The 60-Day Experiment: How I'm Building a Marketing Team of AI Agents

When I took the VP of Marketing role at Rhumbix, I established a goal for myself that would have sounded crazy just a couple of years ago. 

The Goal: Build a marketing department that’s 80% led by AI agents

The first couple of months in any new leadership role can feel like drinking from a fire hose. Learning the business, joining strategy sessions, sizing up the battlefield. But while I was building relationships and reviewing historical performance, budgets, and materials, something was happening in the background.

Case studies were being written. Outbound emails were being sent. Feedback was being provided on campaigns. And warm introduction pathways into our target accounts were being mapped.

All of this, autonomously. All of this, while I was on a Zoom call.

That was an aha moment. Four agents are working in the background while I’m helping someone on my team on a Zoom call. 

It doesn’t make sense for AI agents to replace all marketing roles. But one thing is certain: the way executives build marketing teams and departments has changed significantly with these new AI capabilities. Those who are successful in the future will be those who learn to leverage these innovations.

The Problem with the Old Playbook

Here’s what I’ve learned from a decade of building marketing teams at similar-stage companies: You inherit a backlog of urgent needs. At our stage, you need to improve messaging and positioning. Refine ICP. Build new go-to-market motions. Test those motions. Monitor competitors.

The traditional answer? Scale up. Crawl, walk, run. Hire a BDR for $60K plus commission. Hire another BDR. Hire a PMM, usually the first critical hire at this stage. Bring on fractional contractors and freelancers to fill in the gaps. Bring on junior interns for research.

You’re looking at $150K to $250K annually for basic coverage. And that’s before you factor in the three months of onboarding, the learning curve, and the inevitable turnover as the market changes.

I’ve successfully deployed and made dozens of these hires over my career. They take time to get right. And require management overhead. And they scale somewhat linearly; one person, one salary, one set of deliverables.

This time, I asked a different question: What if, instead of hiring, I created an AI agent for this role? 

The Experiment

I didn’t set out to prove anything radical, but Rhumbix is at a critical time. We’re seeing strong results on product-market fit. We’re also in a competitive space. We don’t have the luxury of a slow ramp-up. I decided to start with these four specific jobs-to-be-done/roles. 

1. The Case Study Agent

Case studies are critical for B2B startups. They create invaluable social proof and clarity on ROI. But they take time, scheduling interviews, transcribing calls, writing drafts, designing layouts, and getting approvals. Industry estimates put the cost between $500-$1,000 per case study, not counting weeks of calendar Tetris.

But using Claude Code, I vibe-coded a template for case studies in an 8×10 PDF format. Then I trained the AI to structure transcripts into a Situation-Opportunity-Result framework. And all we need to do now is upload a transcript from Gong, and we get a fully designed case study.

What does this mean? Within two hours, we created 16 case studies. Generated autonomously. I didn’t even have to push anything; just pointed it to the folder with the transcripts, and it worked its magic. Sixteen customer stories, 2000+ pages of transcripts, incredibly valuable both internally and externally.

That’s $8,000 worth of deliverables created in less than two hours. This usually would have taken weeks, if not months.

2. The BDR Agent

Hiring a Business Development Rep is a classic early-stage GTM move. But they’re expensive, hard to train, and the good ones leave or move up. Still, you need to prove outbound motion at this stage to make the hire.

Instead, I built a BDR agent using Lindy.ai. It runs outbound sales plays autonomously. It emails 50-70 prospects per day that fit our ICP. It averages two responses daily and two positive responses weekly.

We’re still refining this, but one of those outbound responses last week, during that two-hour window, was a CFO at a target account.

A human BDR costs $5,000- $7,000 per month, fully loaded. The agent costs a fraction of that. And it doesn’t take coffee breaks, call in sick, or make excuses. 

The PMM Agent

This one is really interesting. A PMM is the most critical hire at our stage, reviewing messaging, improving campaigns, and providing feedback on personas and ICP. And it’s typically a $150K hire. Very expensive at our stage.

Could I create a PMM agent? A Product Marketing Manager that functions on its own and provides feedback to the rest of the organization?

I started training the agent on customer transcripts from deals we’d won. It ran well on six transcripts. But when I fed it all of our won customer transcripts over the past year – over 2,000 pages – it couldn’t process it. Broke down. Hallucinated. Glitched.

I hit a block, but I didn’t abandon this approach. I asked for help. And after working with an AI consultant, we built sub-agents. Think of it like this: instead of one person handling a task that’s too big, you have four people handling it together. That’s what we built.

Four sub-agents: 

1. Coordinator Agent to manage workflow
2. Analyzer Agent to process transcripts into a sales framework
3. Trend Aggregator Agent to synthesize patterns
4. Report Generator Agent to compile insights in a digestible way.

The result? With ease, the sub-agents together successfully processed 2,000-plus pages of raw transcripts and produced organized training data along with an executive summary of customer trends and quotes. I then uploaded this to Notion so my team could get real-time feedback on messaging and chat with the PMM agent directly.

