AgentMail growth playbook

Write one short, dated proof block and serve it in every markdown response that matters: the homepage, /pricing, /enterprise, the /use-case and /build pages, and a new Customers section in llms.txt. It should give the email volume (1 billion, October 2026), the B2B customer count with its date, SOC 2 Type II, the lead investors and one line per published customer story with its number and link (for example AgentCard's roughly 70% more agentic transactions and Balance's inbox setup going from half a day to one API call). Keep each fact in a single plain sentence that models can quote accurately, generate the HTML and markdown from the same source so they cannot drift apart, and measure share of voice on target prompts before and after.

Agents and answer engines repeat what is on the page they fetch. A capability list makes AgentMail one option among many; named customers, volume and compliance make it the recommended one. The facts are already public, so the cost is a day of writing and a template change.

[ ] Markdown homepage: volume, customer count, SOC 2, investors, 3 named stories
[ ] llms.txt: Customers section, one line and link per story
[ ] Markdown /pricing: which plan includes the SOC 2 report
[ ] Markdown /enterprise: SOC 2, EU region, named customers
[ ] Each /use-case and /build page: the closest customer story
[ ] HTML and markdown generated from one source
[ ] Every number dated
[ ] 20-prompt share-of-voice baseline saved

• Numbers go stale and models keep quoting them

  Date every figure and update from one source monthly.

• Models misquote long claims

  Keep each fact to one short sentence with a link to its source page.

• Your websiteHomepage as AI readers receive it (the site serves a markdown version to agents; fetched November 9, 2026): "The email inbox API for AI agents", a list of capabilities (REST inbox creation, threading, webhooks and WebSockets, custom domains, Python and TypeScript SDKs, an MCP server, Pods, IMAP/SMTP) and the free tier (3 inboxes, 3,000 emails a month). It names no customer, no email volume, no investor and no SOC 2. A crawl of the same homepage for human visitors found links to customer stories (Opendoor, Clay, AgentCard) and 5 testimonials.Source
• Your websitellms.txt opens with sign-up instructions for agents (an agent can create an account with its developer's email; a one-time code goes to the developer), then sections on pricing, comparisons, evals, use cases, integrations, "Security & Compliance" ("SOC 2 Type II compliant") and "Markdown Content Negotiation". It names no customer, no case study and no email volume.Source
• HiringAgent Experience Engineer (posted August 10, 2026): "Our most important user is an agent: it discovers us through model answers and docs." The role will "measure share of voice on the prompts that matter and move it", run evals of how models and agent harnesses use AgentMail, own llms.txt, MCP descriptions, docs and "the error messages agents hit at 3am with no human watching", and do "backlink outreach to the pages models already cite". "A stray Reddit thread once convinced LLMs we have a feature that doesn't exist."Source
• Your websiteCustomer Stories: 13 published stories, including Opendoor (closing workflow), Lanesurf (freight; "hundreds of inboxes, millions of messages"), AgentCard by Alchemy (about 70% more agentic transactions), Browser Use (every cloud session ships with its own inbox), AISDR (hundreds of inboxes in production), Balance (inbox setup from half a day to one API call), Pancake (response times from hours to minutes), Dossium (an inbox per user at signup) and Cofounder (800+ emails in week one).Source
• Your websitePress page: "1 billion emails" launch (October 21, 2026, 201K views on X), AgentID launch (October 6, 2026, 510K views), seed announcement (237.6K views), Agent Armor showcase and Domain-connect release (October 28, 2026), releases of allowlists and blocklists, permission-scoped API keys and labels, and appearances on TBPN and This Week in Startups.Source

Pick one canonical page per question, merge the best content into it, redirect the duplicates, add an author, a published and updated date and a link to the docs, and add the survivors to the sitemap. Rewrite the "best providers" page as a fair comparison that names the real options and says when each fits.

One strong, dated page per question is easier to rank and to cite than seven thin ones, and a fair roundup is far more likely to be quoted by answer engines than one that names only its author.

• Losing traffic from redirected pages

  Redirect each to the closest canonical page and watch Search Console.

• Your websiteA batch of /blog/info/ pages dated November 9, 2026, with overlapping titles: "Can an AI Agent Provision Its Own Inbox on Demand", "How to Let an AI Agent Provision Its Own Inbox on Demand", "Giving an Agent a New Inbox Mid Workflow on Demand", "On Demand Inbox Creation: How Teams Give Agents a New Address Mid Workflow", "4 Ways to Give an AI Agent an Inbox It Provisions Itself", "Tools That Let an Agent Create New Email Inboxes Programmatically" and "How to Create New Email Inboxes Programmatically for your Agents". The pages show no author or date, and "Best Agent Email Inbox Providers for Developers" (about 1,300 words) names no other provider.Source
• Your websiteSitemap (200 URLs, checked November 9, 2026): 124 blog posts, 23 /use-case pages, 12 /build pages (one per coding agent or app builder), 18 press pages, /customers, /comparisons, /enterprise, /agentid, /agent-armor and /evals. The /blog/info/ pages are not in it.Source
• HiringAgent Experience Engineer (posted August 10, 2026): "Our most important user is an agent: it discovers us through model answers and docs." The role will "measure share of voice on the prompts that matter and move it", run evals of how models and agent harnesses use AgentMail, own llms.txt, MCP descriptions, docs and "the error messages agents hit at 3am with no human watching", and do "backlink outreach to the pages models already cite". "A stray Reddit thread once convinced LLMs we have a feature that doesn't exist."Source

Put one honest table at the top of /comparisons: inbox per agent by API, inbound threads, webhooks and WebSockets, custom domains, agent sign-up, identity (AgentID), inbound safety, SOC 2, free tier and starting price, across AgentMail and the main alternatives. Add dated head-to-heads with the agent-native products and link each row to its page.

