Automation~8-10 hours to build$5K/Month goal

Inbox Zero Agent

An AI inbox agent that prioritizes mail, drafts replies in your voice, schedules sends, and surfaces only the threads that need you—Gmail and Outlook.

By John IseghohiPublished July 25, 2026

Opportunity 9/10
Pain 9/10
Timing 9/10
Confidence 8/10

The Problem

Knowledge workers spend more than two hours every day inside email — reading, triaging, drafting, deferring, and re-reading threads they meant to answer yesterday. The inbox is an infinite to-do list owned by everyone else. A founder opens Gmail at 7 a.m.: 47 unread. Fourteen are newsletters, nine are FYI cc's, six are genuine asks with buried deadlines, three are angry customer threads needing careful replies today. The rest is scheduling ping-pong and "quick question" messages that each cost fifteen minutes of context-switching.

The mental load is decision fatigue. Every message asks: Is this urgent? Can I ignore it? Reply now or batch for Friday? Most people develop brittle hacks — color labels, filters that miss edge cases, starred items that become a second inbox, draft replies that sit open for days because the right tone takes energy they do not have at 4 p.m.

Existing tools help at the margins. Filters sort. Snooze hides until Monday. Smart folders surface "important" senders via opaque heuristics. None actually do the work: prioritize with context, draft in your voice, schedule the send, and surface only what needs human judgment. Result: inbox anxiety, slow replies on messages that matter, and a guilt loop around inbox zero that never closes. A consultant billing $150/hour who loses ninety minutes to triage burns $225 before lunch. Email management software is a multi-billion-dollar category because this problem scaled with async work and AI-generated outreach.

The Solution

Inbox Zero Agent (working name: InboxZen) is an AI inbox agent on top of Gmail and Outlook — a chief of staff for your mail. It scores priority from sender history, deadline language, and thread context; drafts replies in your voice; schedules sends; and surfaces a daily attention queue of five to twelve threads that actually need you.

Not another email client. Users keep Gmail or Outlook. A Chrome extension adds a side panel; the agent uses official APIs so UI changes do not break it. Wedge: founders, consultants, account managers, and recruiters who pay $15/month to reclaim an hour without learning Superhuman-style shortcuts.

How it works:

Connect Gmail or Outlook — OAuth through the extension; the agent indexes recent sent mail to learn your tone, sign-off patterns, and typical response length.
Morning attention queue — Each day opens with a ranked list: reply-now, reply-today, delegate, archive. Newsletters and automated noise never appear unless you ask.
One-click draft replies — Tap any thread; Claude generates a reply in your voice with context from the thread and your sent-mail style guide. Edit, approve, or discard.
Schedule sends — Batch non-urgent replies for 9 a.m. Tuesday when you are fresh; the agent queues them through the provider's send API.
Weekly inbox report — See where time went: top senders, average response lag, threads you snoozed three times (a signal to unsubscribe or template).

Retention compounds: approved drafts train voice; "not urgent" dismissals sharpen scoring. After two weeks the agent knows your biggest client always ranks high and vendor cold outreach can auto-archive.

Market Research

Email management software is a large, steady-growth market driven by remote work, async collaboration, and the relentless volume of business communication. Analysts consistently size the category in the multi-billion-dollar range with high single-digit to low double-digit CAGR through the early 2030s.

The global email management software market was valued at roughly USD 6.5 billion in 2023 and is projected to reach USD 13.5 billion by 2032, growing at approximately 8.5% CAGR (DataIntelo).
Market share consolidation and AI feature adoption are accelerating vendor investment in 2025, with enterprise and SMB segments both expanding (OpenPR).
Adjacent email tooling — marketing platforms, tracking, and automation — validates sustained spend on anything that makes email less painful; the broader email software ecosystem continues to grow as businesses treat inbox workflow as operational infrastructure (IMARC Group, Market Research Future).
2+ hours per day in email is the commonly cited knowledge-worker average; even a 30% reduction is 36 minutes daily — over three hours per week — which easily justifies a $15/month subscription on time value alone.
Willingness to pay is proven at multiple price points: Superhuman sustains a $30/month cult following; SaneBox charges $7 and up for filtering alone. A $15 agent that drafts and schedules — not just sorts — sits in a credible gap between free bundled features and premium clients.

SAM: tens of millions of Gmail and Outlook users who pay for productivity tools but refuse to switch clients. Distribution via productivity communities and high-intent "inbox zero" search terms.

