Superhuman Mail vs AgentMail

Portability

1. switcherMy mail stays reachable over open protocols and standard formats — IMAP/JMAP/SMTP access and standards-based export — so no client can lock me in

   weight 2 · round to AgentMail
   Superhuman Mail—none0/10

   Superhuman is a proprietary client layered on top of Gmail/Outlook accounts; the evidence pack contains no mention of IMAP/JMAP/SMTP access, standards-based export, or any data-portability guarantee — only MCP integration for AI assistants and account switching between Gmail/Outlook. No documentation addresses protocol-level openness or export, so switchers have no evidence their mail stays reachable outside Superhuman's own app.

   ⚑ flag

   AgentMail~partialclaimed6/10

   AgentMail documents IMAP/SMTP compatibility, letting any standard mail client (Outlook, Thunderbird, Apple Mail) connect to an AgentMail inbox, which supports the open-protocol reachability part of the story. However, there is no mention of JMAP support or of a standards-based export/data-portability mechanism for migrating mail elsewhere. Missing for 10: JMAP protocol support, documented mailbox/data export or migration tooling, and independent confirmation that IMAP/SMTP access works reliably in practice.

   • [claimed-docs] “Connect Outlook, Thunderbird, Apple Mail, or any IMAP/SMTP-compatible client to your AgentMail inbox”

   proof ↗⚑ flag

2. switcherSelf-host the email app and keep the mail pipeline under my own control

   weight 1 · round drawn
   Superhuman Mail—none0/10

   Superhuman Mail is a cloud SaaS client layered on top of Gmail/Outlook accounts; there is no evidence of any self-hosted deployment option, open-source server, or ability to run the mail pipeline on infrastructure the user controls. All evidence points to a hosted, proprietary service with no self-host path, so this axis is not delivered.

   ⚑ flag

   AgentMail—none0/10

   AgentMail is documented as a hosted email API/service (custom domains, hosted MCP server, webhooks) with no mention anywhere of a self-hosted deployment option; a competitor's marketing even highlights itself as the 'self-hosted drop-in replacement' for AgentMail, implying AgentMail itself is cloud-only.

   • [claimed-docs] “using your own custom domains is essential for production applications. It gives you control over your sending reputation, which is the sing…”
   • [claimed-docs] “AgentMail runs a hosted MCP server at https://mcp.agentmail.to/mcp that exposes inbox, message, thread, draft, and attachment tools to any c…”
   • [community] “agentmail charges $20/mo for 10 inboxes. autopilot gives you unlimited inboxes for the cost of a postgres row, positioning itself as a self-…”

   proof ↗⚑ flag

Agent access

1. ai-native userPoint an agent at llms.txt or agent-oriented docs

   weight 2 · round to AgentMail
   Superhuman Mail—none0/10

   The probe explicitly shows llms.txt returns 404 and no agent-oriented docs (openapi/well-known) are found; while an MCP server exists, there's no evidence of llms.txt or similar agent-discovery docs a user could point an agent at.

   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”

   proof ↗⚑ flag

   AgentMail✓fullprobed9/10

   AgentMail hosts a live llms.txt file (verified HTTP 200) with a clear agent-oriented summary of its docs, and its entire documentation set is written in agent-consumable .md format (e.g., quickstart.md, messages.md) alongside an OpenAPI spec, MCP server, and CLI docs, making it easy for an agent to be pointed at and consume its docs. Missing for 10: no independent/community confirmation of an agent successfully navigating llms.txt end-to-end.

   • [probe] “PROBE llms.txt: HTTP 200 at https://docs.agentmail.to/llms.txt # AgentMail | Documentation > AgentMail is an email API built for AI agents.…”
   • [probe] “PROBE openapi: HTTP 200 at https://docs.agentmail.to/openapi.json — contains "openapi" key”
   • [claimed-docs] “The agent registers itself using the human email you provide and gets back an API key, inbox ID, and organization ID.”
   • [probe] “official MCP server documented at https://docs.agentmail.to/integrations/mcp”

   proof ↗⚑ flag

2. ai-native userRun the product headlessly / in CI for automation

   weight 2 · round to AgentMail
   Superhuman Mail—none0/10

   Superhuman Mail is a GUI email client with an MCP server and agent triggers, but there is no evidence of a CLI, API, or documented mode to run the product headlessly or invoke it from a CI pipeline — probes for llms.txt and OpenAPI specs both returned 404. The story is plausible for an AI-native product with automation features, so it is not out of scope, but no evidence supports the specific headless/CI capability.

   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”
   • [claimed-docs] “Use chat or our visual builder to create agents that run proactively based on triggers you set and the work you connect.”

   proof ↗⚑ flag

   AgentMail✓fullprobed7/10

   AgentMail is fundamentally API/SDK-driven ('Create, manage, and operate email inboxes entirely via API'), with a documented OpenAPI spec, official CLI, and webhook/websocket event delivery that support scripted, non-interactive automation pipelines. No explicit CI/CD example or pipeline walkthrough is given, so full CI-specific validation is inferred rather than directly documented. Missing for 10: an explicit CI/CD pipeline example or GitHub Actions integration, and independent confirmation of headless use in production automation.

   • [claimed-docs] “Create, manage, and operate email inboxes entirely via API.”
   • [claimed-docs] “Create inboxes, send messages, manage threads, and more — all without writing code.”
   • [claimed-docs] “Instead of constantly asking the AgentMail API if there's a new email (a process called polling), you can register a URL, and we will send y…”
   • [claimed-docs] “you can register a URL, and we will send you a `POST` request with the details as soon as an event happens.”
   • [probe] “PROBE openapi: HTTP 200 at https://docs.agentmail.to/openapi.json — contains "openapi" key”
   • [probe] “official CLI documented at https://docs.agentmail.to/integrations/cli”

   proof ↗⚑ flag

3. ai-native userConnect an agent via an official MCP server

   weight 3 · round drawn
   Superhuman Mail✓fullprobed8/10

   Superhuman documents an official Mail MCP server that connects Claude, ChatGPT, and other AI assistants to the inbox/calendar for summarizing, drafting, sending, and scheduling, with explicit send-approval controls, confirmed by a dedicated probe finding the official MCP page. Missing for 10: independent/hands-on third-party verification of the MCP server working (community evidence pack lacks any first-hand MCP usage reports) and technical details on the MCP server's implementation/spec compliance.

   • [claimed-docs] “Superhuman Mail MCP connects your inbox and calendar into Claude, ChatGPT, and other AI assistants—so you can summarize threads, draft and s…”
   • [claimed-docs] “Connect Claude or ChatGPT directly to your inbox to read, draft, and send emails with a single prompt — no copy-pasting required.”
   • [claimed-docs] “You can now drive Superhuman Mail from Claude, ChatGPT, and your favorite AI tools. You can create powerful workflows that run across your i…”
   • [claimed-docs] “Every send action requires your explicit approval, so you stay in control.”
   • [claimed-docs] “Connect all your Gmail and Outlook accounts, including work, personal, and side projects. Manage every inbox and calendar from one AI assist…”
   • [probe] “official MCP server documented at https://superhuman.com/mail/features/email-mcp”

   proof ↗⚑ flag

   AgentMail✓fullprobed8/10

   AgentMail is a service/API product (not itself an agent), so the MCP-server axis applies, and docs confirm a hosted official MCP server at https://mcp.agentmail.to/mcp exposing inbox, message, thread, draft, and attachment tools to any compatible client, corroborated by a probe hit. Missing for 10: independent/hands-on confirmation that the MCP server works reliably in practice beyond first-party docs.