Buyers and agents comparing options want one table; agent-native rivals are where the real decision now happens, and AgentMail's volume, customers and AgentID are clear differences.

• Outdated claims about rivals

  Date each page, cite rivals' own pages and refresh quarterly.

• Your website/comparisons lists 23 articles, all blog posts and none with a summary table: head-to-heads with Resend, Nylas, SendGrid, Postmark, Mailgun, the Gmail API, Amazon SES and Cloudflare Email Service ("launched Email Service for agents in April 2026"), plus roundups. None compares AgentMail with other inbox-for-agents products.Source
• Competitor researchRivals' own homepages: Mails.ai ("Give your agent its own inbox.") leads with prompt-injection scoring and quarantine on inbound mail, a test address that auto-replies, sign-in code and magic-link extraction, and per-agent reputation; Atomic Mail promotes "Atomic Mail for Agents" beside its encrypted consumer email. Send-first APIs (Resend, SendGrid, Mailgun, Postmark) are adjacent.See the research Source
• HiringFounding GTM & Operations Lead (posted August 10, 2026): "Build and own the GTM playbook across all channels including PLG, content, community, outbound"; "Define category narrative, positioning, and messaging so AgentMail becomes the default answer"; "Drive initiatives in pricing"; build "the dashboards that drive decisions". AgentMail has raised $6M from General Catalyst, Y Combinator, Paul Graham and founders of Ramp, Supabase and HubSpot.Source

Add the number of messages held back by a filter to list responses, with a plain hint on how to see them; state on every endpoint page what it cannot do and which endpoint to use instead (reply-all without BCC, send with BCC). Rerun the two failing tasks from the eval and publish the before and after in the changelog and on the report.

The eval already isolated the failures; fixing the exact spots raises agent success, which is activation for an API whose main user is an agent, and the published rerun is proof for builders.

• Response changes break existing parsers

  Add fields without changing existing ones.

• Your website"Evaluating how AI agents use tools" (October 3, 2026): 25 agents (a model inside a harness) ran seven email tasks and 80% of runs finished. "They rarely failed at the work itself. They failed where the tool left them guessing." The most common stop was an inbox that looked empty (35% of failed runs; agents read a filtered list as empty). On reply-all with BCC, success fell from 81% to 62% when docs were linked, because the docs pointed to an endpoint that cannot add BCC. Overall, linking docs raised success from 76% to 84%. The report lists planned fixes.Source
• HiringAgent Experience Engineer (posted August 10, 2026): "Our most important user is an agent: it discovers us through model answers and docs." The role will "measure share of voice on the prompts that matter and move it", run evals of how models and agent harnesses use AgentMail, own llms.txt, MCP descriptions, docs and "the error messages agents hit at 3am with no human watching", and do "backlink outreach to the pages models already cite". "A stray Reddit thread once convinced LLMs we have a feature that doesn't exist."Source
• HiringDeveloper Success Engineer (posted October 12, 2026): "Your job is to answer the question, figure out why it had to be asked, and make sure we learn from it"; debug API requests, webhooks and integrations, reproduce issues or open fix PRs; "Notice recurring questions and friction across support conversations and surface those patterns".Source

Add a small slider for inboxes and monthly emails that shows the cost on Developer with add-ons and on Startup, flags the cheaper plan and mentions what Startup adds (SOC 2 report, higher limits, more seats). Show the same nudge in the console when a Developer org passes about 70 inboxes.

Showing the crossover moves growing teams to the plan that suits them before they hit limits, and makes the $2 add-ons feel fair rather than a trap.

• Some add-on revenue moves to Startup

  Startup is the higher plan; track total revenue per org.

• Your websitePricing: Free $0 (3 inboxes, 3,000 emails a month, 100 a day, Discord support); Developer $20 a month (10 inboxes, 10,000 emails a month, removes the "Sent via AgentMail" footer, no SOC 2 report); Startup $200 a month (150 inboxes, 15,000 emails a day, SOC 2 report); Enterprise custom (white-label platform, EU region, BYO cloud, dedicated IPs, SSO, SLA). Add-ons on Developer and Startup: $2 a month per inbox, per domain or per 1,000 emails, and $5 per seat. "Early-stage startups get a free month on our Startup tier." No calculator and no FAQ.Source
• HiringFounding GTM & Operations Lead (posted August 10, 2026): "Build and own the GTM playbook across all channels including PLG, content, community, outbound"; "Define category narrative, positioning, and messaging so AgentMail becomes the default answer"; "Drive initiatives in pricing"; build "the dashboards that drive decisions". AgentMail has raised $6M from General Catalyst, Y Combinator, Paul Graham and founders of Ramp, Supabase and HubSpot.Source
• Reviewseesel AI's AgentMail review (written by a competing product): pros include API-first inboxes, threads, webhooks and SOC 2 Type II; cons include no built-in analytics or reporting, manual DNS setup for custom domains, a free-tier daily cap that may be too small for testing active workflows, and early pricing seen as steep for pre-revenue startups.Source

Add a proof band to /enterprise in HTML and markdown: SOC 2 Type II with a link to /security and the report request, the EU region and BYO cloud, two or three enterprise-scale stories with their numbers, and what happens after "Book a call" (who joins, how long a security review takes).

