You should be playing with AI agents for marketing
Inside a marketing org with no marketers, just 40 AI agents
👋 Hi, it’s Kyle Poyar and welcome to Growth Unhinged, my weekly newsletter exploring the hidden playbooks behind the fastest-growing startups.
I believe that the ability to hire and manage AI agents will be a foundational skill for, well, everyone.
Shopify’s CEO recently said that “using AI effectively is now a baseline expectation of everyone [at Shopify]” and that “we will add AI usage questions to our performance and peer review questionnaire." Similar company-wide memos have been circulated by CEOs of Amazon, Box, Duolingo, Fiverr, Klarna and countless others. More CEOs have quietly repeated the sentiment at All Hands meetings or as directives to managers.
But if I’m being totally honest, I’m not a huge AI adopter myself. I’ll dabble with ChatGPT or Perplexity for secondary research and synthesis. But I make my graphics. Analyze my survey data. Draft my LinkedIn posts. And I certainly write each Growth Unhinged newsletter.
I’ve been looking for best-in-class examples around how to build and manage AI agents for sales and marketing. Enter Jacob Bank, founder and CEO of Relay.app, a startup that allows anyone to create their own AI agents. Jacob recently shared his AI agent-first marketing org chart and it went absolutely viral on LinkedIn, racking up more than 20,000 (!) comments.
Jacob’s marketing org is just him as the CEO and more than 40 AI agents collectively doing the work of a five person marketing team. I sat down with Jacob to unpack exactly how he did it – and how you can do it, too.
The TL;DR:
AI agents have spiky intelligence and can be surprisingly powerful for marketing tasks.
Modern teams will have hundreds of AI agents in the org chart for every one person.
While that sounds complex, each individual agent is much easier to build than you think.
There’s power in building your own personal AI agent stack that’s specific to your goals and style of working. (This stack is also something you can bring to future roles.)
The AI agent marketing org
You can think of AI agents as software systems that complete tasks on your behalf. Sometimes these tasks can be straightforward or deterministic, which is typically the realm of traditional automation tools (think: “if this, then that” workflows). At other times these tasks might be complex, multi-step processes that require reasoning, memory and autonomous decision making. (This shift toward AI autonomously doing work is leading to changes in pricing models, as I’ve written about previously.)
AI agents act more like people who you hire rather than software that you use. This means founders and business leaders should write their own ideal org charts, then figure out which of these roles can be done by AI versus full-time hires or contractors.
Jacob splits his 40+ AI agents into six core marketing functions:
Social media including content research, content creation and tracking & follow-ups
Blog & website including blog content creation, promotion and tracking & analysis
Email including email newsletter and lifecycle marketing
Lead qualification including qualification research and CRM updates
Community including events, education and Slack
Partner program including application reviews and tracking
This specific combination of AI agents is driven by Relay.app’s marketing strategy (which is heavily oriented around social media and community) as well as the current capabilities of AI agents. It’s worth mentioning that many of Jacob’s AI agents include a human-in-the-loop component similar to how a manager might want to review an intern’s work before shipping it.
Let’s jump into the specific functional applications for AI agents. As we do, I’ll highlight AI agent templates that you can try for yourself.
AI agents for social media
Social media is Relay.app’s top performing marketing channel. It’s also where Relay.app employs the most AI agents. There are content research agents (LinkedIn, X and YouTube), content creation agents (LinkedIn, X and YouTube) and tracking and follow-up agents (LinkedIn, X and Bluesky).
Jacob even has an AI agent to reply to LinkedIn comments – yes, a controversial topic! These are specifically for replying to comments on his own posts and come in handy whenever Jacob uses the comment and I’ll send you this asset strategy.
Jacob’s favorite agent is his LinkedIn engagement coach. Every month it analyzes Jacob’s recent posts and writes a digest of what’s going well and what’s not. This coach recently encouraged him to post on Saturday mornings, for example, which is when he’s most likely to go viral. This use case is fairly easy to copy with opportunities to customize the frequency of coaching, how to receive the output and any special instructions. (Here’s a template you can repurpose.)
I used to think about AI agents as similar to interns, who are typically assigned lower-risk administrative or research tasks. But I wouldn’t think to ask an intern to coach the CEO! While AI agents can handle many intern-level responsibilities like administrative or research tasks, Jacob describes AI models as having spiky intelligence. Sometimes they behave like an intern and at other times they bring the skill of a PhD.