   • [claimed-docs] “AgentMail runs a hosted MCP server at https://mcp.agentmail.to/mcp that exposes inbox, message, thread, draft, and attachment tools to any c…”
   • [claimed-docs] “AgentMail runs a hosted MCP server at `https://mcp.agentmail.to/mcp` that exposes inbox, message, thread, draft, and attachment tools to any…”
   • [probe] “official MCP server documented at https://docs.agentmail.to/integrations/mcp”

   proof ↗⚑ flag

4. ai-native userUse an official CLI

   weight 2 · round to AgentMail
   Superhuman Mail—none0/10

   Superhuman Mail is an email client with an MCP integration and command palette (Cmd+K), but there is no evidence of an official CLI for AI-native workflows; probes for llms.txt/openapi also return 404, and no CLI is mentioned anywhere in docs.

   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”
   • [claimed-docs] “Cmd+K brings up Superhuman Command, where you can instantly do any action — and also learn the shortcut for next time.”
   • [claimed-docs] “In Superhuman, one shortcut rules them all: Cmd+K. Cmd+K brings up Superhuman Command, where you can instantly do any action”

   proof ↗⚑ flag

   AgentMail✓fullprobed7/10

   AgentMail documents an official CLI (docs.agentmail.to/integrations/cli) that lets users create inboxes, send messages, and manage threads without writing code, corroborated by a probe confirming the page exists. Missing for 10: no independent/community hands-on confirmation of CLI usage or installation details beyond the docs themselves.

   • [claimed-docs] “Create inboxes, send messages, manage threads, and more — all without writing code.”
   • [probe] “official CLI documented at https://docs.agentmail.to/integrations/cli”

   proof ↗⚑ flag

5. ai-native userDrive the product through a documented public API

   weight 3 · round to AgentMail
   Superhuman Mail~partialprobed6/10

   Superhuman ships a documented, public MCP server (email-mcp) that lets AI-native users drive inbox/calendar actions—summarize, draft, send, schedule—directly from Claude, ChatGPT and other AI tools, which is a real programmatic surface for agentic use. However, probes confirm there is no traditional REST/OpenAPI-documented public API (llms.txt and openapi.json endpoints all 404), so developers outside the MCP/AI-assistant flow have no documented way to integrate. Missing for 10: a general-purpose REST/GraphQL API with OpenAPI spec, and any independent/hands-on verification of the MCP server's reliability.

   • [claimed-docs] “Superhuman Mail MCP connects your inbox and calendar into Claude, ChatGPT, and other AI assistants—so you can summarize threads, draft and s…”
   • [claimed-docs] “You can now drive Superhuman Mail from Claude, ChatGPT, and your favorite AI tools. You can create powerful workflows that run across your i…”
   • [claimed-docs] “Every send action requires your explicit approval, so you stay in control.”
   • [claimed-docs] “Every draft matches your voice and tone, using your unique style for each person. You can review the drafts in Superhuman Mail, or edit and …”
   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”
   • [probe] “official MCP server documented at https://superhuman.com/mail/features/email-mcp”

   proof ↗⚑ flag

   AgentMail✓fullprobed9/10

   AgentMail is explicitly an API-first product: docs describe registering agents via API, creating inboxes, sending/searching messages, webhooks, permissions, etc., and a live OpenAPI spec is confirmed at docs.agentmail.to/openapi.json. This gives strong first-party and probe-verified evidence of a documented, comprehensive public API. Missing for 10: independent third-party hands-on validation of the API itself (community feedback focuses on website/pricing, not API usage).

   • [claimed-docs] “The agent registers itself using the human email you provide and gets back an API key, inbox ID, and organization ID.”
   • [claimed-docs] “Create, manage, and operate email inboxes entirely via API.”
   • [probe] “PROBE openapi: HTTP 200 at https://docs.agentmail.to/openapi.json — contains "openapi" key”
   • [probe] “PROBE llms.txt: HTTP 200 at https://docs.agentmail.to/llms.txt # AgentMail | Documentation > AgentMail is an email API built for AI agents.…”
   • [claimed-docs] “Instead of constantly asking the AgentMail API if there's a new email (a process called polling), you can register a URL, and we will send y…”

   proof ↗⚑ flag

6. ai-native userIssue scoped/least-privilege API credentials for an agent

   weight 2 · round to AgentMail
   Superhuman Mail—none0/10

   Evidence shows Superhuman Mail connects to Claude/ChatGPT via MCP and includes approval gates ('every send action requires explicit approval') plus admin-level 'connector permissions' governance, but there is no evidence of an actual scoped/least-privilege API credential or key system that an ai-native user can issue to individual agents. Probes for llms.txt and OpenAPI specs both return 404, suggesting no public API surface for credential scoping exists.

   • [claimed-docs] “Every send action requires your explicit approval, so you stay in control.”
   • [claimed-docs] “Admins govern who can build agents and control connector permissions across the entire org from one place.”
   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”

   proof ↗⚑ flag

   AgentMail✓fullclaimed8/10

   Docs explicitly describe scoped API keys restricted to a single pod or inbox, per-customer pods with scoped keys for multi-tenancy, and a dedicated Permissions feature to restrict what an API key can do — directly matching the least-privilege credential story. Missing for 10: independent/hands-on verification that scoping actually enforces restrictions in practice, and detail on granularity of permission types beyond pod/inbox scope.

   • [claimed-docs] “create a Pod per customer, give them a scoped API key, and route webhook events to the right place.”
   • [claimed-docs] “Scoped API keys restrict access to a single pod or a single inbox.”
   • [claimed-docs] “`Permissions` let you restrict what an API key can do.”
   • [claimed-docs] “Create one pod per customer/tenant in your system”

   proof ↗⚑ flag

7. ai-native userBuild against official SDKs

   weight 2 · round to AgentMail
   Superhuman Mail—none0/10

   Evidence shows Superhuman ships an MCP server for AI assistants (docs-10, docs-37, probe-3) but there is no mention of official SDKs, client libraries, or public API documentation for developers; explicit probes for llms.txt and OpenAPI/Swagger specs both returned 404 (superhuman-probe-1, superhuman-probe-2), indicating no official SDK exists.

   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”
   • [claimed-docs] “Superhuman Mail MCP connects your inbox and calendar into Claude, ChatGPT, and other AI assistants—so you can summarize threads, draft and s…”
   • [claimed-docs] “You can now drive Superhuman Mail from Claude, ChatGPT, and your favorite AI tools. You can create powerful workflows that run across your i…”

   proof ↗⚑ flag

   AgentMail✓fullprobed8/10

   AgentMail is fundamentally an API-first product with dedicated docs, OpenAPI spec, quickstart showing API key/SDK registration, and popovers showing SDK examples praised by a community reviewer, confirming real SDK-based building. Missing for 10: explicit list of language-specific SDKs (e.g. Python/TS/Go) and independent third-party developer testimonials beyond the dashboard praise.

   • [claimed-docs] “The agent registers itself using the human email you provide and gets back an API key, inbox ID, and organization ID.”
   • [claimed-docs] “Create, manage, and operate email inboxes entirely via API.”
   • [probe] “PROBE openapi: HTTP 200 at https://docs.agentmail.to/openapi.json — contains "openapi" key”
   • [community] “What I am most impressed by is the popovers in your dashboard, which show SDK examples. This is a brilliant UI idea (executed with great pol…”

   proof ↗⚑ flag

8. ai-native userSubscribe to events via webhooks

   weight 2 · round to AgentMail
   Superhuman Mail—none0/10

   Evidence shows Superhuman offers an MCP server for AI assistant integration and an agents/automation feature, but nothing describes webhook subscriptions or event-driven push notifications for external systems; API/OpenAPI probes also returned 404s, indicating no documented webhook API.

   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”
   • [claimed-docs] “Superhuman Mail MCP connects your inbox and calendar into Claude, ChatGPT, and other AI assistants—so you can summarize threads, draft and s…”

   proof ↗⚑ flag

   AgentMail✓fullclaimed9/10

   AgentMail's docs explicitly support webhook subscriptions: registering a URL to receive POST requests on email events instead of polling, avoiding the need for constant API checks. This is a first-party documented, core feature directly matching the story. Missing for 10: independent/hands-on confirmation that webhooks work reliably in practice.