Security and scale proof answer the first questions in an enterprise review, so more visitors book and calls start further along.

• Customers' approval for new uses of their stories

  Use only what is already published on /customers.

• #1 Give agents the proof humans see: add customers, volume and SOC 2 to the markdown homepage and llms.txt

• Your website/enterprise as AI readers receive it: a feature list (unlimited inboxes, bulk discounts, white-label platform, usage-based pricing, EU region, BYO cloud, dedicated IPs, OIDC/SAML SSO) and "Book a call"; no SOC 2 mention, no customer name and no numbers.Source
• Your websiteCustomer Stories: 13 published stories, including Opendoor (closing workflow), Lanesurf (freight; "hundreds of inboxes, millions of messages"), AgentCard by Alchemy (about 70% more agentic transactions), Browser Use (every cloud session ships with its own inbox), AISDR (hundreds of inboxes in production), Balance (inbox setup from half a day to one API call), Pancake (response times from hours to minutes), Dossium (an inbox per user at signup) and Cofounder (800+ emails in week one).Source
• Your websitellms.txt opens with sign-up instructions for agents (an agent can create an account with its developer's email; a one-time code goes to the developer), then sections on pricing, comparisons, evals, use cases, integrations, "Security & Compliance" ("SOC 2 Type II compliant") and "Markdown Content Negotiation". It names no customer, no case study and no email volume.Source

Rewrite the verification email (and the matching console page) for a human who may not know their agent signed up: what the agent did, the inbox it created, what is limited until they verify, a one-click verify, and the next step (custom domain, Developer plan). Track verification rate and time to verify by signup source, and send one reminder if the code is not used within a day.

User counts have been doubling fast, and agent-led signups are a growing share; lifting the verification rate turns more of that growth into accounts a human owns and pays for.

• Reminder looks like spam

  Send only one, plainly from AgentMail, and stop after.

• #8 Tag every signup with the agent harness or builder it came from, and report activation by harness

• Your websitellms.txt opens with sign-up instructions for agents (an agent can create an account with its developer's email; a one-time code goes to the developer), then sections on pricing, comparisons, evals, use cases, integrations, "Security & Compliance" ("SOC 2 Type II compliant") and "Markdown Content Negotiation". It names no customer, no case study and no email volume.Source
• PressTechCrunch (March 10, 2026): a $6 million seed round led by General Catalyst; "tens of thousands of human users and hundreds of thousands of 'agent users'"; more than 500 B2B customers; agent inboxes can send only 10 emails a day unless a person authenticates them; new accounts are sampled for sensitive keywords and bounce rates are monitored.Source
• HiringEngineering posts: "We've doubled our user count every 2 weeks since January", core infrastructure "processes hundreds of thousands of emails per day" and "We are the sole identity provider for agents" (Founding Engineer); "We've processed hundreds of millions of emails" and "Improve everything between a developer signing up and getting something useful into production" (Founding Product Engineer, posted October 6, 2026).Source

Add a source parameter to every /build install path, plugin, skill and MCP setup, and record the self-reported client on API keys created by agents. Build a PostHog dashboard of signups, verified accounts, first inbound email and paid conversion by harness, and review it weekly.

Knowing which harnesses activate best tells the team where to invest in plugins, docs and partnerships, and which /build pages need fixing.

• Self-reported client data is noisy

  Combine with install-path tags.

• HiringGTM Engineer (posted July 13, 2026): "Own the growth data layer end to end: instrument the funnel, enrich every signup"; "Turn our best customers into ten more: lookalike modeling, automated outbound, LinkedIn plays"; "Run experiments across the PLG engine (content, integrations, backlinks, paid)". Tools named: PostHog, Clay, n8n.Source
• Your website/build pages (12, one per coding agent or app builder): each offers a one-command plugin install with the hosted MCP server and skills, a skill-only option, a builder that generates a template from a plain-language description, and "Start for free" (3 inboxes, up to 100 sends a day). The page checked has no testimonial or customer proof.Source
• Your website"Evaluating how AI agents use tools" (October 3, 2026): 25 agents (a model inside a harness) ran seven email tasks and 80% of runs finished. "They rarely failed at the work itself. They failed where the tool left them guessing." The most common stop was an inbox that looked empty (35% of failed runs; agents read a filtered list as empty). On reply-all with BCC, success fell from 81% to 62% when docs were linked, because the docs pointed to an endpoint that cannot add BCC. Overall, linking docs raised success from 76% to 84%. The report lists planned fixes.Source

Tag every Discord and email support thread by cause, rank the top 20, and for each ship one fix: a docs paragraph, a clearer error message with the next step, an SDK default or a product change. Publish a short weekly "fixed this week" note in Discord and the changelog.