Competitive Landscape

The category splits into premium clients, filter-and-snooze utilities, AI writing add-ons, and free platform bundles. None combine priority intelligence, voice-matched drafting, and scheduled sending in a single lightweight agent that works inside the inbox users already have.

Superhuman — The premium email client benchmark. Blazing UI, split inbox, AI snippets, team read statuses. Requires leaving Gmail's native UI and learning new shortcuts. $30/month individual; team pricing higher. Strong brand, high switching cost, but overkill for users who want assistance without migrating clients.
SaneBox — AI-powered filtering, snooze, attachment sorting, and follow-up reminders. Works with existing clients via IMAP. Excellent at hiding noise but does not draft replies or learn voice. ~$7/month (Snack plan) to $36/month (Dinner) depending on features and mailboxes.
Fyxer AI — Meeting notes and email drafting assistant oriented toward executives. Integrates with Gmail and Outlook; focuses on meeting follow-ups and inbox cleanup. Pricing typically $10–15/user/month range for AI features. Closer competitor on drafting, but less emphasis on daily priority queue and scheduled batch sending.
Mailbutler / Mixmax — Gmail and Outlook add-ons for templates, tracking, scheduling, and smart assistant features. Mailbutler Smart Assistant from roughly $5/month; Mixmax tiers from $13–33/user/month. Broad feature sets; AI is one module among many, not a unified agent experience.
Gmail Smart Features (bundled) — Summarize thread, smart reply chips, priority inbox heuristics. Free with Google Workspace. Good enough for casual users; no voice learning, no cross-thread priority scoring, no scheduled batch workflow, and Outlook users get nothing.

Your Opportunity

Own the "AI chief of staff" slot at $15/month individual and $40/user/month teams — half of Superhuman, more capable than SaneBox, more focused than Mailbutler's Swiss-army approach. Position as: Keep Gmail. Lose the triage. Land with founders and client-facing pros who feel inbox shame daily; expand to team plans where shared SLA dashboards ("who owes a reply") and admin voice guidelines unlock higher ACV. The gap is an agent that reduces decisions, not another surface to learn.

Business Model

Per-user SaaS with a generous trial that demonstrates the first morning queue and three draft replies. Individual tier is the PLG engine; team tier adds shared analytics, SLA tracking, and admin controls for agencies and sales pods.

Free Trial (14 days) — Full attention queue and drafting on one mailbox; capped at 50 AI drafts to prevent abuse.
Individual ($15/month) — Unlimited drafts, priority scoring, send scheduling, weekly reports, one Gmail or Outlook account.
Team ($40/user/month) — Everything in Individual plus shared team dashboard, response-time SLAs, shared templates, admin voice guide, priority rules per role (sales vs. support).
Annual discount — 2 months free on yearly billing to improve cash flow and reduce churn.

Unit Economics (illustrative)

$0.08–0.25 — LLM cost per draft (Claude Haiku for triage, Sonnet for final reply), amortized over daily batch scoring
~80% — Gross margin at Individual tier with sensible caching of sender profiles and thread summaries
$25–60 — Target CAC via productivity content and Chrome Web Store discovery
110%+ — Net revenue retention as solo users invite teammates for SLA visibility

MRR path: 500 Individual = $7.5K/mo; add 50 ten-seat teams = $27.5K/mo total. Voice and priority history reduce churn — retraining elsewhere hurts.

Recommended Tech Stack

The build is a Chrome extension for UI and provider API access, a Next.js backend for auth and AI orchestration, and Postgres for user data and learned preferences. Gmail API and Microsoft Graph cover mail read, draft create, and scheduled send. Avoid scraping DOM — API-first keeps the product stable across UI changes.

Chrome Extension (Manifest V3) — Side panel for attention queue, thread preview, draft editor, and approve/schedule actions. Background service worker syncs new mail on interval and push notifications where supported.
Next.js App Router + Vercel — Landing page, OAuth callback routes, billing portal, weekly report emails, and API routes for extension communication. Server Actions for draft generation jobs.
Gmail API + Microsoft Graph — Read threads, list messages, create drafts, send or schedule send (Gmail: users.messages.send with scheduled time via draft + Apps Script alternative or provider-specific scheduling; Outlook: deferred send via Graph). Store refresh tokens encrypted.
Claude (Anthropic API) — Haiku for priority classification and thread summarization; Sonnet for voice-matched draft generation. System prompt includes user style guide extracted from last 200 sent messages.
Postgres (Supabase or Neon) — Tables: users, mail_accounts, threads, priority_scores, drafts, send_queue, style_profiles, feedback_events. pgvector optional for semantic sender clustering.
Stripe Billing — Individual and per-seat Team subscriptions; usage metering optional if draft volume exceeds fair-use thresholds.