   • [claimed-docs] “Instead of constantly asking the AgentMail API if there's a new email (a process called polling), you can register a URL, and we will send y…”
   • [claimed-docs] “you can register a URL, and we will send you a `POST` request with the details as soon as an event happens.”

   proof ↗⚑ flag

Agentic features

1. ai-native userSet up automations that run autonomously in the background

   weight 2 · round to Superhuman Mail
   Superhuman Mail~partialclaimed6/10

   Superhuman documents proactive automations: Auto Archive/Auto Labels/Auto Drafts run automatically in the background, and a newer 'agents' feature lets users build proactive agents via chat or visual builder with admin-governed triggers and connectors. However, sending still requires explicit human approval, limiting true autonomy, and there is no independent/hands-on corroboration of the agent-builder feature working as described. Missing for 10: third-party verification of the visual builder/agent triggers, detail on failure/error handling for autonomous runs, and evidence of unattended send/execute actions.

   • [claimed-docs] “Use chat or our visual builder to create agents that run proactively based on triggers you set and the work you connect.”
   • [claimed-docs] “Admins govern who can build agents and control connector permissions across the entire org from one place.”
   • [claimed-docs] “Superhuman AI can automatically archive marketing, cold pitches, and social network updates. Turn on Auto Archive, and eliminate inbox clutt…”
   • [claimed-docs] “Auto Drafts now gather context from your inbox, your calendar, and the web. They can use information from other emails, insert availability,…”
   • [claimed-docs] “Every send action requires your explicit approval, so you stay in control.”

   proof ↗⚑ flag

   AgentMail~partialclaimed5/10

   AgentMail provides the building blocks for background automation—webhooks and persistent WebSocket connections that push email events in real time instead of requiring polling—so an agent can react autonomously to inbound mail without a human in the loop. However, AgentMail itself is just email infrastructure; there's no evidence of a built-in scheduler, workflow engine, or persistent agent runtime that executes autonomous logic—that has to be built by the user on top of the webhook/websocket feed. Missing for 10: evidence of a native scheduling/automation engine, documented autonomous background job execution, or independent confirmation that these mechanisms reliably drive unattended agent workflows.

   • [claimed-docs] “Instead of constantly asking the AgentMail API if there's a new email (a process called polling), you can register a URL, and we will send y…”
   • [claimed-docs] “you can register a URL, and we will send you a `POST` request with the details as soon as an event happens.”
   • [claimed-docs] “WebSockets provide a persistent, bidirectional connection to AgentMail for receiving email events in real-time.”
   • [claimed-docs] “WebSockets don't require a public URL or external tools like ngrok.”

   proof ↗⚑ flag

2. ai-native userOperate the product with natural-language commands

   weight 2 · round to Superhuman Mail
   Superhuman Mail✓fullprobed8/10

   Superhuman supports natural-language operation via Ask AI queries, plain-language personalization instructions, AI-prompt-based Auto Labels, and full MCP integration letting users drive inbox/calendar actions (summarize, draft, send, schedule) via Claude/ChatGPT with a single prompt. missing for 10: independent/hands-on verification of natural-language command accuracy and no first-party demo of complex multi-step NL commands beyond simple prompts.

   • [claimed-docs] “Superhuman Mail MCP connects your inbox and calendar into Claude, ChatGPT, and other AI assistants—so you can summarize threads, draft and s…”
   • [claimed-docs] “Connect Claude or ChatGPT directly to your inbox to read, draft, and send emails with a single prompt — no copy-pasting required.”
   • [claimed-docs] “Just ask: _"where is the q2 offsite"_, _"when is my flight"_, _"where am I staying"_, and we'll immediately generate succinct answers.”
   • [claimed-docs] “Add plain-language instructions like "write in lowercase" or "use this booking link" to personalize your writing and scheduling features.”
   • [claimed-docs] “You can even make your own Auto Labels with short AI prompts like "job applications" or "requests to review work".”
   • [claimed-docs] “You can now drive Superhuman Mail from Claude, ChatGPT, and your favorite AI tools. You can create powerful workflows that run across your i…”
   • [claimed-docs] “Every send action requires your explicit approval, so you stay in control.”
   • [probe] “official MCP server documented at https://superhuman.com/mail/features/email-mcp”

   proof ↗⚑ flag

   AgentMail~partialprobed6/10

   AgentMail exposes a hosted MCP server with inbox/message/thread/draft tools that lets any MCP-compatible AI agent operate the product via natural language, and also ships a CLI for no-code operations. However, there's no first-party chat/NL interface of its own, nor independent evidence of successful natural-language operation in practice. Missing for 10: hands-on demonstration of natural-language command execution, a dedicated conversational UI beyond MCP tool-calling.

   • [claimed-docs] “AgentMail runs a hosted MCP server at https://mcp.agentmail.to/mcp that exposes inbox, message, thread, draft, and attachment tools to any c…”
   • [claimed-docs] “AgentMail runs a hosted MCP server at `https://mcp.agentmail.to/mcp` that exposes inbox, message, thread, draft, and attachment tools to any…”
   • [claimed-docs] “Create inboxes, send messages, manage threads, and more — all without writing code.”
   • [probe] “official MCP server documented at https://docs.agentmail.to/integrations/mcp”
   • [probe] “official CLI documented at https://docs.agentmail.to/integrations/cli”

   proof ↗⚑ flag

Api quality

1. ai-native userExplore an interactive API reference with runnable examples

   weight 2 · round to AgentMail
   Superhuman Mail—none0/10

   There is no evidence of an interactive API reference or runnable examples; probes explicitly confirm no OpenAPI/swagger spec and no llms.txt is available, and the MCP server page offers no interactive docs/playground.

   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”
   • [probe] “official MCP server documented at https://superhuman.com/mail/features/email-mcp”

   proof ↗⚑ flag

   AgentMail~partialprobed5/10

   AgentMail exposes a machine-readable OpenAPI spec (agentmail-probe-2) and its docs site includes language-switchable code snippets (Python/curl) plus SDK example popovers praised by users (agentmail-comm-2, agentmail-comm-7), suggesting an interactive reference. However, there is no evidence of a true 'try it now' runnable/executable API console, and independent reports note the docs site itself is slow and glitchy (agentmail-comm-1, agentmail-comm-3), undermining a smooth interactive experience. Missing for 10: confirmation of an actual runnable/sandboxed request execution feature, and resolution of the reported performance/UX issues on the docs site.

   • [probe] “PROBE openapi: HTTP 200 at https://docs.agentmail.to/openapi.json — contains "openapi" key”
   • [community] “The website is really bad. Moving from Python to curl tab on the site shows noticeable lag - I have seen 2-3 whole websites load faster in t…”
   • [community] “What I am most impressed by is the popovers in your dashboard, which show SDK examples. This is a brilliant UI idea (executed with great pol…”
   • [community] “"Application error: a client-side exception has occurred while loading www.agentmail.to"... Failed to create WebGL context... Cookie rejecte…”
   • [community] “I have written about how slow the website is... the amount of lag is something I have genuinely never witnessed in any website... I can't vi…”

   proof ↗⚑ flag

2. ai-native userDownload a machine-readable API spec (OpenAPI or equivalent)

   weight 2 · round to AgentMail
   Superhuman Mail—none0/10

   Direct probes for OpenAPI/Swagger specs and llms.txt all returned 404s, and no documentation mentions a downloadable machine-readable API spec; only an MCP server is offered, which is a separate mechanism, not an API spec.

   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”

   proof ↗⚑ flag

   AgentMail✓fullprobed9/10

   A probe confirms the OpenAPI spec is live and machine-readable at https://docs.agentmail.to/openapi.json (HTTP 200, contains "openapi" key), backed by documented API-first design. Missing for 10: no independent third-party corroboration of spec completeness or SDK auto-generation from it.

