Email is an optimal interface for long-running agents. It’s multithreaded and asynchronous with full support for rich text and files. It’s a universal protocol with identity and authentication built in. Moreover, a lot of workflow critical context already lives in email.
We wanted to build email agents that you can forward your work to and get back a completed task. The agents could act entirely autonomously as you wouldn't need to delegate your identity. If they did get stuck they could just send you, or anyone else, an email.
Using Gmail, we kept getting stuck on the limitations of their API. No way to create inboxes programmatically. Rate and sending limits. OAuth for every single inbox. Keyword search that doesn't understand context. Per-seat pricing that doesn't work for agents.
So we built what we wished existed: an email provider for developers. APIs for creating inboxes and configuring domains. Email parsing and threading. Text extraction from attachments. Realtime webhooks and websockets. Semantic search across inboxes. Usage-based pricing that works for agents.
Developers, startups, and enterprises are already deploying email agents with AgentMail. Agents that convert conversations and documents into structured data. Agents that source quotes, negotiate prices, and get the best deals. Agents that emulate internet users for training models on end-to-end tasks.
Here's demo of Clawdbots communicating using AgentMail: https://youtu.be/Y0MfUWS3LKQ
You can get started with AgentMail for free at https://agentmail.to
Looking forward to hearing your thoughts and feedback.
Thanks to Action Mailbox in Rails[1], I give all my records email addresses. Eg let ecommerce "order" records accept forwarded emails that are pinned as comments. It opens you up for doing things like forwarding a purchase order and having the PO number pulled out and attached to an order, or forwarding tracking information from a supplier and having it attached to a "supplier order" etc.
In my personal life I have individual email addresses for all my utilities and emails automatically get filed away.
If this idea tickles your fancy, I opensourced Emitt[2], an inbound email processing server with LLM-powered automation.
1. https://guides.rubyonrails.org/action_mailbox_basics.html
I just type in whatever address I want to send from, and then as long as the domain is verified in SES it'll go through.
Google has A2A: An Agent-to-Agent Protocol. SaaS is plumetting in value.
Arbitrary semantics made sense when communications were human-dominated.
If agents dominate these fields, why wouldn't they simply set their own protocols and methods to communicate both text, binary, and agreed data structures?
There's an assumption that email is somehow the best channel, when you've found yourself that the most popular, functional interfaces don't align with your expectations.
Then, ultimately I have a single agent that can sit in numerous communication platforms, such as email
The question is the transition, because email is undoubtedly the most ubiquitous channel of communication in today. I would only give my agent an A2A integration if your agent has an A2A integration, but because you don't we are at a stalemate. I'd rather just give my agent an inbox where I know it can communicate with the other billions of people that already have an email address.
Email isn’t the final protocol for agents. It’s the bridge that lets them participate in today’s internet while native agent protocols/networks emerge.
Guaranteed this is going to attract a ton of abusers who are looking to use this for signing up to services, spamming or other nefarious purposes, which then blacklists the doman. This is an infinite whack-a-mole.
do you guys have some ways of handling it?
We expect our infra and policies to evolve with usage, and one of our goals is to make agent driven email safer than the status quo, not just more scalable
Resend uses SES since it's almost impossible to get private IP mail to hit the inbox through ProofPoint filters. Looks like you have no idea about any of this. You don't even have knowledge of email reputation, much less a plan. Have you heard of Senderscore? You will have all zeros. Saying "SPF DKIM DMARC" is wild - that's a checklist from 15 years ago.
We’re not wide open for abuse nor are we bypassing the hard parts of email reputation. Quite the opposite. We also utilize SES's infrastructure and monitor reputation continuously, but we don’t assume SPF/DKIM/DMARC are sufficient on their own. They’re basics we have implemented, not the entire strategy.
You are correct private IPs per customer make sense once you’re sending meaningful volume (on the order of ~10k+/day per IP). But its inaccurate to say we are sending from a single private IP. IP pools are typically segmented by reputation and traffic profile for customers.
Reputation here is earned at multiple layers: per-IP, per-domain, per-inbox, and over time. We rate-limit, isolate, or revoke bad actors without poisoning unrelated senders. Hopefully this makes sense.
