How to Build a Team of AI Agents From Someone Who Has Done It
Definition: AI teammates differ from agents; they're custom GPT tools requiring human interaction and context.
Start: Build AI tools where resources are scarce and skill gaps exist, especially starting with content tasks.
Time: Integrate AI development into small time slots; incrementally build and refine AI ecosystem.
Data: Use real, research-based data for AI personas and avoid including any personal identifying information.
Future: Automate current job tasks, freeing time for strategic work and skills development in AI environments.
This interview was originally conducted on The Digital Project Manager Podcast, where host Galen Low spoke with Megan Ratcliff about what it actually takes to build and manage a team of AI teammates — not someday, but right now, with today's technology. It has been reformatted into an article for easier reading.
Megan Ratcliff, a go-to-market consultant and coach at Clarity & Motion Collective, built a team of AI teammates during her time as Head of Marketing at SaaS career tech company Dice — an ecosystem that eventually did her job well enough that she had time to invent her next one. She had no technical background, no budget, and no spare hours on her calendar. What follows is her playbook.
What is an AI teammate, anyway? (And what it isn't)
First, some definitional housekeeping, because the terminology is getting muddy. "The word AI Agent is being tossed around really freely right now," Ratcliff says. "An agent is an AI tool that is acting autonomously on its own. And I think when people say they're making agents on LinkedIn, they're actually just making a teammate, which is a custom GPT with a human in the loop."
Unlock for Free
Create a free account to finish this piece and join a community of forward-thinking leaders unlocking tools, playbooks, and insights for thriving in the age of AI.
Have an account? Log In
That distinction matters. A teammate is a purpose-built assistant — a campaign copywriter, a strategist, a persona simulator — trained on your real context and used with you at the controls. And critically, building one doesn't require you to become an engineer. Ratcliff is blunt about her own ceiling: "I am not somebody who's gonna go build something in JSON or HTML or React. That is Greek to me."
Nor does everyone need the same depth of skill. "Learning AI is like learning another language," she says. "There will be maybe 10 to 20% of the population that becomes AI fluent, and they're the ones that are building all kinds of things that you have never dreamed of. And then, to use the language learning metaphor, everyone else is only able to order a glass of water and ask for the bathroom." The baseline, in her framing, is simple: "can you interact with an LLM chat bot. And, can you use it effectively?"
Learning AI is like learning another language
Step 1: Start where the pain is (usually content)
Ratcliff's first teammate wasn't born from a grand AI strategy. It was born from scarcity. "I was the head of demand at the time," she recalls. "I didn't have a lot of resources. I didn't have a lot of cash. I didn't have a lot of time and I didn't have a lot of team members – human team members – to do the work."
So she started with the sharpest pain point: "I started by building out the first teammate, which was a campaign copywriter. Content creation is a great place for anyone new to AI to start. It's really easy because you can easily understand what good looks like."
Content creation is a great place for anyone new to AI to start. It's really easy because you can easily understand what good looks like.
From there, the logic extended to her own skill gaps. Coming from an agency account director background rather than classical demand gen training, she says, "I was building teammates to help augment where my gaps were in my skillset." That's the pattern to copy: don't build AI for the sake of AI — build it where you're stretched thinnest.
Step 2: Build in the margins — you don't need a free afternoon
The most common objection to building anything new is time, and Ratcliff dismantles it. She didn't block off half-days. "It was like in between meetings. If I had 15 minutes, I would go work on something and then I would prompt a set of instructions, go to my meeting, come back, see what the output was, refine, update."
Not everything survived. "I would keep some teammates, I would ditch some teammates, and then I started linking them together" — and that discard pile is progress, not failure. The compounding payoff came fast: "Over time, I built an ecosystem in probably four months that started doing my job pretty well for me. And then I had more time to innovate."
Over time, I built an ecosystem in probably four months that started doing my job pretty well for me. And then I had more time to innovate.
Two accelerants helped. The first was mentorship — a fractional CMO at Dice who reviewed her first custom GPT instructions and pushed her to iterate. As Ratcliff puts it about the people she now trains: "Once they get going they become unstoppable. But they just need that little bit of help first." The second was temperament. "AI really rewards ideas-people. So if you're an ideas person, now is your time."
Step 3: Feed it real information — never garbage, and never PII
Teammates are only as good as their inputs, and Ratcliff draws two hard lines. The first is on personas: "You cannot build an AI persona based on AI information. You have to have actual real information. You have had to have done the research." Interviews, qualitative data, quantitative data — that's what makes something like Ratcliff’s persona simulator worth consulting.
The second line is absolute: "I never put PII into AI and neither should you."
Step 4: Scale from solo tool to shared intelligence
Once individual teammates work, the question becomes what to build next — and the answer starts with outcomes, not tools. "Understanding the outcome that you're going towards, the skillset you already have on your team helps you identify what teammates you even need. You don't need to build AI for the sake of AI," Ratcliff says. "There's gonna be gaps. And so how do you fill those gaps? You build a teammate to fill those gaps."
