What we can ship while the review runs, who actually owes what, and the two builds that change what this product is.
12 October 2026 For Stevan and Tom. Companion to the WhatsApp gate briefing.
The App Review went in on 1 October. Meta's wording is "most submissions are reviewed within 20 days", and decisions often arrive inside a week, so the realistic window for this plan is zero to nine more days. That cuts one way: prefer work that is still valuable the day after approval lands. All of it below is.
Today is day 11. Not a deadline: a decision can land any morning, including tomorrow's.
Everything else on the list makes the product better. These two change what it is.
Today the bot never talks listings, by design: without data, stock talk is how bots lie. Wave 1 flips the weakness into the moat. SHO's Kyero-format feed loads into a matcher (area via gazetteer, beds, type, budget band), and guards allow a property to be named only if it is in the feed block, price and status from the feed alone.
Startable today: Kyero is a public format, so the loader, matcher and guards get built against a synthetic sample and Mediaelx's real export drops in whenever it finally arrives. It is a month overdue. This also answers Sellebrate's matching module, with truthfulness discipline attached.
Wave 4: lead emails from A Place in the Sun and Rightmove forward to an inbox we read; parse, create the lead record, queue the opener. Plus dedupe, cool-off rules for repeat enquirers, and a scheduler for openers and the 2-day and 7-day nudges. The difference between a chatbot demo and a product an agency forgets is running.
The parser needs three raw lead emails from Tom. The scheduler, dedupe and cool-off scaffolding need nothing and can be built first.
| Item | Waiting on | Build | |
|---|---|---|---|
| 1 | Stock feed (lever A)The bot offers real properties, guarded against invention | Ready now Mediaelx for live data | 3 to 4 days |
| 2 | Lead-email ingest (lever B)Enquiry to conversation with no human typing; closes the loop | Tom: 3 raw lead emails | 2 to 3 days |
| 3 | Backlist importerRevive dead leads; the pilot's commercial proof, built to our opt-in rules | Skeleton now Tom: source + 20 rows | 1 day |
| 4 | Wave 2b + Tom's 15 Sep fixesHonour contact preference, hand off dated-trip-less cash buyers, challenge past dates, kill recap openers, La Torre false fallback, money-verdict phrase, question-cap bug | Ready now | 1 to 2 days |
| 5 | Copilot approve queueCallum approves or edits each send from his phone; the safety story for real traffic | Ready now | 2 to 3 days |
| 6 | Marketing site + nameThe sellable surface; its domain doubles as the stable host /onboard needs anyway | Stevan: name + round-3 notes | 2 days |
| 7 | Exec routing + hand-off watcherSheet routed by area, one-tap outcome, re-escalate silent hand-offs: leads never fall into a void | Adam: roster, coverage, hours | 2 days |
| 8 | Voice-note handlingBuyers' voice notes stop dead-ending: transcribe, then the normal pipeline. New scope | Stevan: green-light | 1 to 2 days |
| 9 | Independent review of onboarding codeCommit 01ddf9f creates subaccounts and writes credentials; harden before any agency touches it | Ready now | Delegated |
| 10 | Bundled-brief vs one-question A/BAn open product decision, settled with data on the same 12 personas | Ready now | Half a day |
Deliberately absent: the 1,200 lead-generator cohort stays on hold until the consent answer is in writing, per the standing decision. And nothing here touches the "Agency AI" app while Meta is looking at it.
The scarce resource this fortnight is not build time. It is five people's inputs. Chasing the left four columns is worth more than any single feature.
One possible shape, biased to cheap-and-sharpening first, then the levers. Solid bars need nobody. Dashed bars start the moment their input lands, and the dates shift with them.
Estimates, not commitments. If approval lands mid-stream, the post-approval steps (Access Verification, the Twilio ticket, signup config) slot in alongside; nothing above gets discarded. The devbox demo is never restarted while Tom is mid-test.
Every build above exits through the same gate the last three waves used:
just test before any deploy, including the tenant-isolation and voice-rule suites.It's scope. The discipline for the fortnight: ship the ranked list top-down, chase the owed inputs daily, and let the review land whenever it lands. The product that greets approval day should already qualify harder, know the stock, and have a pilot sender warm.