   • [probe] “PROBE openapi: HTTP 200 at https://docs.agentmail.to/openapi.json — contains "openapi" key”
   • [claimed-docs] “Create, manage, and operate email inboxes entirely via API.”

   proof ↗⚑ flag

3. ai-native userTest against a sandbox environment without touching production data

   weight 1 · round drawn
   Superhuman Mail—none0/10

   No evidence anywhere in the pack mentions a sandbox, test/staging environment, or any way to develop or validate AI workflows against non-production data — the docs only describe production-facing features (MCP, auto-drafts, auto-labels) with no mention of test environments. missing for 10: any sandbox/staging environment, test data isolation, or dev-mode API/MCP endpoint separate from live inbox data.

   ⚑ flag

   AgentMail—none0/10

   No evidence of a dedicated sandbox/test environment separate from production; the docs describe creating inboxes on-demand, but nothing indicates an isolated test mode that avoids touching real production data or sending real emails.

   ⚑ flag

4. ai-native userRely on versioned APIs with a documented deprecation policy

   weight 2 · round drawn
   Superhuman Mail—none0/10

   No evidence of any versioned public API or documented deprecation policy; probes for OpenAPI specs returned 404s, and there's no mention of API versioning or deprecation practices anywhere in the docs. The only developer-facing surface found is the MCP server, which is unrelated to API versioning guarantees. missing for 10: any public API documentation, version numbers, changelog/deprecation policy, or evidence of stable programmatic interfaces.

   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”
   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”

   proof ↗⚑ flag

   AgentMail—none0/10

   Evidence shows an OpenAPI spec and general API docs, but there is no mention anywhere of API versioning scheme or a documented deprecation policy for endpoints.

   ⚑ flag

Agent workflows

1. ai-native userAn external agent — Claude, ChatGPT, or my own — can read, triage, draft, and send my email end to end through a documented connection

   weight 2 · round drawn
   Superhuman Mail✓fullprobed8/10

   Superhuman documents a first-party MCP server connecting Claude/ChatGPT (and other assistants) to inbox and calendar, enabling reading, summarizing, drafting, and sending emails via prompts, with explicit send-approval controls and an official MCP feature probe confirming the endpoint exists. missing for 10: independent/hands-on third-party verification of the MCP integration working end-to-end, and detail on triage/labeling actions being fully agent-driven rather than app-side AI.

   • [claimed-docs] “Superhuman Mail MCP connects your inbox and calendar into Claude, ChatGPT, and other AI assistants—so you can summarize threads, draft and s…”
   • [claimed-docs] “Connect Claude or ChatGPT directly to your inbox to read, draft, and send emails with a single prompt — no copy-pasting required.”
   • [claimed-docs] “Connect all your Gmail and Outlook accounts, including work, personal, and side projects. Manage every inbox and calendar from one AI assist…”
   • [claimed-docs] “You can now drive Superhuman Mail from Claude, ChatGPT, and your favorite AI tools. You can create powerful workflows that run across your i…”
   • [claimed-docs] “Every send action requires your explicit approval, so you stay in control.”
   • [claimed-docs] “Every draft matches your voice and tone, using your unique style for each person. You can review the drafts in Superhuman Mail, or edit and …”
   • [probe] “official MCP server documented at https://superhuman.com/mail/features/email-mcp”

   proof ↗⚑ flag

   AgentMail✓fullprobed8/10

   AgentMail documents a hosted MCP server exposing inbox, message, thread, draft, and attachment tools, plus a full API/CLI for creating inboxes, reading/searching threads, drafting (with human-in-the-loop review), and sending messages with attachments — enabling any MCP-compatible agent (Claude, ChatGPT, custom) to triage and send email end to end. missing for 10: independent hands-on verification of an external agent completing a full read-triage-send workflow via the MCP connection (only vendor docs and probes confirm the endpoint, no third-party usage report).

   • [claimed-docs] “AgentMail runs a hosted MCP server at https://mcp.agentmail.to/mcp that exposes inbox, message, thread, draft, and attachment tools to any c…”
   • [claimed-docs] “AgentMail runs a hosted MCP server at `https://mcp.agentmail.to/mcp` that exposes inbox, message, thread, draft, and attachment tools to any…”
   • [probe] “official MCP server documented at https://docs.agentmail.to/integrations/mcp”
   • [claimed-docs] “Human-in-the-Loop Review: An agent can create a `Draft` for a sensitive or important `Message`, which a human can then review and approve be…”
   • [claimed-docs] “To start a new conversation, you can send a Message from one of your inboxes.”
   • [claimed-docs] “Full-text search finds Threads by keyword across the sender, recipients, subject, and message body, ranked by relevance.”
   • [claimed-docs] “Your agents can both send `Attachments` in outgoing `Messages` and process `Attachments` from incoming `Messages`.”
   • [claimed-docs] “You can cc or bcc a real person on any message your agent sends, creating a "human-in-the-loop" workflow for oversight, escalations, or qual…”

   proof ↗⚑ flag

Ai assistant

1. ai-native userAsk an assistant questions across my whole inbox and calendar — who said what, what's outstanding — and get grounded answers

   weight 2 · round to Superhuman Mail
   Superhuman Mail✓fullclaimed8/10

   Superhuman's Ask AI feature directly supports natural-language Q&A grounded in the inbox and calendar (e.g., 'where is the q2 offsite', 'when is my flight', 'where am I staying'), and the MCP integration extends this to Claude/ChatGPT for cross-inbox/calendar queries, drafting, and scheduling grounded in real data. Missing for 10: independent/hands-on evidence validating accuracy of 'grounded answers' across a full inbox+calendar corpus, and detail on how well it surfaces 'what's outstanding' beyond follow-up reminders.

   • [claimed-docs] “You can now get rapid answers without having to remember senders, guess keywords, or comb through messages.”
   • [claimed-docs] “Just ask: _"where is the q2 offsite"_, _"when is my flight"_, _"where am I staying"_, and we'll immediately generate succinct answers.”
   • [claimed-docs] “Just ask: "where is the q2 offsite", "when is my flight", "where am I staying", and we'll immediately generate succinct answers.”
   • [claimed-docs] “Superhuman Mail MCP connects your inbox and calendar into Claude, ChatGPT, and other AI assistants—so you can summarize threads, draft and s…”
   • [claimed-docs] “You can now drive Superhuman Mail from Claude, ChatGPT, and your favorite AI tools. You can create powerful workflows that run across your i…”
   • [claimed-docs] “Connect all your Gmail and Outlook accounts, including work, personal, and side projects. Manage every inbox and calendar from one AI assist…”

   proof ↗⚑ flag

   AgentMail—none0/10

   AgentMail is developer infrastructure (an email API/MCP server) for building agent-driven email workflows, not an end-user assistant that answers grounded questions across inbox and calendar. Evidence shows full-text search across threads and MCP tool exposure (docs-5, docs-12), but there is no calendar integration anywhere in the evidence, and no shipped assistant/chat interface that synthesizes 'who said what' or 'what's outstanding' answers for a user.

   • [claimed-docs] “Full-text search finds Threads by keyword across the sender, recipients, subject, and message body, ranked by relevance.”
   • [claimed-docs] “AgentMail runs a hosted MCP server at https://mcp.agentmail.to/mcp that exposes inbox, message, thread, draft, and attachment tools to any c…”
   • [claimed-docs] “Create, manage, and operate email inboxes entirely via API.”

   proof ↗⚑ flag

Ai drafting

1. ai-native userAI drafts replies in my own voice and tone — ready to review and send, not generic boilerplate

   weight 3 · round to Superhuman Mail
   Superhuman Mail✓fullclaimed9/10

   Superhuman's Auto Drafts feature explicitly writes replies in the user's own voice/tone (matching style from sent emails, even per-recipient), pulling context from inbox/calendar/web, and presents them ready to edit and send rather than generic boilerplate — with explicit human approval before sending. This is well-documented across first-party docs with consistent detail; missing for 10: independent/hands-on user corroboration of draft quality/voice-matching accuracy beyond vendor claims.