You don't need to build AI for the sake of AI.
The real transformation happens when teammates stop being personal productivity hacks and become shared infrastructure. "Once you start to build that shared intelligence layer, and then you have a shared orchestration layer, shared teammates that you guys are using," she explains, "that's when you break down the silos between the organizations and you start moving as one, and the AI is your connective tissue."
What this looks like in practice: two examples
The president simulator. Ratcliff had an idea she wanted to pitch to the president of Dice — someone she didn’t know well. Rather than risk the cold pitch, she built a simulator from transcripts of his all-hands meetings, a job description for his role, and language from his emails. The simulator hated her idea. So she used it to reframe the pitch around what he actually cared about: "We didn't change the sentiment, we just changed the way that it was framed." The real pitch landed. When she confessed the experiment, he loved it — and the simulator was repurposed as the backbone for building his personal brand content.
The go-to-market strategist. After a product launch, something odd happened: "We launched it and it went differently than we all thought. And the people we thought would buy that product were not buying it." The ICP needed redefining, and no one was accountable for it. "Nobody owns the go to market. So then I was like, why not me?" She built a strategist teammate, fed it conversion rates, the original ICP, and account-level buying data, and used it to build a new buyer profile and design validation tests. Shared across teams, that teammate "started to break down the silos between marketing sales enablement, rev ops" — a single source of truth everyone could consult.
Common misconceptions, busted
"AI will take your job." Ratcliff's answer is a deliberate yes-and. "You're going to be able to build a system of tools that supports the work that you do from an execution standpoint, and you can also use these tools to help uplevel your strategic influence on the organization. So what you should be doing is using AI to replace your current job while building your new job. That is what the future of this looks like." She's honest about the clock, though. For anyone doing purely repetitive, task-oriented work: "You will get away with that for probably two more years, and then you're gonna not have a job anymore."
What you should be doing is using AI to replace your current job while building your new job. That is what the future of this looks like.
"I can just sit this one out." She compares this moment to a previous inflection point: "If you were resistant to the internet, you know, like you didn't fare that well." Even AI certifications strike her as a temporary signal — when someone mentioned certifying their team, her reaction was "cool, that's like getting certified in the internet."
"Managing AI teammates is just like managing people." "Myth. False. AI teammates don't have emotions," she says — though they do deserve a diagnostic review "probably at a regular interval quarterly, probably twice a year in some cases."
"This tool can replace your whole marketing team." "Any tool claiming it's gonna replace your marketing team is marketing. It's bad marketing at the least," Ratcliff says. Humans still hold the taste, judgment, and nuance. "It won't replace your marketing team. It will make them better."
"Putting your AI agents on your resume is a gimmick." "Valid. I've seen this before," she says of candidates who interview alongside their agent teams — a signal of exactly the fluency she'd hire for. She'd even flip the classic interview question: "Maybe the new question isn't what is your weakness? It's what teammates have you built to augment your skills?"
The warning: why the human stays in the loop
For all the enthusiasm, Ratcliff's playbook comes with a hard boundary. "Only humans have taste and only humans can have the understanding of nuance and the historical reference and information that an AI can't pick up," she says. Her rule for automation placement: "I'm automating the edges of those workflows, not the heart of those workflows." Clean the data automatically, draft the report automatically — but the judgment call in the middle stays human.
Skip that boundary and you get slop. She points to a McDonald's AI ad produced in Europe: "It was bad, and people really pushed back and they pulled it because it wasn't good. And that's where the human element of it is so critically important."
Her favorite analogy is laundry. Washers and dryers automate two stages, but a human decides what goes in, sets the dials, and pulls the leggings out before they shrink. "There's going to be processes that we're doing manually right now that just don't make sense" to keep manual, she notes — but also processes that are critically important for humans to maintain, because "it has to do with judgment, it has to do with ethics."
That human-centered boundary is also the answer to team anxiety, which is why her change management work always begins the same way: "The first step is dispelling the fear." From there, the real work is helping people redesign their roles around what they actually want to keep. "The mindset shift is so important because without it, without reimagining the work, nothing fundamentally changes. You're just doing the same work, maybe a little faster."
The mindset shift is so important because without it, without reimagining the work, nothing fundamentally changes.
The playbook's paradox
Here is the uncomfortable core of Ratcliff's advice: the safest career move available right now is to automate yourself out of your current job on purpose — because the alternative is waiting for someone, or something, to do it to you. The reward isn't unemployment; it's the time and standing to raise your hand for work nobody owns yet, the way Ratcliff claimed go-to-market when no one else would. And in the future she describes, where org charts give way to work charts and teams assemble around outcomes rather than titles, that's the durable position. As she puts it: "Get rid of the title, keep the skills, and go do the work that's best aligned with what you're good at and what you're interested in."