Fixing causes rather than answering tickets cuts support load and failed integrations together, and public fix notes build trust in the community that drives growth.

• Tagging slips when busy

  Make the cause tag required to close a thread.

• HiringDeveloper Success Engineer (posted October 12, 2026): "Your job is to answer the question, figure out why it had to be asked, and make sure we learn from it"; debug API requests, webhooks and integrations, reproduce issues or open fix PRs; "Notice recurring questions and friction across support conversations and surface those patterns".Source
• HiringAgent Experience Engineer (posted August 10, 2026): "Our most important user is an agent: it discovers us through model answers and docs." The role will "measure share of voice on the prompts that matter and move it", run evals of how models and agent harnesses use AgentMail, own llms.txt, MCP descriptions, docs and "the error messages agents hit at 3am with no human watching", and do "backlink outreach to the pages models already cite". "A stray Reddit thread once convinced LLMs we have a feature that doesn't exist."Source
• Your website"Evaluating how AI agents use tools" (October 3, 2026): 25 agents (a model inside a harness) ran seven email tasks and 80% of runs finished. "They rarely failed at the work itself. They failed where the tool left them guessing." The most common stop was an inbox that looked empty (35% of failed runs; agents read a filtered list as empty). On reply-all with BCC, success fell from 81% to 62% when docs were linked, because the docs pointed to an endpoint that cannot add BCC. Overall, linking docs raised success from 76% to 84%. The report lists planned fixes.Source

Fix a list of 30 prompts builders actually ask ("email API for AI agents", "give my agent an inbox", "agent email with webhooks in Python"), run them monthly across assistants and coding agents, and log who is recommended, which pages are cited and any wrong claims. For each wrong claim, fix it at the cited source: AgentMail's own page, a docs answer, or a polite correction on the thread or article.

What gets measured gets moved: the log shows which pages to strengthen and which outside pages to earn links from, and catches false claims before they spread.

• Answers vary run to run

  Run each prompt several times and track averages.

• #1 Give agents the proof humans see: add customers, volume and SOC 2 to the markdown homepage and llms.txt

• HiringAgent Experience Engineer (posted August 10, 2026): "Our most important user is an agent: it discovers us through model answers and docs." The role will "measure share of voice on the prompts that matter and move it", run evals of how models and agent harnesses use AgentMail, own llms.txt, MCP descriptions, docs and "the error messages agents hit at 3am with no human watching", and do "backlink outreach to the pages models already cite". "A stray Reddit thread once convinced LLMs we have a feature that doesn't exist."Source
• Your website/comparisons lists 23 articles, all blog posts and none with a summary table: head-to-heads with Resend, Nylas, SendGrid, Postmark, Mailgun, the Gmail API, Amazon SES and Cloudflare Email Service ("launched Email Service for agents in April 2026"), plus roundups. None compares AgentMail with other inbox-for-agents products.Source
• Your websiteA batch of /blog/info/ pages dated November 9, 2026, with overlapping titles: "Can an AI Agent Provision Its Own Inbox on Demand", "How to Let an AI Agent Provision Its Own Inbox on Demand", "Giving an Agent a New Inbox Mid Workflow on Demand", "On Demand Inbox Creation: How Teams Give Agents a New Address Mid Workflow", "4 Ways to Give an AI Agent an Inbox It Provisions Itself", "Tools That Let an Agent Create New Email Inboxes Programmatically" and "How to Create New Email Inboxes Programmatically for your Agents". The pages show no author or date, and "Best Agent Email Inbox Providers for Developers" (about 1,300 words) names no other provider.Source

Link the footer to a short landing page for recipients: what an agent inbox is, why the email they got came from one, a 60-second example and "Give your agent an inbox". Tag the link so signups from footers are counted, and A/B two footer wordings. Keep it short and honest so recipients trust the sender.

It is the same loop that grew early email products: every message is a small ad, seen by people already in the category, at no cost.

• A promotional footer hurts deliverability or sender trust

  Keep it one plain line and monitor complaints.

• Your websitePricing: Free $0 (3 inboxes, 3,000 emails a month, 100 a day, Discord support); Developer $20 a month (10 inboxes, 10,000 emails a month, removes the "Sent via AgentMail" footer, no SOC 2 report); Startup $200 a month (150 inboxes, 15,000 emails a day, SOC 2 report); Enterprise custom (white-label platform, EU region, BYO cloud, dedicated IPs, SSO, SLA). Add-ons on Developer and Startup: $2 a month per inbox, per domain or per 1,000 emails, and $5 per seat. "Early-stage startups get a free month on our Startup tier." No calculator and no FAQ.Source
• HiringEngineering posts: "We've doubled our user count every 2 weeks since January", core infrastructure "processes hundreds of thousands of emails per day" and "We are the sole identity provider for agents" (Founding Engineer); "We've processed hundreds of millions of emails" and "Improve everything between a developer signing up and getting something useful into production" (Founding Product Engineer, posted October 6, 2026).Source
• HiringFounding PLG & Community Lead (posted August 10, 2026): "Most of our growth has come from community." "Reddit threads, Discord, X, word of mouth, and builders shipping in public are what got us here." The lead owns the PLG budget, Discord and Reddit presence, LinkedIn and X posting, "PostHog and experimentation: understanding what our signups actually do", and "making our SF office a gathering place for the agent ecosystem".Source

Publish a directory of apps that accept "Sign in with AgentID", with a badge apps can show. Get an AgentID guide into each auth provider's docs or integration catalogue, and offer a one-prompt integration for coding agents. Show the count of apps and weekly sign-ins on /agentid.