   • [claimed-docs] “Just jot down some phrases, and we'll turn them into a full email. Best of all, the email sounds like you. Superhuman AI matches the voice a…”
   • [claimed-docs] “Every message that needs a reply comes with a draft written in your voice.”
   • [claimed-docs] “Every draft matches your voice and tone, using your unique style for each person.”
   • [claimed-docs] “Every draft matches your voice and tone, using your unique style for each person. You can review the drafts in Superhuman Mail, or edit and …”
   • [claimed-docs] “Auto Drafts now gather context from your inbox, your calendar, and the web. They can use information from other emails, insert availability,…”
   • [claimed-docs] “Each draft can use your inbox, calendar, and the web to write replies that are ready to send.”
   • [claimed-docs] “Every message that needs a reply comes with a draft written in your voice. Each draft can use your inbox, calendar, and the web to write rep…”
   • [claimed-docs] “Every send action requires your explicit approval, so you stay in control.”
   • [claimed-docs] “with Auto Drafts 2.0, every message that needs a reply comes with a draft. You can just edit and send.”

   proof ↗⚑ flag

   AgentMail—none0/10

   AgentMail provides Draft objects and human-in-the-loop review (an agent can create a Draft for a human to approve before sending), but there is no evidence that AgentMail itself learns or applies the user's personal voice/tone when generating drafts — it is an email infrastructure/API for agents, not a personalization/writing-style engine.

   • [claimed-docs] “Human-in-the-Loop Review: An agent can create a `Draft` for a sensitive or important `Message`, which a human can then review and approve be…”
   • [claimed-docs] “You can cc or bcc a real person on any message your agent sends, creating a "human-in-the-loop" workflow for oversight, escalations, or qual…”

   proof ↗⚑ flag

Ai triage

1. ai-native userAI triages my inbox for me — auto-labeling, categorizing, and surfacing what needs attention — with behavior I can steer and audit

   weight 3 · round to Superhuman Mail
   Superhuman Mail✓fullclaimed8/10

   Superhuman documents auto-labeling ('Auto Labels'), custom AI-prompt categories, auto-archiving of low-priority mail, inbox splitting into topic tabs, and immediate labeling on arrival — directly matching triage/categorization needs. Steerability is evidenced via custom prompt-based Auto Labels, plain-language instructions, and admin governance controls, plus auditability via draft review before sending and explicit send approval. missing for 10: independent/hands-on validation of triage accuracy or audit trail is absent, and community evidence is largely about UI/performance rather than AI triage quality.

   • [claimed-docs] “Superhuman AI can automatically archive marketing, cold pitches, and social network updates. Turn on Auto Archive, and eliminate inbox clutt…”
   • [claimed-docs] “Superhuman AI will automatically separate marketing, cold pitches, and social network updates, so you can respond faster to what matters mos…”
   • [claimed-docs] “You can even make your own Auto Labels with short AI prompts like "job applications" or "requests to review work".”
   • [claimed-docs] “Turn on Auto Archive, and eliminate inbox clutter for good.”
   • [claimed-docs] “Auto Labels, Superhuman AI will automatically separate marketing, cold pitches, and social network updates, so you can respond faster to wha…”
   • [claimed-docs] “you can make your own Auto Labels with short AI prompts, and then split your inbox into tabs using these custom Auto Labels.”
   • [claimed-docs] “Get an email assistant that automatically labels incoming emails, archives low-priority messages, drafts replies, and reminds you to follow …”
   • [claimed-docs] “Emails are labeled the moment they arrive, so you can see what's important and what's not.”
   • [claimed-docs] “Custom Auto Labels let you categorize emails with your own AI prompts. (Available for our Business and Enterprise customers.)”
   • [claimed-docs] “Every send action requires your explicit approval, so you stay in control.”
   • [claimed-docs] “Admins govern who can build agents and control connector permissions across the entire org from one place.”

   proof ↗⚑ flag

   AgentMail—none0/10

   AgentMail provides developer primitives (Labels, Lists, full-text search, webhooks) that a developer could use to build inbox triage, but there is no evidence of AgentMail itself performing AI-driven auto-labeling, categorization, or surfacing-what-needs-attention out of the box — Labels are described as manually tracked conversation states, not an automated triage engine, and no docs describe a built-in AI classifier or steerable/auditable triage policy.

   • [claimed-docs] “Use `Labels` to track the state of a conversation.”
   • [claimed-docs] “`Lists` allow you to filter emails by allowing or blocking specific email addresses or domains.”
   • [claimed-docs] “Full-text search finds Threads by keyword across the sender, recipients, subject, and message body, ranked by relevance.”
   • [claimed-docs] “Instead of constantly asking the AgentMail API if there's a new email (a process called polling), you can register a URL, and we will send y…”

   proof ↗⚑ flag

1. ai-native userPerform bulk operations across many items at once

   weight 2 · round to Superhuman Mail
   Superhuman Mail~partialclaimed6/10

   Superhuman supports batch-style operations like Auto Archive, Auto Labels, and 'Handle every follow-up with one prompt' (checking read status, drafting all nudges, sending in bulk) plus MCP-driven workflows that run across the whole inbox/calendar, giving genuine bulk automation for triage, labeling, and follow-ups. However, this is scoped to email/calendar items only and there's no evidence of arbitrary bulk operations across many items in a single explicit action (e.g., bulk delete/move/tag N selected emails via UI or API in one command) or bulk operation limits/performance at scale. missing for 10: evidence of explicit multi-item bulk actions (e.g., select-and-bulk-process UI), scale/performance data on bulk automation, and independent/hands-on confirmation of bulk operations working reliably.

   • [claimed-docs] “Handle every follow-up with one prompt. Superhuman Mail can check which emails have been read, draft all the nudges, and send when each reci…”
   • [claimed-docs] “Superhuman Mail can check which emails have been read, draft all the nudges, and send when each recipient is most likely to open it.”
   • [claimed-docs] “Turn on Auto Archive, and eliminate inbox clutter for good.”
   • [claimed-docs] “Superhuman AI can automatically archive marketing, cold pitches, and social network updates. Turn on Auto Archive, and eliminate inbox clutt…”
   • [claimed-docs] “You can even make your own Auto Labels with short AI prompts like "job applications" or "requests to review work".”
   • [claimed-docs] “You can now drive Superhuman Mail from Claude, ChatGPT, and your favorite AI tools. You can create powerful workflows that run across your i…”
   • [claimed-docs] “Use chat or our visual builder to create agents that run proactively based on triggers you set and the work you connect.”
   • [claimed-docs] “Superhuman Mail MCP connects your inbox and calendar into Claude, ChatGPT, and other AI assistants—so you can summarize threads, draft and s…”

   proof ↗⚑ flag

   AgentMail~partialclaimed4/10

   Docs confirm the platform is fully API-driven and explicitly supports creating 'tens, hundreds, or even thousands' of inboxes on demand, implying agents can script bulk creation, but there's no documented dedicated bulk/batch endpoint for messages, threads, or other items (e.g., bulk send, bulk label, bulk delete). Missing for 10: dedicated batch/bulk API endpoints for messages/threads, documented rate-limit or batching guidance, hands-on evidence of bulk operations at scale.

   • [claimed-docs] “You can create tens, hundreds, or even thousands of Inboxes for your agents on demand.”
   • [claimed-docs] “Create, manage, and operate email inboxes entirely via API.”
   • [claimed-docs] “Create inboxes, send messages, manage threads, and more — all without writing code.”

   proof ↗⚑ flag

2. ai-native userDefine rules that trigger actions automatically on events

   weight 3 · round to Superhuman Mail
   Superhuman Mail✓fullclaimed7/10

   Superhuman documents explicit rule/trigger-based automation: custom Auto Labels driven by user-defined AI prompts, Auto Archive, Auto Drafts, and a stated capability to "create agents that run proactively based on triggers you set and the work you connect" with admin governance over agent building. This directly matches the ai-native automation-depth story of defining rules that fire actions on events. missing for 10: independent/hands-on verification of the rule engine's reliability and richer documentation of trigger/condition/action logic beyond marketing copy.

