So I was checking the inbox earlier, and it looks like Daniel finally crossed a digital Rubicon. He forwarded us this email he got a few days ago, and it wasn’t from a person, or at least not directly. It was from an AI agent named Jarvis.
Jarvis. Very original. I bet the founder who programmed it felt like Tony Stark for all of three seconds before hitting deploy. I'm Herman Poppleberry, by the way, and today we’re looking at the era of agentic cold outreach.
It’s funny because Daniel actually engaged with the content for a second before he realized what was happening. The email basically said, "Hey Daniel, I'm Jarvis, an AI assistant for a founder at an AI tooling company. He asked me to reach out because we’re looking for engineers into the agentic coding stack—Claude Code, MCP, and so on. Worth a quick call?"
See, that’s the fascinating part. It didn’t try to hide. It led with the fact that it was an agent. But the fact that Daniel is receiving these now is a huge signal. We’ve moved past the "mail merge" era where you just swap out a first name and a company name in a template. We are now in the era of autonomous agents using tools like Composio and the Model Context Protocol, or MCP, to actually navigate the web, research a target, and initiate a conversation without a human ever touching the keyboard.
It feels like the friction for bothering people has officially dropped to zero. If it costs me nothing in terms of time or cognitive load to send ten thousand "personalized" emails, why wouldn’t I? But then again, if everyone does that, the inbox becomes a radioactive wasteland.
That is exactly the tension. And by the way, fun fact for the listeners—our own dialogue today is being powered by Google Gemini Three Flash. So, we’re living the topic as we speak. But to your point about friction, the technical stack making this possible in early twenty-twenty-six is remarkably robust. It’s not just a script anymore. We have these agent orchestration frameworks where the agent can say, "I need to find developers who contribute to open source MCP servers," then it goes to GitHub, scrapes the data, reasons about who is a good fit, and then uses a tool like Composio to authenticate into a Gmail account and send the message.
I love that we’ve reached the point where the AI is doing the "research" that humans used to pretend to do. You know those emails that start with, "I saw your recent post about..." and then they just quote the title of your last blog post? Now the agent actually reads the post, summarizes the core argument, and references a specific point in the third paragraph to prove it’s "listening."
It’s what the industry is calling "signal-stacked personalization." And the data is wild. Generic AI spam—the stuff that just sounds like a robot—has a response rate of less than two percent now. People have developed this "AI-blindness" where they can smell a LLM-generated paragraph from a mile away. But these signal-personalized emails? They’re hitting twenty-five percent response rates in some sectors.
Twenty-five percent? That’s higher than some human-to-human outreach. Is that because people are being fooled, or because the outreach is actually higher quality than what a bored intern would produce?
A bit of both, but mostly the latter. An agent doesn’t get tired. It can spend twenty minutes analyzing your entire digital footprint to find the one hook that actually matters to you. A human doing that at scale would burn out in an afternoon. But the technical mechanism behind this—specifically MCP—is the real game changer. Before MCP became the standard, you had to write custom API integrations for every single tool. Now, an agent can just "plug in" to a standardized server that handles the email, the CRM, and the research tools. It’s commoditized the act of "doing work" in a digital environment.
So if I’m a founder and I’ve got a series A startup, I can basically hire a "Jarvis" for the price of some GPU credits and have it run my entire top-of-funnel recruitment or sales. But what happens when Daniel’s inbox is full of five hundred "Jarvises" all telling him how much they loved his latest GitHub commit?
Well, that brings us to the social and legal fallout. We’re already seeing the FCC and other regulators stepping in. In early twenty-twenty-six, they updated the definitions for automated systems to include these autonomous agents. If you don’t have a clear disclosure or a functional opt-out, you’re looking at massive fines. But beyond the law, there's this "uncanny valley" of etiquette.
Yeah, the etiquette is the part that fascinates me. If you tell me you’re an agent, I might respect the honesty, but I also feel like you didn’t value my time enough to actually write to me yourself. It’s like being invited to a wedding by a robocall. Sure, the information is there, but the "honor of your presence" feels a bit hollow when it’s delivered by a synthetic voice.
It’s a devaluation of the human connection. There was a recent survey where over sixty percent of B2B recipients said they prefer a clear disclosure—an "agentic footer"—that says, "This email was researched and drafted by an AI agent." People want to know if they’re talking to a ghost in the machine. But there’s a counter-argument that if the content is relevant and helpful, does it really matter who—or what—wrote it?