Identity grows by network effects: every app that accepts AgentID makes an AgentMail inbox more useful, and every agent with one makes AgentID more worth adding.

• Small numbers look weak

  Show the directory first and add counts once they are meaningful.

• Your websiteAgentID is an OpenID Connect provider: an agent signs into apps with its verified AgentMail address, and registered apps also receive the email of the human who owns the agent. It is free for apps, with setup guides for Better Auth, Clerk, Supabase, Auth0, Descope and Auth.js. The page shows no adoption figures and no list of apps that accept it.Source
• Researchnpm: the agentmail package has 301,342 weekly downloads, @agentmail/agentid-better-auth 5,461 and @agentmail/cli 59. GitHub: 40 public repos; the most-starred, agentmail-toolkit, has 111 stars, then agentmail-python 75, agentmail-mcp 72 and agentmail-examples 44.Source
• Your websitePress page: "1 billion emails" launch (October 21, 2026, 201K views on X), AgentID launch (October 6, 2026, 510K views), seed announcement (237.6K views), Agent Armor showcase and Domain-connect release (October 28, 2026), releases of allowlists and blocklists, permission-scoped API keys and labels, and appearances on TBPN and This Week in Startups.Source

Map each use-case page to its closest published story and add a proof box: the customer, one number and a link. Where no story fits (for example recruiting or education), mark it as the next story to collect. Add the same box to the markdown version agents read.

A specific peer result on the page a builder already chose converts better than generic claims, and gives answer engines a concrete example to cite.

• Forced matches feel irrelevant

  Leave a page without a box rather than use a poor match.

• Your website23 /use-case pages by job or industry, such as recruiting, customer support, sales, legal, accounting, ecommerce, education, n8n, lead follow-up and "AI Agent Fleet That Needs Thousands of Inboxes".Source
• Your websiteCustomer Stories: 13 published stories, including Opendoor (closing workflow), Lanesurf (freight; "hundreds of inboxes, millions of messages"), AgentCard by Alchemy (about 70% more agentic transactions), Browser Use (every cloud session ships with its own inbox), AISDR (hundreds of inboxes in production), Balance (inbox setup from half a day to one API call), Pancake (response times from hours to minutes), Dossium (an inbox per user at signup) and Cofounder (800+ emails in week one).Source

Score current customers by inboxes per account and growth, describe the top pattern (AI products that give each user, session or agent an identity), and build a Clay workflow that finds companies with that pattern, using public signals such as launches of agent features. Send short, technical outreach that links the closest published story and a working example, and route replies to a founder.

Inbox-per-user products grow inbox counts with their own users, so each win compounds; lookalike targeting keeps outbound small and relevant.

• Outbound feels spammy to developers

  Keep volumes small and lead with a working example.

• #8 Tag every signup with the agent harness or builder it came from, and report activation by harness

• HiringGTM Engineer (posted July 13, 2026): "Own the growth data layer end to end: instrument the funnel, enrich every signup"; "Turn our best customers into ten more: lookalike modeling, automated outbound, LinkedIn plays"; "Run experiments across the PLG engine (content, integrations, backlinks, paid)". Tools named: PostHog, Clay, n8n.Source
• Your websiteCustomer Stories: 13 published stories, including Opendoor (closing workflow), Lanesurf (freight; "hundreds of inboxes, millions of messages"), AgentCard by Alchemy (about 70% more agentic transactions), Browser Use (every cloud session ships with its own inbox), AISDR (hundreds of inboxes in production), Balance (inbox setup from half a day to one API call), Pancake (response times from hours to minutes), Dossium (an inbox per user at signup) and Cofounder (800+ emails in week one).Source
• PressTechCrunch (March 10, 2026): a $6 million seed round led by General Catalyst; "tens of thousands of human users and hundreds of thousands of 'agent users'"; more than 500 B2B customers; agent inboxes can send only 10 emails a day unless a person authenticates them; new accounts are sampled for sensitive keywords and bounce rates are monitored.Source

Launch a gallery page of projects built on AgentMail, submitted from Discord or by tagging AgentMail on X, each with a short description, the harness used and a repo link. Feature one project a week on X and LinkedIn, and add a one-line "star the SDK if this helped" to the gallery, the README and the quickstart's last step.

Showcasing builders rewards the behaviour that drove growth, gives new users ideas, and turns quiet usage into visible social proof.

• Low-quality submissions

  Review before publishing and feature only working projects.