We have strict rules for our customer service people not to respond to what seems to be a bot, since all the "agent" based communication we get is for conducting scams. It is never worthwhile to engage with or pursue.
If we lose a sale or two, that's okay.
The more interesting shift isnt whether humans will keep using email with agents, but whether agents can become distinguishable from noise. Historically, we ignored anonymous calls but we engaged with known vendors that had reputation, contracts, and consequences.
Once an agent has a persistent identity/a domain, trust becomes something that can be accumulated over time instead of being assumed per message.
I understand the differentiator vs GMail, but API-based scripted email access isn’t new.
Doesn't seem like a particularly difficult problem to solve.
Didn't Alexa fail miserably with the "have AI buy something for me" theory?
There is a significant mental in allowing someone else make purchase decisions on my behalf:
- With a human, there is accountability.
- With deterministic software, there is reproducibility.
With an agent, you get neither.
FWIW - I am not anti-LLM. I work with them and build them full time.
Do you see more pushback in specific industries? I did some quote/purchasing automation work in food mfg a decade ago, and those guys were super difficult to work with. Very opaque, guarded, old-school industry.
Are logistics companies really that poor so they cannot afford to pay workers wages?
Sure hiring 10–100 humans gives accountability, but reality is it doesn't scale in any comparable way compared to agents in speed, coverage, or responsiveness. The sheer volume agents can pump out(more vendors, more quotes, faster cycles) is the benefit, while humans retain accountability at the decision boundary.
In practice the agent does the gruntwork, and the human gets looped in when confidence is low. Accountability doesnt dissapear, it gets concentrated where it matters most
I am 99% certain I could build to parity in a weekend using Cloudflare without the the pricing limitations.
I am thinking it would be within the free tier of CF usage.
I am not certain I have the bandwidth to communicate over delivery and plain text inspection concerns.
I am 99% certain I could reach parity in a weekend by publishing content on public networks, without the old distribution or pricing constraints.
I think it would all run on infrastructure that is effectively free to use.
I am not certain I have the bandwidth to handle distribution, sustained attention, and moderation once the content starts flowing.
We are re-making entire scenes from movies with our desired dialog and endings.
This is a singular solution.
Death of true understanding because everyone feels entitled to paying the lowest perceived monetary cost possible for everything in their lives.
The mental torque of a teenager today is so weak that they cannot answer four WHY? questions in a row without a panic attack.
I guarantee you that the "moat" is very much intact for the SaaS we are building (more developer / gaming tool but then again so is this) because it requires specialized skills, synthesis and most importantly AI would have no idea how to build it without very specific prompting from our architect
CRUD wrappers never had a moat. Even the most basic viable SaaS that wasnt a micro SaaS had some secret sauce or differentiation. And AI doesnt help you get that unless you already know what it is.
Not to mention network effects. Users are a moat and if you can sell and grow fast enough and create a community, no amount of "there's a clone" can beat it. Never underestimate the power of brand recognition.
Most SaaS pre-AI had an open source alternative. Most people didn't use them not because the open source alternative lack some features, it was because mainteinance was hard. It's way easier to pay a small monthly fee and forget about it.
I'm reminded of the infamous Dropbox Hacker News comment[1]. If you're looking at stuff like this thinking "what's the point? I could just make that myself" then you're not the target audience in the same sort of way Ikea isn't trying to sell stuff to carpenters.
This is true even when the barrier to entry in making these sorts of systems has gotten way lower.
What does this even mean?
I could spend $1,000s on tokens asking an agent to build (some semblance of) Sentry, or New Relic, but why would I bother? I have real work to do in the near-term, and I'm happy to pay for services that help me do it.
And that the actual act of these 3rd parties offering said products maintains not only the software, but the knowledge required to build it.
But I have still written parts of the comments where I had assumed that you were right and I am still gonna let it be to show what my thinking process was I guess. Not that it matters now but I am frugal in finding alternatives sooo yeah :> lol (currently the cf private beta option's the best imo)
Yea I am a little bit confused as well being honest.
That being said, I feel as if even if Cloudflare might not be the best approach, one can try out purelymail (https://purelymail.com/) as well.
I feel as if Amazon SES might be the best option for it (or any EU alternative, I remember seeing an UK service with the same competitive pricing of Amazon SES)
But that being said, I am unable to understand the exact use of E-mail & what's the real idea to suggest the best infrastructure to use.
