Ten features, shipping now

Shift to an assistant-centric architecture where every user belongs to a team — solo users get a team of one. Eliminates the solo-vs-team code branching and enables shared assistants, shared contacts, and shared inboxes out of the box.

Data Model

• Team owns usage, contacts, and assistants (all team-level, all visible to all members)
• User joins teams via TeamMembership with a role (owner / admin / member)
• Roles: owner/admin create & configure assistants; members can view everything

Why this unlocks virality

Once a user adds a colleague, that colleague instantly sees the shared inbox & contacts — turning a personal AI into a team tool without migration.

Senior callers (boss, significant other, major client, major partner) feel insulted when the AI handles them, causing churn. Neutral fix: exclude them from AI. Better fix: upgrade the experience — AI formally greets VIP and immediately tries a warm transfer.

Behaviour

• AI greets formally & immediately attempts warm transfer
• Long VIP-unique urgent sound + Lock Screen Live Activity w/ caller info & live transcript
• User can join or re-schedule ASAP; AI keeps VIP engaged ~30s, otherwise re-schedules
• AI/UI advises user not to miss calls from VIPs

VIP defaults: Boss, Significant Other, Major Client, Major Partner.

Continuous observation of AI-handled calls to detect issues (dissatisfaction, mis-routing, prompt regressions) and drive prompt & behaviour improvements that compound into retention gains.

Scope doc in progress — see roadmap comments for status.

Most users don't want or know how to customize their AI. Magical if it pre-fills based on their (sub) industry, with examples to follow or edit.

What ships

• Select (sub) industry at onboarding → fields pre-filled
• Add missing global fields: caller scenarios for new client / existing client / partner, products & services, pricing, areas served, open hours
• Industry-specific fields (e.g. accepted insurance for healthcare)
• Simulation + prompt learning to ensure the values actually work
• Expand from 3 sub-industries (pharmacy, equipment rental, real estate) to 27 (dental, chiropractic, pest control…)

Scheduled outbound calls and texts that boost user revenue. Built on contact events — logging each time a service is rendered — plus caller ID so recipients recognise the sender.

Three versions

• v1 Renewal — "it's the end of your {service} soon, do you need a new one?" (e.g. pharmacy refill due, lift rental return date, real-estate homeiversary)
• v2 Check-in — "day 1 of your {service}, how's it going? need any help?"
• v3 Referral — "sounds like you need {partner_service}, would you like us to arrange it?" — builds the network flywheel

HIPAA-aware: oncology SMS only invites a call-back; no PHI in text. Default send mode for oncology = call.

Users aren't shown the revenue impact of YouSquared — this contributes to churn. Solution: a stats page that quantifies revenue uplift, shows how to grow it (more channels, more users), and is shareable.

Features

• Calculate $ lead value handled (Price × ΔLTV × U2 attribution)
• Recommend how to increase revenue uplift further
• Highlight impressive calls to show off socially

Worked examples

Equipment rental — 86ft boom quote: price $5,775 · ΔLTV $11.5k · U2 attribution 85% → $10k

Pharmacy — GLP-1 refill: price $199/mo · ΔLTV $2.4k · U2 attribution 10% → $240

Real estate — angry $1M listing client: commission $22k · ΔLTV $12k · U2 30% → $3.6k

Auth-gated Next.js console at app.yousquared.ai, hitting the same api-service backend as iOS. Unlocks self-serve sign-up and team admin flows that are awkward on mobile.

Phase 1 — MVP

• Next.js in clone-backend/web-app/ (monorepo), deployed on Vercel Pro
• app.yousquared.ai CNAME to Vercel, CORS allow-listed
• Firebase Auth web SDK, ID-token validation server-side (same as iOS)
• Marketing CTA on www.yousquared.ai → app.yousquared.ai

Phase 2 — Development workflow

Push-to-branch → password-protected Vercel preview (hits prod API, like TestFlight); merge to main → auto-deploy to prod. Mixpanel + Amplitude JS SDKs, Sentry TBD.

Phase 3 — Long-term

• Proper staging environment (public api-service behind IAP / Cloud Armor)
• GKE migration from Vercel (cost / operational consistency trigger)
• Fold Webflow marketing site into Next.js (unlocks voice-onboarding landing page with auth context)

AI predicts the next step after a call and makes it easy to act — re-engaging silent callers, pinning user follow-ups, and assigning work to colleagues (which is the viral path into Team plans).

Three versions

• v0 Re-engage — caller hung up without speaking → AI drafts/sends a personalized text-back to qualify & re-engage. Basic users see the draft, upgrade to send.
• v1 To-do + follow-up — caller needed user action → AI pins the action, keeps thread top-of-inbox, loads a draft follow-up text when user marks done.
• v2 Assign — AI suggests action for AI / user / colleague, assigns it, reminds the assignee, informs caller once done. Colleague assignment adds a user to the Team (30 days free, then requires right-size Team plan).

Autonomous Prompt Learning (APL) — a coding agent labels production data, flags ambiguous cases for human review, writes & evaluates task prompts, and iteratively self-refines them in a single agentic loop.

Production results

• B2C2B lead classification: 95–97% F1 (up from ~3–5% rule-based)
• Post-call next-step prediction: near-zero false positives (down from ~16%)
• Expert-advice drafts grounded in call content

Each optimization cycle: 15–60 min, production data, one human supervisor. No specialized infra (unlike DSPy, OPRO, TextGrad) — just a capable coding agent and an eval harness the agent builds itself.

Next steps

• Formalize the methodology as a reproducible protocol
• Controlled ablations (active vs random sampling, self-reflection vs single-shot, LLM vs human judge agreement)
• Apply to 2–3 more downstream tasks; literature review; workshop paper

Drive per-minute AI call cost down to $0.04/min — the threshold at which aggressive pricing and higher-volume plans become sustainable.