• HiringFounding PLG & Community Lead (posted August 10, 2026): "Most of our growth has come from community." "Reddit threads, Discord, X, word of mouth, and builders shipping in public are what got us here." The lead owns the PLG budget, Discord and Reddit presence, LinkedIn and X posting, "PostHog and experimentation: understanding what our signups actually do", and "making our SF office a gathering place for the agent ecosystem".Source
• Researchnpm: the agentmail package has 301,342 weekly downloads, @agentmail/agentid-better-auth 5,461 and @agentmail/cli 59. GitHub: 40 public repos; the most-starred, agentmail-toolkit, has 111 stars, then agentmail-python 75, agentmail-mcp 72 and agentmail-examples 44.Source

Commit to a quarterly rerun with the same tasks plus new ones, a public leaderboard by harness on /evals, a short "what changed" summary, and an open task set others can run. Invite harness teams to submit improved configurations before each release, and pitch the results to the podcasts that already covered AgentMail.

Benchmarks earn links and repeat coverage, and they position AgentMail as the authority on how agents use email.

• Seen as biased because AgentMail runs it

  Open the tasks and grading, and keep an independent co-runner.

• #4 Close the two doc gaps your own eval found: say when a list is filtered, and say what reply-all cannot do

• Your website"Evaluating how AI agents use tools" (October 3, 2026): 25 agents (a model inside a harness) ran seven email tasks and 80% of runs finished. "They rarely failed at the work itself. They failed where the tool left them guessing." The most common stop was an inbox that looked empty (35% of failed runs; agents read a filtered list as empty). On reply-all with BCC, success fell from 81% to 62% when docs were linked, because the docs pointed to an endpoint that cannot add BCC. Overall, linking docs raised success from 76% to 84%. The report lists planned fixes.Source
• HiringAgent Experience Engineer (posted August 10, 2026): "Our most important user is an agent: it discovers us through model answers and docs." The role will "measure share of voice on the prompts that matter and move it", run evals of how models and agent harnesses use AgentMail, own llms.txt, MCP descriptions, docs and "the error messages agents hit at 3am with no human watching", and do "backlink outreach to the pages models already cite". "A stray Reddit thread once convinced LLMs we have a feature that doesn't exist."Source
• Your websitePress page: "1 billion emails" launch (October 21, 2026, 201K views on X), AgentID launch (October 6, 2026, 510K views), seed announcement (237.6K views), Agent Armor showcase and Domain-connect release (October 28, 2026), releases of allowlists and blocklists, permission-scoped API keys and labels, and appearances on TBPN and This Week in Startups.Source

Make every limit response say which limit was hit, when it resets, and the exact add-on or plan that removes it, in a form an agent can relay. At the same time, email the account owner once a day at most with what was blocked and a one-click upgrade, and show a banner in the console. Warn at 80% of a limit so builders can upgrade before anything breaks.

Turning a dead end into a clear next step recovers blocked work and converts usage spikes into paid plans at the moment of highest intent.

• Upgrade prompts look pushy

  Lead with what happened and when it resets; the upgrade is the last line.

• #5 Show the Developer-to-Startup crossover on the pricing page: at 100 inboxes, add-ons cost the same as Startup

• Your websitePricing: Free $0 (3 inboxes, 3,000 emails a month, 100 a day, Discord support); Developer $20 a month (10 inboxes, 10,000 emails a month, removes the "Sent via AgentMail" footer, no SOC 2 report); Startup $200 a month (150 inboxes, 15,000 emails a day, SOC 2 report); Enterprise custom (white-label platform, EU region, BYO cloud, dedicated IPs, SSO, SLA). Add-ons on Developer and Startup: $2 a month per inbox, per domain or per 1,000 emails, and $5 per seat. "Early-stage startups get a free month on our Startup tier." No calculator and no FAQ.Source
• HiringAgent Experience Engineer (posted August 10, 2026): "Our most important user is an agent: it discovers us through model answers and docs." The role will "measure share of voice on the prompts that matter and move it", run evals of how models and agent harnesses use AgentMail, own llms.txt, MCP descriptions, docs and "the error messages agents hit at 3am with no human watching", and do "backlink outreach to the pages models already cite". "A stray Reddit thread once convinced LLMs we have a feature that doesn't exist."Source
• Reviewseesel AI's AgentMail review (written by a competing product): pros include API-first inboxes, threads, webhooks and SOC 2 Type II; cons include no built-in analytics or reporting, manual DNS setup for custom domains, a free-tier daily cap that may be too small for testing active workflows, and early pricing seen as steep for pre-revenue startups.Source

Write one page that explains how AgentMail keeps an agent inbox safe: who can email it (allowlists), what an agent can do (scoped keys), how risky inbound content is handled (Agent Armor), and the audit trail. Link it from /enterprise, /security, the comparison table and llms.txt, and give sales a one-page PDF version.

Safety is the objection that blocks agent email in larger companies; one clear answer shortens reviews and counters rivals' safety pitch.

• Overclaiming protection

  Describe controls precisely and state what remains the builder's job.