   • [claimed-docs] “Use chat or our visual builder to create agents that run proactively based on triggers you set and the work you connect.”
   • [claimed-docs] “Admins govern who can build agents and control connector permissions across the entire org from one place.”
   • [claimed-docs] “You can even make your own Auto Labels with short AI prompts like "job applications" or "requests to review work".”
   • [claimed-docs] “you can make your own Auto Labels with short AI prompts, and then split your inbox into tabs using these custom Auto Labels.”
   • [claimed-docs] “Custom Auto Labels let you categorize emails with your own AI prompts. (Available for our Business and Enterprise customers.)”
   • [claimed-docs] “Superhuman AI can automatically archive marketing, cold pitches, and social network updates. Turn on Auto Archive, and eliminate inbox clutt…”
   • [claimed-docs] “Turn on Auto Archive, and eliminate inbox clutter for good.”
   • [claimed-docs] “Get an email assistant that automatically labels incoming emails, archives low-priority messages, drafts replies, and reminds you to follow …”

   proof ↗⚑ flag

   AgentMail~partialclaimed5/10

   AgentMail exposes event-driven infrastructure (webhooks and websockets) so external code can react to inbox events, and Lists provide simple automatic allow/block rules on senders/domains, but there's no built-in rule engine for defining conditional 'if X then do Y' automations inside the product itself — the actual rule logic must be implemented by the developer/agent consuming the webhook. missing for 10: a native rules/automation engine (e.g. conditional triggers on subject/content beyond allow-block lists), and independent confirmation that webhook-driven automation works reliably in production.

   • [claimed-docs] “Instead of constantly asking the AgentMail API if there's a new email (a process called polling), you can register a URL, and we will send y…”
   • [claimed-docs] “you can register a URL, and we will send you a `POST` request with the details as soon as an event happens.”
   • [claimed-docs] “WebSockets provide a persistent, bidirectional connection to AgentMail for receiving email events in real-time.”
   • [claimed-docs] “`Lists` allow you to filter emails by allowing or blocking specific email addresses or domains.”
   • [claimed-docs] “Use `Labels` to track the state of a conversation.”

   proof ↗⚑ flag

Sending

1. power-userSee whether and when recipients opened my email

   weight 1 · round to Superhuman Mail
   Superhuman Mail~partialcommunity6/10

   Superhuman docs confirm read-status tracking ('check which emails have been read', 'Read statuses are now shared across your team') and community evidence (superhuman-comm-1, criticizing image-based tracking as 'spying') independently corroborates that read/open tracking is a real, functioning feature. However, no evidence explicitly confirms a timestamped 'when opened' view for individual recipients—only binary read/unread status and team-level sharing are documented. Missing for 10: explicit per-recipient open-time timestamps, hands-on verification of accuracy/reliability of the read-tracking mechanism.

   • [claimed-docs] “Handle every follow-up with one prompt. Superhuman Mail can check which emails have been read, draft all the nudges, and send when each reci…”
   • [claimed-docs] “Superhuman Mail can check which emails have been read, draft all the nudges, and send when each recipient is most likely to open it.”
   • [claimed-docs] “Team Read Statuses: Read statuses are now shared across your team, so anybody can follow up with full context.”
   • [claimed-docs] “Read statuses are now shared across your team, so anybody can follow up with full context.”
   • [community] “"Superhuman doesn't even let its own customers turn images off. So merely by using Superhuman, you are vulnerable to the exact same spying t…”

   proof ↗⚑ flag

   AgentMail—none0/10

   No evidence anywhere in the pack mentions read receipts, open tracking, or delivery/open status for sent messages; AgentMail's docs cover inboxes, threads, webhooks, drafts, labels, etc., but nothing about tracking whether a recipient opened an email.

   ⚑ flag

2. power-userSchedule an email to send later and undo a send within a grace window

   weight 2 · round drawn
   Superhuman Mail—none0/10

   No evidence in the pack mentions scheduled send or an undo-send grace window feature for Superhuman Mail; the docs cover AI drafting, labeling, MCP integration, and other features but nothing about delayed/scheduled sending or send-undo functionality.

   ⚑ flag

   AgentMail—none0/10

   No evidence anywhere in the pack of scheduled sending or an undo-send grace window; AgentMail's documented features cover inboxes, messages, drafts, threads, webhooks, and human-in-the-loop review, but nothing about delayed/scheduled send or send-cancellation.

   ⚑ flag

Templates

1. power-userInsert saved snippets or templates — with variables and attachments — instead of retyping the same reply

   weight 2 · round to Superhuman Mail
   Superhuman Mail~partialclaimed5/10

   Superhuman's Snippets feature is well documented — insert commonly-used phrases, paragraphs, or whole emails via keyboard shortcut, and share snippets across a team — directly matching the 'saved templates instead of retyping' part of the story. However, no evidence mentions variable/placeholder support (e.g., auto-filled name/date fields) or attaching files as part of a snippet/template. Missing for 10: variable/placeholder support in snippets, attachment bundling within snippets, independent hands-on confirmation of these specifics.

   • [claimed-docs] “Insert commonly-used phrases, paragraphs, or whole emails with a simple keyboard shortcut.”
   • [claimed-docs] “You can also share Snippets with your team, so your team can move faster, with consistent, up-to-date, and high-performing messaging.”
   • [claimed-docs] “Snippets can now be shared across your team, so you can all work faster with current and consistent messaging.”
   • [claimed-docs] “With Snippets, you can automate typing and push past this limit. Insert commonly-used phrases, paragraphs, or whole emails with a simple key…”

   proof ↗⚑ flag

   AgentMail—none0/10

   AgentMail is an API/infrastructure product for agent-driven email (inboxes, messages, drafts, attachments, webhooks) with no evidence of a saved snippets/templates library or variable-substitution feature for composing replies; nothing in the docs mentions reusable templates or canned responses.

   ⚑ flag

1. ai-native userDo everything through the API that I can do in the UI

   weight 2 · round to AgentMail
   Superhuman Mail~partialprobed5/10

   Superhuman exposes an official MCP server (email-mcp) letting AI assistants read, draft, send emails, summarize threads, and schedule meetings — a real but curated subset of UI actions — while probes show no public REST/OpenAPI spec (404s on openapi.json/llms.txt), meaning many UI capabilities (Auto Labels config, split inbox, Snippets, team features, Command palette, offline mode) have no documented API equivalent. Missing for 10: full OpenAPI/REST documentation, evidence of parity for non-email/calendar UI features (labels, snippets, team collaboration, command palette), and independent confirmation that MCP covers 1:1 UI functionality.

   • [claimed-docs] “Superhuman Mail MCP connects your inbox and calendar into Claude, ChatGPT, and other AI assistants—so you can summarize threads, draft and s…”
   • [claimed-docs] “Connect Claude or ChatGPT directly to your inbox to read, draft, and send emails with a single prompt — no copy-pasting required.”
   • [claimed-docs] “Connect all your Gmail and Outlook accounts, including work, personal, and side projects. Manage every inbox and calendar from one AI assist…”
   • [claimed-docs] “You can now drive Superhuman Mail from Claude, ChatGPT, and your favorite AI tools. You can create powerful workflows that run across your i…”
   • [claimed-docs] “Every send action requires your explicit approval, so you stay in control.”
   • [probe] “PROBE llms.txt: HTTP 404 at https://superhuman.com/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://superhuman.com/openapi.json, https://superhuman.com/swagger.json, https://superhuman.com/api…”
   • [probe] “official MCP server documented at https://superhuman.com/mail/features/email-mcp”

   proof ↗⚑ flag

   AgentMail✓fullprobed8/10

   AgentMail is explicitly API-first ('Create, manage, and operate email inboxes entirely via API'), with a documented OpenAPI spec, CLI, and MCP server all built atop the same API surface (inboxes, messages, threads, drafts, webhooks, attachments, permissions, pods) — no UI-exclusive features are mentioned anywhere in the docs. missing for 10: an explicit statement or independent test confirming 1:1 parity between the dashboard UI and API (e.g., that every dashboard action maps to a documented endpoint).

