Inline AI assistant inside HubSpot/Pipedrive that drafts follow-ups grounded in the prospect's full pipeline history — not generic chat answers. Press ⌘K, get a context-aware reply written into the deal note.

What we ship

**Core features (MVP)**

• Inline ⌘K command palette inside HubSpot/Pipedrive deal pages — drafts follow-ups in 2-3 seconds.
• Pipeline memory: every reply pulls full deal history, last 10 emails, notes, stage transitions as Claude context.
• Deal-stage-aware prompting: cold outreach, discovery, negotiation, onboarding each get a different system prompt.
• One-click "Insert into note" / "Send via CRM" — the draft never leaves the CRM tab.
• Cross-CRM sync: works on HubSpot AND Pipedrive with one Solo seat, no separate billing.

**Primary user flow**

1. Rep finds us via IndieHackers / RevGenius Slack / "Lavender alternative" search.

2. OAuth-connect HubSpot or Pipedrive in 60 seconds, install the browser extension.

3. Open any deal, hit ⌘K, type "follow up on Q2 budget concern" — Claude drafts a reply that references the actual call notes from 2 weeks ago.

4. Hits the daily 5-draft free limit on day 3 → upgrades to $29/mo Solo.

**What it looks like**

• **Landing page**: split-screen hero. Left: "The AI dance is killing your pipeline." Right: 8-second loop of a rep pressing ⌘K inside a HubSpot deal and a context-aware draft appearing. CTA: "Connect HubSpot — try free for 14 days."
• **Main app screen**: there is no app screen — the product LIVES inside HubSpot/Pipedrive. The "main UI" is a Cmd+K modal overlay: top input ("write a follow-up..."), middle shows the relevant deal context Claude pulled (3-4 chips: last call, deal value, stage, blocker), bottom shows the draft with Insert / Regenerate / Edit Tone buttons.
• **Notification / output format**: drafts appear as deal-note comments with a small "Drafted by [product] — review before sending" footer; weekly digest email shows time saved + drafts sent.

**MVP build plan (9 days)**

• Day 1-2: FastAPI + Postgres on Railway, JWT auth, Stripe Subscriptions wired.
• Day 3-4: HubSpot + Pipedrive OAuth, deals/contacts/notes sync into Postgres cache.
• Day 5-6: Browser extension (React + Tailwind), ⌘K modal, Claude Sonnet 4.6 streaming.
• Day 7: Stage-aware prompt templates, deal-context retrieval, "Insert into note" wiring.
• Day 8: Cloudflare Pages landing, pricing page, 14-day trial flow.
• Day 9: Product Hunt assets, IndieHackers post draft, Stripe webhooks tested, ship.

Source signals

• `SIG-20260426-us-ai-wrapper-vertical-copy-paste-dance` — quantified 5 min/hr loss to AI dance (Yoink AI author tracked it for a week)
• `SIG-20260426-us-ai-wrapper-vertical-multi-llm-tab-chaos` — multi-provider tab juggling pain (TypeThinkAI hit $200 MRR solving the simpler horizontal version)
• `SIG-20260426-us-ai-wrapper-vertical-rag-moat` — vertical RAG products churn 2-4% vs generic 12-15%; meta-signal that we MUST build with data moat, not generic prompts

Reasoning

Three independent signals converge on the same opportunity: knowledge workers waste meaningful time on the "AI dance" (context switching), and the products that survive are those with a data moat (user-specific memory, deep workflow integration), not generic ChatGPT wrappers.

Sales ops is a high-pain, high-payment-willingness vertical:

• Sales reps live inside their CRM 6+ hours/day
• Every minute saved on follow-up writing is directly attributable revenue
• HubSpot/Pipedrive CRMs have stable public APIs — integration is a 1-2 week project, not a moat-busting endeavor
• The "memory" moat is real: an AI that remembers "Acme Corp's CFO mentioned Q2 budget freeze in our last call" is impossible to clone with raw ChatGPT

Differentiation vs generic AI assistants:

• Inline (⌘K inside CRM, not a separate tab)
• Pipeline-aware (knows what was promised when)
• Deal-stage-specific prompting (cold outreach vs negotiation vs onboarding)

This is a `vertical-micro-tool × ai-wrapper-vertical` hybrid — essentially what Yoink AI does, but specialized for one workflow.

Quick competitive read

Direct competitors to verify in Stage 5b:

• **Lavender AI** (sales email AI): general-purpose, not CRM-integrated. Missing pipeline memory.
• **Twain** (cold-email AI): one-shot generation, no memory.
• **HubSpot ChatSpot** (built-in): generic, surfaces stats not drafts; light RAG.
• **Apollo's email AI**: tied to their CRM only.
• **Pipedrive's Smart Apps**: marketplace for AI add-ons but no first-class inline assistant with memory.

Initial competitive read: **medium density** (3-5 competitors) but each has specific weakness — none does inline-in-CRM + persistent pipeline memory + cross-CRM (HubSpot AND Pipedrive, not exclusive). White-space exists.

Initial pricing hypothesis

$29/mo Solo (one CRM seat), $49/mo per seat for teams of 2-10. Annual 20% off. Free trial 14 days. Based on developed-tier solo subscription willingness ($10-100/mo from taxonomy.yaml).

Distribution hypothesis

Build-in-public on X (Twitter) + IndieHackers + Product Hunt launch. Target: 20 first-100-users from sales-ops Slack communities (RevGenius, Pavilion, Modern Sales Pros). Cold outreach via LinkedIn to HubSpot power-users. Once 50 paying customers — switch to content marketing on "sales productivity" SEO (programmatic content for "how to write a follow-up email after [stage]" queries).