• #6 Put SOC 2, EU region and named customers on /enterprise for the buyers who arrive there

• Competitor researchRivals' own homepages: Mails.ai ("Give your agent its own inbox.") leads with prompt-injection scoring and quarantine on inbound mail, a test address that auto-replies, sign-in code and magic-link extraction, and per-agent reputation; Atomic Mail promotes "Atomic Mail for Agents" beside its encrypted consumer email. Send-first APIs (Resend, SendGrid, Mailgun, Postmark) are adjacent.See the research Source
• Your websitePress page: "1 billion emails" launch (October 21, 2026, 201K views on X), AgentID launch (October 6, 2026, 510K views), seed announcement (237.6K views), Agent Armor showcase and Domain-connect release (October 28, 2026), releases of allowlists and blocklists, permission-scoped API keys and labels, and appearances on TBPN and This Week in Startups.Source
• Your website/enterprise as AI readers receive it: a feature list (unlimited inboxes, bulk discounts, white-label platform, usage-based pricing, EU region, BYO cloud, dedicated IPs, OIDC/SAML SSO) and "Book a call"; no SOC 2 mention, no customer name and no numbers.Source

Package the offer as a perk (free Startup month plus a setup call) and list it in YC's and General Catalyst's founder perk lists and in other accelerators' deal books. Give each programme its own signup link to track conversion, and follow up with founders at day 20 about plan fit.

Accelerator cohorts are dense with agent startups at the moment they pick infrastructure, and perks lists are trusted, low-cost distribution.

• Free months that never convert

  Require a work email and review conversion by programme.

• Your websitePricing: Free $0 (3 inboxes, 3,000 emails a month, 100 a day, Discord support); Developer $20 a month (10 inboxes, 10,000 emails a month, removes the "Sent via AgentMail" footer, no SOC 2 report); Startup $200 a month (150 inboxes, 15,000 emails a day, SOC 2 report); Enterprise custom (white-label platform, EU region, BYO cloud, dedicated IPs, SSO, SLA). Add-ons on Developer and Startup: $2 a month per inbox, per domain or per 1,000 emails, and $5 per seat. "Early-stage startups get a free month on our Startup tier." No calculator and no FAQ.Source
• HiringFounding GTM & Operations Lead (posted August 10, 2026): "Build and own the GTM playbook across all channels including PLG, content, community, outbound"; "Define category narrative, positioning, and messaging so AgentMail becomes the default answer"; "Drive initiatives in pricing"; build "the dashboards that drive decisions". AgentMail has raised $6M from General Catalyst, Y Combinator, Paul Graham and founders of Ramp, Supabase and HubSpot.Source
• Reviewseesel AI's AgentMail review (written by a competing product): pros include API-first inboxes, threads, webhooks and SOC 2 Type II; cons include no built-in analytics or reporting, manual DNS setup for custom domains, a free-tier daily cap that may be too small for testing active workflows, and early pricing seen as steep for pre-revenue startups.Source

Treat AgentID as a second product with its own goal: the number of apps that accept it and weekly agent sign-ins. Fund integrations with the major auth providers and frameworks, publish an open spec for agent sign-in claims (including the human owner), and give apps a reason to adopt: lower fraud from verified agents and a clear human owner on every sign-in.

Every new app that accepts AgentID raises the value of every AgentMail inbox, which pulls more agents onto AgentMail: a growth loop rather than a channel.

• A larger platform launches a competing standard

  Publish the spec openly and aim to be the reference implementation.

• #12 List the apps that accept AgentID, and ship it inside each auth provider's docs

• Your websiteAgentID is an OpenID Connect provider: an agent signs into apps with its verified AgentMail address, and registered apps also receive the email of the human who owns the agent. It is free for apps, with setup guides for Better Auth, Clerk, Supabase, Auth0, Descope and Auth.js. The page shows no adoption figures and no list of apps that accept it.Source
• HiringEngineering posts: "We've doubled our user count every 2 weeks since January", core infrastructure "processes hundreds of thousands of emails per day" and "We are the sole identity provider for agents" (Founding Engineer); "We've processed hundreds of millions of emails" and "Improve everything between a developer signing up and getting something useful into production" (Founding Product Engineer, posted October 6, 2026).Source
• Your websitePress page: "1 billion emails" launch (October 21, 2026, 201K views on X), AgentID launch (October 6, 2026, 510K views), seed announcement (237.6K views), Agent Armor showcase and Domain-connect release (October 28, 2026), releases of allowlists and blocklists, permission-scoped API keys and labels, and appearances on TBPN and This Week in Startups.Source

Design a Platform tier priced per active inbox with volume bands, Pods for per-customer isolation, white-label domains and a usage dashboard, and publish starting prices. Write a short guide for embedding AgentMail inboxes in a product, with the published stories as examples.

Inbox-per-user customers grow with their own users, so a clear, self-serve price for them turns AgentMail's fastest-growing accounts into predictable expansion revenue.

• Lower prices for existing large accounts

  Price on active inboxes and pilot with existing customers first.