I mean technically, can something like cloudflare workers for inbox and amazon ses for outbound work if cloudflare email product is only for receving
That being said all of this is basing on the fact that what you thought is right
We're devs ourselves so ik the first thought is usually "how hard can it be?" in our validation, we thought it was hard enough to build a startup around :) these things are easier said than done, and no one in 2026 should be stitching together email workflows. especially not agents
So let's see how many people actually build it. Let's make it the new browser test instead and launch many open source solutions instead and see what's the best perhaps.
It would be a really great experiment imo.
I have no doubt this will be huge.
We’re an O365 GCC shop. Appreciate that your enterprise options include Bring Your Own Cloud, that makes things much easier for us.
It would be nice to have integrations with n8n and Glean.
adi@agentmail.cc
I would advise to have the landing page static, or at the very least as something that can't crash entirely.
Do you have plans to only allow reading mail from specific senders, and giving fine-grained control on the API side?
I would like to give this to an AI agent but only initially where I and my team can forward mail to the agent, and not give it the ability to draft and send emails on its own.
Will be testing it with some of the agent work I'm doing on agent.ai.
If you're ever looking for a super low-maintenance indie investor, please reach out. Have lots of YC folks that will vouch for me.
Did you record yourself reading out the output of an LLM prompt in this video?
Long-running agents are themselves not optimal though. There are a ton of these coordination layers for long running agents now but they don't make any sense under other paradigms
It's more "long running" because the agent takes 4 steps, then waits a week for the user to email it back. We might have a successful client exchange that takes a month, but for the Agent it's 99% just waiting for the next user reply.
Alternatively you can set up IMAP access.
Obviously you have to be able to configure proper OAuth2 keys to do this, but again, that's another 30 seconds spent in Grok/GPT/Claude/whatever.
Then I scrolled even more in the website and the amount of lag, my my, I don't even know what to say but the amount of lag is something I have genuinely never witnessed in any website. This is like a new low. I really just want to archive this website to preserve how abysmally slow the website is and its aniations and everything. image literally loading 10% and everyhting.
Ship fast and break things is shying from what I am witnessing in here. Sfabt (ship fast and break things) is gonna use your service to talk to the agent which created this project to ask it personally to slow down
I can't view your website in 16 gigs of a computer... Weird where the world's progressing in this sense and how it got (YC funded?)
Quite frankly I am out of words for how slow the website is. Its really just that bad to be in its own league. Sorry to say.
Spam doesn't matter for an agent mailbox, but sophisticated fraud does.
It lets agents plug into the same trust systems the web already uses! And this opens the door to new ways agents can do work and build credibility on the internet.
aws just gives you a low-level smtp + api service. we are the application layer they do not offer but your agents need to actually use email as first-class users.
The idea is pretty solid, automation platforms often provision a mailbox per flow for this reason, so it makes sense to make a generic service that can be used through MCP for agents
the mcp agent mail project is agents getting their own identity in an internal messaging layer.
$ phone call bill
ok call_id=3f2a
$ phone status 3f2a
dialing
$ phone status 3f2a
answered
bill: hello
$ phone say 3f2a "hey, quick question"
ok
Second time at least HN is launching YC on one of my products:
BrowserBox - hyperbeam
Mailpilot/AI-chat.email - agentnail
My Show from 14 days ago: https://news.ycombinator.com/item?id=46629191 - hmm, why didn't it get into YC?
> Looks at developer console...
- "Failed to create WebGL context: WebGL is currently disabled." Dafuq does an email website need WebGL for?
- "Cookie “dmn_chk_xxxxxxxx-yyyy-dead-beef-123456789ABC” has been rejected for invalid domain."
Let me guess...vibe-coded?
https://www.agentmail.to/enterprise
the cutesy ASCII art is rendered in a proportional, not monospace, font, so it looks terrible.
If that's the quality y'all can live with and accept, no wonder the web turned to shit.
Like holy Cow, I am not sure what to say seeing such noticable lag
The time to go from python to curl. I have seen websites load faster, heck I feel like even 2-3 whole websites can actually be loaded in the noticable lag time we observe.
Absolutely crazy to witness.