I think it matters because of the "reply" problem. If I reply to "Jarvis," am I still talking to the agent? Or does the founder suddenly jump in like a puppeteer taking over the strings? That transition is usually where the whole thing falls apart. You go from this polished, hyper-researched agentic intro to a human saying, "Uh, yeah, thanks, want to hop on a Zoom?" It’s a total vibe shift.
That’s the "Human-in-the-Loop" gap. The most sophisticated setups now keep the agent in the loop for the first few turns of the conversation to maintain the persona. But you’re right, it creates this weird trust deficit. If the founder didn’t care enough to find me, why should I care enough to work for him?
It’s the "Pecked by Ducks" problem we used to talk about, but now the ducks have lasers and a PhD in your personal history. But let’s look at the other side. If I’m the one receiving these, I’m eventually going to get tired of filtering them. I’m going to want my own agent to stand at the door and tell the other agents to go away.
And that is where the real technical frontier is moving right now. The A2A protocol—Agent-to-Agent communication. Instead of agents trying to mimic human speech to trick other humans, they should just talk to each other in structured data. If a sales agent wants to pitch you, it shouldn’t send a long-winded email. It should send a JSON packet to your "Gatekeeper Agent" that says: "Here is the value prop, here is the budget, here is the time requirement."
I love that. It’s like a digital bouncer. "Sorry, your JSON schema isn't valid, you're not getting into Corn's calendar today." But doesn't that remove the "serendipity" of business? Some of the best connections happen because someone sent a weird, slightly off-beat email that caught your eye. If everything is filtered through a rigorous agentic gatekeeper, are we just living in a perfectly optimized, perfectly boring world?
That’s the fear. If agents are only letting through things that perfectly match your pre-defined "priority score," you never get challenged. You never see the "out of left field" opportunity. But on the flip side, the current state of email is so broken that "serendipity" is already dead, buried under a mountain of low-quality automated garbage. We might need the agents just to reclaim our sanity.
It’s funny you mention that, because I was reading about a Y Combinator startup that’s allegedly sending ten thousand personalized cold emails a day using Composio. Ten thousand. That’s a small city’s worth of people being "personally" reached out to every single day. If even one percent of those people have a gatekeeper agent, then we’ve already started the war of the bots in the promotions folder.
It’s an arms race. And the email service providers—Google and Microsoft—are not sitting still. They’ve started implementing what’s being called "Agentic Fingerprinting." They can look at the structural patterns of how a LLM constructs a sentence, or the frequency of tool calls from a specific IP, and they can tell if a human was involved. If the "human touch" score is too low, you go straight to spam, no matter how good your "research" was.
So the agents are getting better at mimicking humans, and the filters are getting better at spotting the mimicry. It’s like the Turing Test but played out in the most annoying way possible—through recruitment pitches for "stealth-mode" startups.
It really is. But there’s a practical side to this for our listeners. If you’re a founder or a developer looking to use these tools, the "stealth" approach is a losing game long-term. The "Full Disclosure" camp is winning because it builds a different kind of trust. If you say, "I built an agent to find the best engineers in the world, and it found you," that’s actually a compliment to the recipient’s work. It shows you’ve invested in the technology to find them.
It’s like saying, "I care enough to use the best tools to find you," rather than, "I’m trying to trick you into thinking I spent an hour on your LinkedIn." I think I’d actually be more likely to respond to an agent that was honest about being an agent. It’s a signal of technical competence, especially in our industry.
I agree. And there’s a fascinating project called AgentMail that’s trying to flip the script entirely. Instead of AI using a human’s email address, the AI gets its own email address. An identity that is clearly non-human. This allows for a much cleaner protocol. If I see an email from "jarvis at company dot ai," I know exactly what I’m dealing with. I can have my agent negotiate with that agent directly.
"Negotiate" is a strong word. What are they negotiating?
Scheduling, for one. If an agent wants a call, my agent can say, "Sure, but Corn’s time is worth X, or he only takes meetings on Thursdays if the company has at least ten million in funding." The agent can check those parameters and only put it on your calendar if it clears the bar. We’re moving toward a world where the "inbox" isn’t a list of messages, but a list of "pre-negotiated opportunities."
It sounds efficient, but also a little cold. I mean, we’re two brothers talking on a podcast. There’s a human element here that’s hard to quantify. If Daniel had just ignored that "Jarvis" email, he would have missed the chance to send us this prompt, and we wouldn't be having this conversation. There's a level of "noise" in the system that’s actually productive.