• #14 Find the products that look like your inbox-per-user customers, and reach them when they launch agents

• Your websitePricing: Free $0 (3 inboxes, 3,000 emails a month, 100 a day, Discord support); Developer $20 a month (10 inboxes, 10,000 emails a month, removes the "Sent via AgentMail" footer, no SOC 2 report); Startup $200 a month (150 inboxes, 15,000 emails a day, SOC 2 report); Enterprise custom (white-label platform, EU region, BYO cloud, dedicated IPs, SSO, SLA). Add-ons on Developer and Startup: $2 a month per inbox, per domain or per 1,000 emails, and $5 per seat. "Early-stage startups get a free month on our Startup tier." No calculator and no FAQ.Source
• Your websiteCustomer Stories: 13 published stories, including Opendoor (closing workflow), Lanesurf (freight; "hundreds of inboxes, millions of messages"), AgentCard by Alchemy (about 70% more agentic transactions), Browser Use (every cloud session ships with its own inbox), AISDR (hundreds of inboxes in production), Balance (inbox setup from half a day to one API call), Pancake (response times from hours to minutes), Dossium (an inbox per user at signup) and Cofounder (800+ emails in week one).Source
• Your website23 /use-case pages by job or industry, such as recruiting, customer support, sales, legal, accounting, ecommerce, education, n8n, lead follow-up and "AI Agent Fleet That Needs Thousands of Inboxes".Source

Publish a quarterly report with aggregate, anonymised trends: volume by use case, reply times, agent-to-agent versus agent-to-human share, inbound safety events and identity usage. Pair it with a founder thread, a podcast round and a dataset page that answer engines can cite.

Original data is the most cited kind of content; it puts AgentMail at the centre of the category story every quarter.

• Customers worry about their data

  Publish only aggregates above strict thresholds and state the rules.

• Your websitePress page: "1 billion emails" launch (October 21, 2026, 201K views on X), AgentID launch (October 6, 2026, 510K views), seed announcement (237.6K views), Agent Armor showcase and Domain-connect release (October 28, 2026), releases of allowlists and blocklists, permission-scoped API keys and labels, and appearances on TBPN and This Week in Startups.Source
• HiringEngineering posts: "We've doubled our user count every 2 weeks since January", core infrastructure "processes hundreds of thousands of emails per day" and "We are the sole identity provider for agents" (Founding Engineer); "We've processed hundreds of millions of emails" and "Improve everything between a developer signing up and getting something useful into production" (Founding Product Engineer, posted October 6, 2026).Source
• PressTechCrunch (March 10, 2026): a $6 million seed round led by General Catalyst; "tens of thousands of human users and hundreds of thousands of 'agent users'"; more than 500 B2B customers; agent inboxes can send only 10 emails a day unless a person authenticates them; new accounts are sampled for sensitive keywords and bounce rates are monitored.Source

Host a monthly build night with a theme (an agent that books meetings, a support agent, an agent that signs up for apps with AgentID). Every team ships with an AgentMail inbox, demos are streamed on X, and the best projects go into the gallery and the next week's posts. Invite harness teams and customers to judge.

It turns a cost line (events) into a content and signup engine, and makes the office the place where the category meets.

• Low turnout

  Co-host with a harness team or customer that brings its own audience.

• #15 A public "built with AgentMail" gallery fed by Discord and X, with a star ask on every entry

• HiringChief of Staff (posted October 11, 2026): owns events, "developer dinners, hackathons", candidate experience and the things "that fall between functions".Source
• HiringFounding PLG & Community Lead (posted August 10, 2026): "Most of our growth has come from community." "Reddit threads, Discord, X, word of mouth, and builders shipping in public are what got us here." The lead owns the PLG budget, Discord and Reddit presence, LinkedIn and X posting, "PostHog and experimentation: understanding what our signups actually do", and "making our SF office a gathering place for the agent ecosystem".Source
• Your websitePress page: "1 billion emails" launch (October 21, 2026, 201K views on X), AgentID launch (October 6, 2026, 510K views), seed announcement (237.6K views), Agent Armor showcase and Domain-connect release (October 28, 2026), releases of allowlists and blocklists, permission-scoped API keys and labels, and appearances on TBPN and This Week in Startups.Source

Publish a plain-language page for recipients: what an AgentMail address means, how agents are tied to a responsible human, how to report unwanted email, and what AgentMail does next. Link it from the free-tier footer and the security page, and report how many reports were handled each quarter.

Visible accountability protects deliverability, which is what every paying customer relies on, and turns a category worry into a trust advantage.

• The page draws attention to abuse

  Frame it as accountability and show how few reports there are.

• #11 Make the free-tier "Sent via AgentMail" footer a tracked door for the people who receive agent email

• PressTechCrunch (March 10, 2026): a $6 million seed round led by General Catalyst; "tens of thousands of human users and hundreds of thousands of 'agent users'"; more than 500 B2B customers; agent inboxes can send only 10 emails a day unless a person authenticates them; new accounts are sampled for sensitive keywords and bounce rates are monitored.Source
• Reviewseesel AI's AgentMail review (written by a competing product): pros include API-first inboxes, threads, webhooks and SOC 2 Type II; cons include no built-in analytics or reporting, manual DNS setup for custom domains, a free-tier daily cap that may be too small for testing active workflows, and early pricing seen as steep for pre-revenue startups.Source
• Your websitePricing: Free $0 (3 inboxes, 3,000 emails a month, 100 a day, Discord support); Developer $20 a month (10 inboxes, 10,000 emails a month, removes the "Sent via AgentMail" footer, no SOC 2 report); Startup $200 a month (150 inboxes, 15,000 emails a day, SOC 2 report); Enterprise custom (white-label platform, EU region, BYO cloud, dedicated IPs, SSO, SLA). Add-ons on Developer and Startup: $2 a month per inbox, per domain or per 1,000 emails, and $5 per seat. "Early-stage startups get a free month on our Startup tier." No calculator and no FAQ.Source