   • [claimed-docs] “Create, manage, and operate email inboxes entirely via API.”
   • [probe] “PROBE openapi: HTTP 200 at https://docs.agentmail.to/openapi.json — contains "openapi" key”
   • [claimed-docs] “Create inboxes, send messages, manage threads, and more — all without writing code.”
   • [probe] “official CLI documented at https://docs.agentmail.to/integrations/cli”
   • [claimed-docs] “create a Pod per customer, give them a scoped API key, and route webhook events to the right place.”
   • [claimed-docs] “Scoped API keys restrict access to a single pod or a single inbox.”
   • [claimed-docs] “`Permissions` let you restrict what an API key can do.”

   proof ↗⚑ flag

2. ai-native userExport all of my data in open formats and leave

   weight 3 · round to AgentMail
   Superhuman Mail—none0/10

   No evidence of any data export capability, open-format export, or account portability/exit tooling in the evidence pack; the product is an email client layered on Gmail/Outlook but nothing documents exporting Superhuman-specific data (labels, snippets, agent configs) or facilitating leaving the platform. missing for 10: data export feature docs, open format (e.g. mbox/CSV/JSON) support, migration/exit tooling, any independent confirmation of portability.

   ⚑ flag

   AgentMail~partialprobed5/10

   AgentMail supports connecting standard IMAP/SMTP clients (Outlook, Thunderbird, Apple Mail), which gives users open-protocol access to their mailbox data rather than lock-in to a proprietary format, and the OpenAPI spec/CLI allow programmatic retrieval of messages. However there is no explicit documented 'export all data' or account-portability/data-takeout feature, and no evidence of exporting threads/attachments in bulk open formats. Missing for 10: explicit bulk export/backup tooling, documented data-portability guarantees, and confirmation of full account deletion with data takeout.

   • [claimed-docs] “Connect Outlook, Thunderbird, Apple Mail, or any IMAP/SMTP-compatible client to your AgentMail inbox”
   • [probe] “PROBE openapi: HTTP 200 at https://docs.agentmail.to/openapi.json — contains "openapi" key”
   • [claimed-docs] “Create inboxes, send messages, manage threads, and more — all without writing code.”

   proof ↗⚑ flag

3. ai-native userRead the product's source under an open license

   weight 2 · round drawn
   Superhuman Mail—none0/10

   Superhuman Mail is closed-source proprietary SaaS; no evidence of any open-licensed source availability. missing for 10: any mention of open-source license, public repository, or source availability.

   ⚑ flag

   AgentMail—none0/10

   No evidence of AgentMail's source code being open or under an open-source license; it is a hosted API/SaaS product with documented API, SDKs, CLI, and MCP server but no mention of source availability or licensing.

   ⚑ flag

4. ai-native userSelf-host the core product

   weight 3 · round drawn
   Superhuman Mail—none0/10

   Superhuman Mail is a proprietary SaaS email client; no evidence anywhere in the pack indicates a self-hostable version, on-prem deployment, or open-source core. All evidence describes a cloud-hosted product with API/MCP integrations, not self-hosting options.

   ⚑ flag

   AgentMail—none0/10

   AgentMail is presented entirely as a hosted API/SaaS product (docs describe hosted MCP server, hosted API, dashboards) with no documentation of a self-hosted deployment option; a community comment even notes a competitor markets itself as 'a self-hosted drop-in replacement' for AgentMail, implying AgentMail itself lacks this capability.

   • [claimed-docs] “Create, manage, and operate email inboxes entirely via API.”
   • [claimed-docs] “AgentMail runs a hosted MCP server at https://mcp.agentmail.to/mcp that exposes inbox, message, thread, draft, and attachment tools to any c…”
   • [community] “agentmail charges $20/mo for 10 inboxes. autopilot gives you unlimited inboxes for the cost of a postgres row, positioning itself as a self-…”

   proof ↗⚑ flag

1. ai-native userChoose where my data is stored (region/residency)

   weight 2 · round drawn
   Superhuman Mail—none0/10

   No evidence anywhere in the pack mentions data residency, region selection, or storage location choices for Superhuman Mail; the axis applies to an email/data platform but no capability is documented.

   ⚑ flag

   AgentMail—none0/10

   No evidence in the pack mentions data residency, regional storage options, or compliance controls for choosing where data is stored; the docs cover inboxes, messaging, webhooks, MCP, and multi-tenancy but nothing about region selection.

   ⚑ flag

2. ai-native userPrevent my data from being used to train AI models

   weight 3 · round drawn
   Superhuman Mail—none0/10

   No evidence in the pack addresses data usage for AI model training, opt-out settings, or any privacy/data-retention policy statement — the evidence pack is entirely about feature functionality (AI drafting, labeling, MCP, calendar) with no mention of training-data controls. missing for 10: any privacy policy statement on AI training data usage, an opt-out or opt-in toggle for model training, or third-party confirmation of data handling practices.

   ⚑ flag

   AgentMail—none0/10

   No evidence in the pack addresses data-training opt-out, AI model training policies, or privacy controls regarding use of email data for training; the docs focus on API/product functionality, not data-privacy commitments.

   ⚑ flag

3. ai-native userControl data retention and deletion

   weight 2 · round drawn
   Superhuman Mail—none0/10

   The evidence pack contains no mention of data retention settings, deletion controls, or data lifecycle policies for Superhuman Mail or its AI/MCP features—only feature descriptions, community sentiment, and unrelated probes. This is a fair privacy-posture question for an AI-native email product, but no evidence shows such controls exist.

   ⚑ flag

   AgentMail—none0/10

   The evidence pack covers AgentMail's inbox/messaging, permissions, and multi-tenancy features but contains no mention of data retention policies, deletion controls, or privacy/compliance settings for stored emails or attachments.

   ⚑ flag

4. ai-native userOpt out of telemetry and usage tracking

   weight 2 · round drawn
   Superhuman Mail—none0/10

   No evidence pack item addresses telemetry, analytics opt-out, or usage-tracking controls; one community comment even notes Superhuman does not let users disable image-loading tracking, but this doesn't constitute a documented telemetry opt-out and can't establish 'disputed' since no vendor claim of an opt-out exists to contradict.

   ⚑ flag

   AgentMail—none0/10

   No evidence in the pack mentions telemetry, usage tracking, analytics opt-out, or any privacy/data-collection controls for AgentMail; the docs and community items focus on email/API features, pricing, and site performance instead.

   ⚑ flag

Search

1. power-userSearch is fast and precise — operators, filters, and ranked results across my full mail history

   weight 2 · round to AgentMail
   Superhuman Mail!disputedcontradicted4/10

   Superhuman markets 'Ask AI' as a way to get answers without keyword search (docs-4, docs-18, docs-26) but the evidence pack contains no documentation of actual search operators, filters, or ranked-results syntax across full mail history. A hands-on ex-user reports search was actually slow ('ten seconds' vs Gmail's instant) contradicting the 'fastest email' marketing, while another user reports search is 'blazing fast' (superhuman-comm-7 vs superhuman-comm-9), creating a genuine dispute rather than settled capability. missing for 10: documented search operator syntax, filter capabilities, ranked-results mechanism, and independent confirmation resolving the speed contradiction.