“AI agents struggle with some things that we think are easy, but are WAY better than we are at other things you'd never expect an intern to be able to do like: coaching you on how to improve on sales calls, analyzing hundreds of pages of reports into a concise summary or critiquing your marketing strategy.”
- Jacob Bank, founder and CEO of Relay.app
Another broadly applicable AI agent worth sharing is the LinkedIn content research agent. Jacob’s agent writes a report on the top-performing content themes from five relevant influencers: Dharmesh Shah, Jason Lemkin, Greg Isenberg, Nathan Latka and… me.
You can easily adjust the frequency of the report, which influencers to include, which LLM to employ to summarize the responses (it defaults to Claude 4 Sonnet), specific instructions for what the report should include and where to receive the report. (Here’s a template you can repurpose.)
AI agents for blogging and website
Jacob employs AI agents to help write Relay.app blogs, promote those blog posts on LinkedIn, update existing posts and track performance.
A personal highlight for me was the blog post updater, which monitors blog posts to ensure they’re up to-do-date. Not only does this improve the user experience for any website visitors, it helps Relay.app rise up the organic search rankings. For this AI agent, Jacob has found that AI can’t quite take the last mile action on its own; he stays in the loop to review blog post updates before they go live.
As part of the blog post updater, Jacob built an AI agent that tracks competitor pricing changes.
It monitors competitor pricing changes by scraping competitor websites, analyzing the plans and sending alerts when there are significant updates. (Here’s a template you can repurpose.)
AI agents for email
Jacob has AI agents for Relay.app’s email newsletter as well as for personalized lifecycle marketing emails.
Relay.app’s newsletter is currently just a weekly summary of the things they post during the prior week. The AI agent looks up recent webinars, YouTube videos and other pieces of content. It summarizes the new posts and then writes a draft newsletter in Relay.app’s tone of voice. Any time there’s content going out to Relay.app’s audience, Jacob includes a human-in-the-loop step for Jacob to review/approve drafts before sending.
The lifecycle marketing agents look up product and payment information to recommend next steps for new users. There’s even an agent to identify and check in with churned users.
AI agents for lead qualification
Employing AI agents for lead qualification is a popular use case, and for good reason. It’s hard to imagine going back to the days of manual enrichment, research, CRM updates and qualification!
I wouldn’t necessarily say Jacob is reinventing the wheel with his demo request qualifier agent, but it’s still a good use case to highlight.
Relay.app collects demo requests through a HubSpot form, specifically asking for a LinkedIn URL (optional) and company name (mandatory). The AI agent then sends that information to OpenAI to make a decision to classify the demo request as either: qualified (if so, send a scheduling link), unqualified (if so, politely decline) or not sure (if so, escalate to the team via Slack). For each decision, OpenAI needs to give the reason for its decision and provide a summary of the requestor's profile. (Here’s a template you can repurpose.)
AI agents for community and partnerships
Relay.app has a few different community-led growth (CLG) plays including a live event series, self-paced course course, a Slack community and a template gallery.
The self-paced course is effectively a Google Doc that’s personalized to the community member. AI agents add new participants to the right courses, send them a welcome email, personalize the course materials and then send course progress reminders on a periodic basis.
Jacob has AI agents for their certified partner program, too. Certified partners are AI and automation agencies and consultants who provide extra support to Relay.app users (for example, building complex AI agents). While these are valuable partners for Relay.app, they’re resource-intensive as well. Certified partners get free access to the product and a 30% revenue share for the first year of subscriptions.
An AI agent reviews all new applicants and recommends whether to accept or reject them (turnaround time is typically <48 hours). Another checks in on partner engagement. There’s even an AI agent for recommending the best partner based on a customer’s requirements.
Wrap up
In reflecting on Jacob’s experience, the folks who get the most out of AI agents won’t be super individual contributors (ICs). They’ll be super delegators. They’ll map out existing workflows, find discrete tasks that can be handed off, and then patiently create the instructions and training for an AI agent to own it 24x7.
The good news is that you don’t need to be an AI expert to unlock this power. With a few hours after work or over the weekend, you could soon hire your own army of marketing AI agents.
This was amazing and insightful to read. To know that this app could so much people really need to know about it.
This is exactly the kind of thinking PMs need to embrace—owning more of the full stack, from R&D to GTM. I’ve been encouraging product leaders to think beyond specs and sprint planning, and start automating user discovery, qualification, and even experimentation like this. In the relm of "you can build anything", the discovery part becomes way more critical. AI agents don’t just scale execution—they collapse the distance between insight and action.