Weekend MVP: Gmail only, queue + draft + manual approve. Week two: Outlook and scheduled send.

AI Prompts to Build This

Copy and paste these into Claude, Cursor, or your favorite AI tool.

1. Chrome Extension + Gmail OAuth Shell

Build a Chrome extension (Manifest V3) plus Next.js 14 App Router backend for an AI inbox agent called Inbox Zero Agent.
 
Extension:
- Side panel UI showing "Attention Queue" (ranked thread list with priority badge, subject, sender, one-line summary)
- Background service worker that polls backend every 5 minutes for queue updates
- Content script minimal — no DOM scraping; all data via backend API
- OAuth flow: extension opens popup to /api/auth/gmail on the Next.js app; store session token in chrome.storage.local
 
Backend routes:
- GET /api/auth/gmail — start Google OAuth with scopes: gmail.readonly, gmail.compose, gmail.send
- GET /api/auth/gmail/callback — exchange code, encrypt refresh token, store in mail_accounts table
- GET /api/queue — return today's attention queue for authenticated user
- POST /api/draft — given thread_id, generate and return draft reply
 
Postgres schema:
- users(id, email, plan, stripe_customer_id, created_at)
- mail_accounts(id, user_id, provider ENUM[gmail,outlook], refresh_token_encrypted, last_sync_at)
- threads(id, mail_account_id, provider_thread_id, subject, sender, last_message_at, priority_score, summary, status ENUM[pending,drafted,scheduled,done,archived])
- drafts(id, thread_id, body, status ENUM[generated,approved,sent,scheduled], scheduled_for, created_at)
- style_profiles(id, user_id, tone_notes, avg_length, sign_off, example_snippets JSONB)
 
Landing page headline: "Your inbox, minus the triage." Subhead: AI priority queue and voice-matched drafts for Gmail.
Use TypeScript throughout. Environment variables for Google client ID/secret and Anthropic API key.

2. Priority Scoring + Attention Queue

Implement the daily attention queue pipeline for Inbox Zero Agent.
 
Sync job (runs every 15 min per connected account):
1. Gmail API: list threads with label INBOX, max 100, ordered by internalDate desc.
2. Skip threads already scored in the last 6 hours unless new messages arrived.
3. For each new/changed thread, fetch snippet + last 3 messages (plain text, strip quoted replies).
 
Priority classifier (Claude Haiku, JSON mode):
Input: thread subject, sender, snippet, user role (optional), historical sender_stats (avg response time, past priority labels).
Output schema:
{
  priority: "reply_now" | "reply_today" | "delegate" | "archive",
  score: 0-100,
  reason: "one sentence",
  detected_deadline: "ISO date or null",
  summary: "one sentence thread summary"
}
 
Scoring rules in system prompt:
- reply_now: explicit urgency, angry customer, deadline within 24h, VIP sender marked high in style profile
- reply_today: direct question, meeting request, client thread without urgency markers
- delegate: FYI with action item for someone else, internal routing
- archive: newsletter, automated notification, cold outreach, no-reply needed
 
Store results in threads table. Attention queue API returns top 15 threads where priority in (reply_now, reply_today) sorted by score desc.
 
Feedback loop: when user marks "not urgent" or archives from queue, log feedback_events and downgrade sender priors.

3. Voice-Matched Draft Replies + Scheduled Send

Build draft generation and scheduled sending for Inbox Zero Agent.
 
Style profile bootstrap (on first connect):
- Fetch last 200 sent messages via Gmail API.
- Claude analysis: extract tone (formal/casual), typical length, common sign-offs, greeting patterns, 5 example reply pairs (incoming → sent).
- Store in style_profiles.
 
Draft generation (POST /api/draft):
Input: thread_id, optional user_notes ("softer tone", "push meeting to next week").
Fetch full thread. Build prompt:
 
System: "You write email replies matching this user's voice exactly."
Include style_profiles fields and 3 example snippets.
User: thread transcript + user_notes.
Output: plain-text reply body only, no subject line, no markdown.
 
Return draft for side-panel editor. User can edit inline.
 
On approve:
- Gmail API users.drafts.create with threadId binding.
- If send_immediately: users.messages.send with draft id.
- If scheduled_for timestamp: store in send_queue table; cron job (Vercel Cron every 5 min) checks due items, sends via API, marks thread done.
 
Track: drafts generated, approval rate, edits before send (measure voice accuracy). Gate unlimited drafts behind Individual plan; trial capped at 50.