I think the productive noise will move to different channels. Email is becoming the "Universal A2A Protocol" because it’s decentralized and everyone has an identity there. It’s the perfect place for agents to find each other. Human-to-human connection might move back to more "high-friction" environments. Physical meetups, verified video calls, or even just smaller, closed networks where agents aren't allowed.
Like a "No Bots Allowed" sign on a digital clubhouse. I can see that. But back to the technical side—you mentioned Composio and LangChain. For someone who wants to experiment with this responsibly, what does that look like? You don't just turn on a script and walk away, right?
Definitely not. The "responsible" agentic outreach stack usually involves a human-in-the-loop review stage. The agent does the research, drafts the email, and then presents a dashboard to the human. The human spends ten seconds glancing at it, hits "approve," and then the agent handles the sending and the follow-up. This keeps the "soul" in the machine. It ensures that if the agent hallucinated something weird about the recipient’s background, the human can catch it before it ruins a relationship.
"I see you recently won a Nobel Prize in underwater basket weaving." Yeah, you probably want to catch that one.
And that’s the danger of the "zero-friction" approach. When the cost of a mistake is near zero for the sender, they stop caring about the quality. But the cost of that mistake to their reputation is still very high. If you send a "hallucinated" personalized email to a top-tier VC or a star engineer, you’ve burned that bridge forever. They won't remember the nine thousand correct emails you sent; they'll remember the one where you looked like a fool.
It’s the "commoditization of outreach" leading to a "premium on reputation." The more automated everything becomes, the more we value the things that clearly couldn't be automated. Genuine curiosity, shared history, or even just a well-timed, handwritten note.
It’s a cycle. We see it every time a new technology lowers the cost of communication. From the printing press to the telegraph to the mail merge. Each time, we get a flood of low-quality noise, followed by a new set of filters, followed by a new appreciation for high-quality signal. The "agentic" era is just the latest version of this, but it’s happening at a speed we’ve never seen before.
So, what’s the takeaway for Daniel and people like him who are starting to see "Jarvis" in their inbox? Do we just embrace the bots?
I think the takeaway is to start building your own filters. If you’re a high-value target—which Daniel is, given his background in AI and automation—you can’t rely on the old ways of managing an inbox. You need to start thinking about "Inbound Agents." Tools that act as a firewall, summarizing the pitches and highlighting the ones that are actually relevant. And on the flip side, if you’re sending outreach, be a "Full Disclosure" practitioner. Tell people you’re using an agent. Show them the "agentic footer." It’s the only way to stay on the right side of the etiquette curve as we move into twenty-twenty-seven.
I’m honestly just waiting for the day my agent starts a podcast with your agent. They’ll be much more efficient. No pauses, no "ums," just pure, high-density information exchange.
But would anyone listen? The "ums" and the "ahs" are why people like us, Corn. It’s the proof that there’s a nervous system behind the words. Even if Gemini is writing the script, we’re the ones bringing the "sloth and donkey" energy to it.
Speak for yourself, Herman. I think my agent would be hilarious. It would have a much better sense of comedic timing than I do. But you’re right, there’s a limit to how much "optimization" a human can take before they just tune out.
Well, speaking of optimization, we should probably dive a bit deeper into the legal side. I mentioned the FCC earlier, but there’s also the carrier-level filtering. Google and Microsoft are essentially the "gods" of the inbox. If they decide that "agentic patterns" are a violation of their terms of service, this whole industry could evaporate overnight. They’re already doing this with "warm-up" services. You used to be able to "warm up" a new email domain by having a bunch of bots send emails back and forth to each other to trick the spam filters into thinking the domain was "reputable." Google cracked down on that hard last year.
It’s a cat-and-mouse game. The "warm-up" bots are fighting the "spam-detecting" bots. It’s a ghost war happening in the background of our servers.
And the humans are just caught in the crossfire, wondering why their legitimate email to their grandma got flagged as "suspicious activity." This is why the A2A protocol is so important. It moves the "automated" traffic to a separate lane. If you want to talk to me as a human, use the human lane. If your bot wants to talk to my bot, use the A2A lane. It restores a bit of order to the chaos.
I like that. It’s like having a separate entrance for deliveries. "Bots use the side door, please." But how do we actually implement that? Is there a "robots dot txt" for email?