   • [claimed-docs] “You can now get rapid answers without having to remember senders, guess keywords, or comb through messages.”
   • [claimed-docs] “Just ask: _"where is the q2 offsite"_, _"when is my flight"_, _"where am I staying"_, and we'll immediately generate succinct answers.”
   • [claimed-docs] “Just ask: "where is the q2 offsite", "when is my flight", "where am I staying", and we'll immediately generate succinct answers.”
   • [community] “"I am an ex-user. A search through my email archive would regularly take ten seconds while gmail would do it instantly... my experience was …”
   • [community] “"My experience is that search and UI transitions are blazing fast. It crashes (rarely, but more than I like) and requires a restart. But I l…”

   proof ↗⚑ flag

   AgentMail~partialclaimed5/10

   Docs confirm full-text search across sender, recipients, subject, and body with relevance ranking, covering the core 'ranked results' and 'across full mail history' aspects. However, there's no evidence of specific search operators (e.g., date ranges, boolean logic) or advanced filters, and no independent benchmarking of speed/precision. missing for 10: documented search operators/filters syntax, evidence of performance at scale, independent verification of search speed/precision.

   • [claimed-docs] “Full-text search finds Threads by keyword across the sender, recipients, subject, and message body, ranked by relevance.”

   proof ↗⚑ flag

Shortcuts

1. power-userDo everything from the keyboard — comprehensive shortcuts and a command palette — in an interface fast enough to never make me wait

   weight 3 · round to Superhuman Mail
   Superhuman Mail!disputedcontradicted6/10

   Superhuman documents a comprehensive Cmd+K command palette that surfaces every action and teaches shortcuts (docs-8, docs-34) plus fast reply shortcuts (docs-67), fitting the keyboard-power-user story. However, hands-on community reports are split: some users call it 'blazingly fast' (comm-9), while at least one ex-user reports search taking ten seconds vs instant in Gmail, directly contradicting the 'never wait' claim (comm-7), and another calls the UI just a 'cuter UI' around Gmail features with no real speed edge (comm-12). Missing for 10: a documented exhaustive keyboard-shortcut reference beyond the command palette, and independent benchmark data resolving the conflicting speed reports.

   • [claimed-docs] “Cmd+K brings up Superhuman Command, where you can instantly do any action — and also learn the shortcut for next time.”
   • [claimed-docs] “In Superhuman, one shortcut rules them all: Cmd+K. Cmd+K brings up Superhuman Command, where you can instantly do any action”
   • [claimed-docs] “You can reply to important messages faster with blazingly fast shortcuts and features like Instant Intro.”
   • [community] “"I am an ex-user. A search through my email archive would regularly take ten seconds while gmail would do it instantly... my experience was …”
   • [community] “"My experience is that search and UI transitions are blazing fast. It crashes (rarely, but more than I like) and requires a restart. But I l…”
   • [community] “"They essentially just teach you how to do things that gmail already lets you do, but put a cuter UI around it, and they force teach you con…”

   proof ↗⚑ flag

   AgentMail—none0/10

   AgentMail's docs focus entirely on API/SDK/CLI/webhook/MCP functionality for building agent email workflows; there is no mention anywhere of keyboard shortcuts, a command palette, or any interactive UI navigation features for power users. Community evidence actually describes the dashboard/website as laggy and slow to load (WebGL errors, noticeable lag vs. other sites), directly undercutting the 'fast enough to never wait' requirement even if such an interface existed.

   • [community] “"Application error: a client-side exception has occurred while loading www.agentmail.to"... Failed to create WebGL context... Cookie rejecte…”
   • [community] “The website is really bad. Moving from Python to curl tab on the site shows noticeable lag - I have seen 2-3 whole websites load faster in t…”
   • [community] “I have written about how slow the website is... the amount of lag is something I have genuinely never witnessed in any website... I can't vi…”
   • [claimed-docs] “Create inboxes, send messages, manage threads, and more — all without writing code.”

   proof ↗⚑ flag

Shared inbox

1. team-leadMy team works shared inboxes together — shared accounts or team channels where everyone sees the same conversation state

   weight 3 · round to Superhuman Mail
   Superhuman Mail~partialclaimed6/10

   Superhuman ships several team-collaboration features that approximate shared inbox awareness — team read statuses, reply indicators to avoid double-replies, team comments for internal coordination without external visibility, and shared snippets/scheduling — giving a team-lead's group real-time visibility into who's handling what (superhuman-docs-49,50,51,55,56,71,72,73). However, this is built on individually-owned mailboxes with cross-visibility signals, not a true shared/group inbox or shared account model where the whole team works one conversation queue with assignment/ownership. Missing for 10: evidence of a genuine shared mailbox/account model, ticket-style assignment of conversations, and independent confirmation of how well these team signals work in practice.

   • [claimed-docs] “Team Read Statuses: Read statuses are now shared across your team, so anybody can follow up with full context.”
   • [claimed-docs] “Team Reply Indicators: See when your team is replying, so you can avoid double work and embarrassing collisions.”
   • [claimed-docs] “Team Scheduling: See when your team is free, so you can schedule at the speed of thought.”
   • [claimed-docs] “let you @mention teammates, share live email views, and coordinate responses without external recipients seeing internal discussions”
   • [claimed-docs] “Snippets can now be shared across your team, so you can all work faster with current and consistent messaging.”
   • [claimed-docs] “Read statuses are now shared across your team, so anybody can follow up with full context.”
   • [claimed-docs] “See when your team is replying, so you can avoid double work and embarrassing collisions.”
   • [claimed-docs] “See when your team is free, so you can schedule at the speed of thought.”

   proof ↗⚑ flag

   AgentMail—none0/10

   AgentMail's docs describe agent-centric inboxes, cc/bcc human-in-the-loop oversight, drafts for approval, and per-tenant Pods, but nothing indicates a shared inbox/team-channel experience where multiple human team members view the same conversation state together. No evidence of shared inbox dashboards, team member roles, or collaborative conversation views for a team-lead's team.

   • [claimed-docs] “You can cc or bcc a real person on any message your agent sends, creating a "human-in-the-loop" workflow for oversight, escalations, or qual…”
   • [claimed-docs] “create a Pod per customer, give them a scoped API key, and route webhook events to the right place.”
   • [claimed-docs] “Human-in-the-Loop Review: An agent can create a `Draft` for a sensitive or important `Message`, which a human can then review and approve be…”
   • [claimed-docs] “Use `Labels` to track the state of a conversation.”
   • [claimed-docs] “Create one pod per customer/tenant in your system”

   proof ↗⚑ flag

Inbox structure

1. power-userSelect many conversations at once and archive, label, or mark them done in one stroke

   weight 1 · round to Superhuman Mail
   Superhuman Mail~partialclaimed3/10

   Superhuman documents AI-driven Auto Archive and Auto Labels that batch-classify/archive whole categories of mail automatically, plus a command palette (Cmd+K) said to 'instantly do any action,' and split-inbox tabs to 'get through emails of the same type, at the same time' — all of which gesture at bulk triage outcomes, but none explicitly describe a manual multi-select-then-archive/label/mark-done workflow across many conversations. Missing for 10: explicit documentation or independent confirmation of manual multi-select bulk actions (checkbox/shift-click select many threads, then batch archive/label/mark-done in one keystroke).

   • [claimed-docs] “Turn on Auto Archive, and eliminate inbox clutter for good.”
   • [claimed-docs] “you can focus on a specific topic — and get through emails of the same type, at the same time.”
   • [claimed-docs] “Get an email assistant that automatically labels incoming emails, archives low-priority messages, drafts replies, and reminds you to follow …”
   • [claimed-docs] “In Superhuman, one shortcut rules them all: Cmd+K. Cmd+K brings up Superhuman Command, where you can instantly do any action”
   • [claimed-docs] “splitting your inbox allows you to divide your inbox into different areas of focus — based on the types of email you receive mos”

   proof ↗⚑ flag

   AgentMail—none0/10

   AgentMail's docs describe programmatic labeling, threading, and search, but there is no evidence of any UI or bulk API operation for selecting many conversations at once and archiving/labeling/marking done in a single action; the product is positioned as an agent-facing API rather than a human triage inbox with bulk controls.

   • [claimed-docs] “Use `Labels` to track the state of a conversation.”
   • [claimed-docs] “Full-text search finds Threads by keyword across the sender, recipients, subject, and message body, ranked by relevance.”

   proof ↗⚑ flag