This enables the agent to grade marketing campaigns. Provided specific, actionable feedback. Score these overall campaigns. I now have a PMM who knows the business better than anyone in the industry. No big expense. Faster results. Greater effectiveness than I expected. 

The Intro Finder Agent

Relationship selling dominates construction. Everybody knows everyone. Warm introductions are gold.

But mapping these pathways takes hours. Manually searching LinkedIn, Google, identifying mutual connections, assessing relationship strength, and prioritizing outreach. Very difficult.

I built an agent that automates this entirely. It identifies connections to target prospects and recommends warm-introduction pathways.

In one run, it identified over 20 possible intro pathways into ENR Top 600 prospects.

The Numbers That Matter

Let’s get concrete.

Executing all of this with human resources would cost $15,000-$25,000 per month. That’s conservative, multiple BDRs running simultaneously, PMM, writing, and research.

What am I spending on AI tools, API credits, and subscriptions? A tiny fraction of that cost.

But here’s what matters more than the cost savings: these agents work in parallel. They work while I’m in meetings. They work if I take a break or am off for the day. They don’t need to be managed as often as a person does. And, most importantly, their output is proving to be greater than some hires I would have made, and faster!

In one two-hour window this past week, while I was on a Zoom call working on yet another agent with one of my team members, the Case Study Agent produced 16 case studies, the BDR Agent contacted over 20 prospects and got a response from a CFO at an ICP account, the PMM Agent graded and provided feedback on a six-touch nurture campaign to our contract marketer, and the Intro Finder Agent mapped 20 mutual connections across 10 prospect accounts.

That’s not incremental improvement. This is a paradigm shift.

The AI Roster Keeps Growing

In addition to these four agents, I’ve deployed six more: 

• Competitor Monitoring Agent that tracks rivals’ moves and messaging shifts
• Market Insights Agent that synthesizes industry trends
• Content Specialist Agent for ongoing content production
• List Building Agent that compiles targeted prospect lists
• Prospect Research Agent that deep-dives on accounts before outreach and pulls in trends.
• Video/Podcast/Image Creation AI process for multimedia assets.

Ten agents. Each handles a discrete function that would traditionally require either a dedicated hire or significant hours from existing team members. And this doesn’t even include the sub-agents that fall under some of these agents.

The compound effect is what matters. These agents don’t just save time individually; they create a system in which insights from one feed inform the work of another.

The Market Insights Agent surfaces a trend; the Content Specialist Agent blogs about it in the voice of our customer. A human checks messaging, posts, and promotes it. As we get leads, the BDR Agent can reach out and reference this content. The Prospect Research Agent pulls specific information from their website to make outreach more contextualized. The PMM agent follows the customer journey through the transcripts. 

This is not just a collection of tools. I am starting to see something that could function like a department.

What I’ve Learned

First of all, this isn’t about replacing humans. Humans need to be in the loop with the current state of AI and what the foreseeable future holds. 

This is about augmenting what a small team can accomplish.

We saw the same thing with digital years ago. Digital technology enabled small companies to compete with larger brands. AI is appearing to have a similar effect.

The agents handle the volume work, the repetitive, time-consuming tasks that would otherwise eat up human hours and 100s of thousands of marketing dollars. This frees the team to focus on strategy, creativity, and judgment calls that still require human insight.

This is a shift from headcount to jobs-to-be-done workflows.

The second lesson is about architecture. The PMM 1.0 agent failed when I fed it too much data. The solution wasn’t better prompting; it was better system design. Breaking the task into specialized sub-agents to process large amounts of data.

This is the future of knowledge work. Not a monolithic AI that does everything, but an orchestrated system of specialized sub-agents working together.

The third lesson is about speed. We’re in the construction technology industry, which is notoriously slow to adopt new tools. Our customers are contractors who still use paper timesheets. But marketing technology doesn’t have that luxury. The competitive landscape moves fast. The companies that figure out how to leverage AI agents effectively will outmaneuver their slower competitors.

The Question I’m Left With

At similar stages in my career, when joining companies like Rhumbix, I’ve hired many people to prepare for repeatable, efficient growth: BDRs, PMMs, content writers, and marketing coordinators.

Now I’m asking a different question: Can we build AI agents to fill these roles instead of hiring full-time?

The early signs are very positive. These agents are more productive than I expected. They’re more consistent. They scale. They are faster. The result is proving to be better.

And here’s the question I’m asking the rest of my organization, and it’s a question you might want to ask yourself, too:

If you could make one full-time hire to increase your productivity by 3- 5x, who would you hire?

And could you build an agent for that same role?

Because if the answer is yes, we’re not just talking about cost savings. We’re talking about fundamentally restructuring how small teams compete with larger organizations in go-to-market.

I’m sixty days in. The experiment continues. But the results so far suggest that the marketing department of the future won’t scale through headcount but through its effectiveness in building and orchestrating teams of agents and humans that work together to achieve 5-10x results.

Happy to talk through what I’m doing and building if you’d like to learn.

Please send me a DM.

David Gabriel, Vice President of Marketing, Rhumbix