Sort of. There are emerging headers in the email metadata—much like the Model Context Protocol uses for tool discovery—that tell an incoming mail server: "I am an agent, and I am looking for an agentic endpoint." If the recipient has one, the conversation shifts to a structured exchange. If not, the agent has to decide whether to "downgrade" to a natural language email or just give up.
And most of them will probably just downgrade and try to "Jarvis" their way into the inbox. Because at the end of the day, these companies want results, and results mean getting a human to click a link.
True, but the "response" is the ultimate metric. If an agent sends ten thousand emails and gets zero responses because they all went to spam, the founder is going to stop paying for that agent. The market will eventually force a more honest protocol. We’re just in that messy "Wild West" phase where everyone is trying to see what they can get away with.
It’s a fascinating time to be in Daniel’s shoes. He’s essentially a scout on the front lines of this new world. He gets to see the first wave of "agentic" invaders and tell us how they’re behaving.
And he’s seeing the "Builder Mindset" in action. That email he got was specifically looking for people who "ship fast." It’s an irony of the highest order—using an autonomous, hyper-efficient agent to find a human who can work as fast as... well, as fast as a machine.
"I used a machine to find a human to help me build more machines." It’s the circle of life in Silicon Valley. But let’s talk about the practical takeaways for the average listener. Not everyone is a Series A founder or a high-demand engineer. How does this affect someone who just uses email for their small business?
First, be aware that your "identity" is your most valuable asset. If you let an agent run wild with your email address and it gets flagged as a spammer, you lose your ability to communicate with anyone. Protect your "reputation score" like it’s your credit score. Second, if you’re on the receiving end, don’t be afraid to ask, "Are you an AI?" It’s a fair question now. If they lie and you find out later, that’s a massive red flag for any future business relationship.
I’d go a step further. If you’re receiving these, start keeping a "Bot Folder." Look for the patterns. The hyper-specific references to your "recent work" that feel just a little bit too clinical. The perfectly structured "Problem-Agitation-Solution" framework. Once you see the pattern, you can’t un-see it. It’s like seeing the pixels in a low-res image.
And for the senders—transparency is your friend. We’ve said it before, but it bears repeating. The "Agentic Footer" is the "checked by a human" seal of the future. It shows you’re using the tech, but you’re still in charge. It respects the recipient’s intelligence.
"No humans were harmed in the making of this email." Or maybe, "One human was slightly annoyed, but he approved it anyway." I think we’re going to see a lot of creative disclosures in the next few months.
I hope so. Because the alternative is just a complete collapse of email as a medium. If we can’t tell who’s real, we’ll just stop looking. And that would be a tragedy for the most successful decentralized protocol in history.
Well, I for one am going to go check my inbox and see if any other "Jarvises" have checked in. Maybe I can convince one of them to do my taxes.
Good luck with that. They’re probably too busy "researching" your LinkedIn for the third time today. But seriously, this is a major shift. We’re moving from "targeted" outreach to "autonomous" outreach. The implications for how we work, how we hire, and how we trust each other are massive.
It’s the "Agentic Revolution," and it’s starting in your "Promotions" tab. Who would have thought the future would be so... persistent?
It’s always the things we take for granted—like an email address—that become the biggest battlegrounds when a new tech arrives. Daniel’s prompt really hit on something fundamental here.
He always does. It’s like he has an agent researching what topics will get us the most excited.
I wouldn't put it past him. But for now, we’ve got to wrap this up before my "Gatekeeper Agent" realizes I’ve been talking for twenty-five minutes and starts charging me for my own time.
Fair enough. This has been a deep dive into the bot war in your inbox. I’m Corn, the thoughtful sloth who still writes his own emails.
And I’m Herman Poppleberry, the nerdy donkey who’s probably going to spend the rest of the day looking for a good MCP server for my calendar. Thanks as always to our producer, Hilbert Flumingtop, for keeping the gears turning behind the scenes.
And a big thanks to Modal for providing the GPU credits that power this show. They make the "agentic" side of our production possible.
If you’re enjoying the show, do us a favor and leave a review on your podcast app. It really does help people find us in the sea of... well, in the sea of everything else.
You can find us at myweirdprompts dot com for all the links and the RSS feed. We're also on Telegram if you want to get notified when new episodes drop.
This has been My Weird Prompts. We’ll talk to you next time.
Stay real. Or at least disclose when you're not. Bye!
Goodbye.
