thing
The scan, first pass — the market runs its cloud deep-dive (county records, water, permits, satellite).
Goal
A real scan comes back for the home.
Done when
Go-deep produces real public-record findings for the home.
05
Tuesday
Jun 23
Thing
The scan → card layers + a sharper score — the advisor weaves the scan onto the card; the score moves from floor toward “pays for itself.”
Goal
The scan visibly deepens the card and the score.
Done when
A scan on 229 Colorado Ave (real data) lands public-record layers and a visibly sharper score. M3 done
M4The walkthrough scan
06
Wednesday
Jun 24
Thing
The walkthrough capture, first pass — the home agent runs a Granola-style tour capture; notes split three ways (you / market / home).
Goal
A walkthrough produces structured notes onto the card.
Done when
A recorded or typed walkthrough lands attributed notes on the home's card.
07
Thursday
Jun 25
Thing
Walkthrough onto the card + build premium in — wire the walkthrough into the briefing; stand up the metered wall + Clerk $25/mo. (Could pull earlier if it's ready.)
Goal
The walkthrough deepens the home's picture, and the paywall is live.
Done when
A walkthrough sharpens the card and hitting the wall prompts a real $25/mo Clerk checkout. M4 done (+ billing)
M5The deep scan
08
Friday
Jun 26
Thing
The deep scan, first pass — the return-visit inspection (measurements, photos, guided self-inspection) as the deepest layer on the card.
Goal
The deepest scan completes the ladder.
Done when
A return-visit capture lands measurements/photos as the deepest card layer. M5 done; full ladder runs E2E
+Round it out + retune
09
Saturday
Jun 27
Thing
Multiplayer — invite the others in. Add Melana and Tony to the Linear team (co-buyer + scoped helper). Board-sharing and per-person attribution are native; the work is making the advisor handle multiple people's comments and moves cleanly — per-person sessions, with Linear telling it who.
Goal
You, Melana, and Tony on one Linear board, and the advisor keeps who-said-what straight.
Done when
An invited person comments or moves an issue and the advisor responds with their input attributed to them.
10
Sunday
Jun 28
Thing
Export your advisor + email reach. One-click “take it with you” zip of the advisor's brain; the advisor's email skill (send + act on replies), alongside the wired WhatsApp and Linear's own notifications.
Goal
You can walk away with your advisor, and reach it by email.
Done when
The export produces an openable brain zip and the advisor sends and acts on an email.
11
Monday
Jun 29
Thing
Retune — the first real pass at sharpening cost-to-zero and the score (heuristic → calibrated), and deepening the briefs now that the spine runs end-to-end.
Goal
The score gets honest; the briefs get real.
Done when
The score reflects a tuned cost-to-zero on real homes, and the mini / deep briefings are visibly better than their first pass.
12
Tuesday
Jun 30
Thing
Integrate + dogfood + re-cut the landing — full end-to-end pass on the real shared board through all five scans; re-cut the landing to the $25/mo story; close the seams.
Goal
The not-so-MVP works for the three of you, end to end.
Done when
Pete + Melana + Tony run a real home through advisor → market → scan → walkthrough → deep scan, paying at the wall, and the landing tells the new story.

The cadence at a glance

DayDateMilestoneThe thing
1Fri Jun 19M1 · Advisor worksStand up Linear + wire the advisor to it (zero → issue)
2Sat Jun 20M1 · Advisor worksFill the issue — mini-brief + first-pass score → loop works E2E
3Sun Jun 21M2 · Market anywhereGo-deep spins up a market for any place
4Mon Jun 22M3 · The scanThe cloud deep-dive, first pass
5Tue Jun 23M3 · The scanScan → card layers + sharper score
6Wed Jun 24M4 · Walkthrough scanThe tour capture, first pass
7Thu Jun 25M4 · Walkthrough scanWalkthrough onto card + build premium in
8Fri Jun 26M5 · The deep scanReturn-visit inspection → full ladder E2E
9Sat Jun 27Round outMultiplayer — invite the others to the Linear team
10Sun Jun 28Round outExport your advisor + email reach
11Mon Jun 29RetuneSharpen cost-to-zero, the score, the briefs
12Tue Jun 30RetuneIntegrate, dogfood, re-cut the landing