Angela Ferrante (Enterprise Marketing at Zapier) and Alex Lieberman (co-founder of Morning Brew and CEO of Tenex) had locked in an agenda for a webinar. Then a wave of new tools dropped, making the content obsolete.
So instead, sans script or deck, they covered three ideas every marketer needs to think about right now: the AI tools that actually matter, how to navigate the tension between sanctioned and unsanctioned AI use at work, and real workflows you can build today with tools like Claude Code, Codex, Cursor, and Zapier MCP.
The biggest takeaway? The tools that engineers have been using to write code are now some of the most powerful tools in a marketer’s stack. And you don’t need to be technical to use them.
Table of contents:
Why should marketers care about coding agents?
If you assume coding agents aren’t for you based on their name, you’re not alone. But AI coding tools can do more than just write code. They can build workflows, connect to APIs, scrape data, generate content, and automate multi-step processes, all through plain English instructions.
Coding agents aren’t just for software engineers. Every function, including marketers, can find a game-changing use case for these tools. Even Zapier’s own YouTube channel manager is saving hours of work with automated workflows he built himself.
I always [assumed]…I didn’t have the right to play around with a lot of these tools. In an era where engineers aren’t even typing code anymore and everyone’s just writing in the English language, it was a massive unlock for me.
Alex Lieberman
Inside Alex’s AI content machine
The most concrete example from the webinar was Alex’s “content machine,” a multi-step system he built in Claude Code. It turns one piece of content into dozens of posts formatted for different platforms.
Here’s how it works:
The oracle: The system connects to Slack and Notion, scans conversations and meeting transcripts, and identifies “content spikes,” moments with strong stories, opinions, or anecdotes that could become content. It scores each spike on story quality, emotional resonance, and lesson depth. The goal: eliminate the blank page problem.
The interview panel: Once a spike is selected, a panel of AI interviewers (modeled after real interviewers) asks questions designed to draw out the best version of the idea. Each interviewer has a distinct style based on research into how they actually conduct interviews. Alex responds using voice-to-text (“yap to text,” as he calls it), which removes the filtering that happens when you type and lets ideas flow more naturally.
Production and refinement: The system generates drafts for multiple content types, including long-form website playbooks, short and long-form LinkedIn and X posts, and newsletter content. Each type has its own prompt trained on Alex’s previous posts.
The editor board: Similar to the interview panel, a board of AI editors reviews each draft, scores it, and suggests changes. The drafts go through revision loops until they meet quality standards.
The feedback loop: Every piece of feedback Alex gives on a draft is used to improve the underlying prompts. Over time, the system gets better. Alex says he’s now at the point where he can post content directly from the machine without editing.
The total build time? About 25 hours, with the initial setup taking roughly 30 minutes and the remaining time spent refining prompts and providing feedback. As Alex described it: “It’s very much like a really high-octane junior employee. I give them all the context they need, tell them where we’re headed, they start doing work, check in with me, I give them feedback, they get better.”
The core of the AI slop problem
One of the most provocative points from the webinar: AI slop is the result of a skill issue, not a deficiency in the technology.
“The models are good enough now that slop is actually, ironically, a skill error,” Alex said. “It’s self-deprecation of your own inability to work with this technology and give it feedback.”
The fix comes down to two things:
Give the model enough context. Generic output comes from generic input. The more specific the context you provide (your brand voice, past content, audience data, product positioning), the more specific the output. This is why Alex’s content machine connects to Slack and Notion rather than starting from a blank prompt.
Know what you want before you build. If you haven’t clarified your own thinking, the model can’t either. Using AI as a thought partner to plan before building is a step most people skip, even though it makes the biggest difference.
There’s also a third factor emerging. As models improve, the amount of direction you need to give is going down. Context is becoming more important than prompting technique. “Very soon, the only answer is going to be not enough context,” Alex predicted, “because as these models get better, the amount of direction you need to give is going to actually go down.”
Sanctioned vs. unsanctioned AI usage
The webinar didn’t shy away from one of the most uncomfortable topics in enterprise AI right now: the gap between the tools companies have officially approved and the tools employees are actually using.
“Unsanctioned” doesn’t mean unethical. It often means that a company’s approved tools are too slow, don’t do the job, or haven’t kept pace with what’s available. A new tool that’s 10x better might exist, but the organization’s operations might not have caught up yet.
Zapier MCP is a great option for a sanctioned tool. Whatever coding agent you choose, Zapier MCP gives you a single connection point to 9,000+ app integrations and 30,000+ actions with managed authentication, scoped permissions, and enterprise controls. Your agent gets to do real work across your entire tech stack while Zapier governs what flows through its platform.
Real workflows marketers are building right now
Beyond the content machine, the webinar highlighted several practical use cases for using Zapier MCP with an AI coding tool:
Lead routing QA/QC: Alex built a workflow using Codex, Zapier MCP, and Playwright that audits the entire customer experience for marketing qualified leads. It maps every email, every automation flow, and every sales touchpoint a lead receives, something that previously took 35 minutes per lead to check manually. He ran it on 194 marketing qualified leads to diagnose routing issues.
Scaled ad creative testing: A growth marketer featured on Alex’s podcast built a system where Claude Code mines subreddits for customer language, generates hundreds of ad creatives, pushes them into Meta campaigns, and then builds custom landing pages that match winning creatives. The entire loop runs continuously with $100/day budgets.
Podcast-to-content repurposing: Alex’s content machine includes a podcast promo skill that takes a YouTube link, transcribes the episode, identifies the top five compelling moments, generates hooks, creates drafts across multiple content types, and runs them through the editor board. The whole process is triggered with a single command.
How to get started with AI coding agents
If this feels overwhelming, Alex’s advice is that you don’t need to try to be a hero with your first build. Here’s a practical path even newcomers can follow:
Pick one repetitive task. Look at your week and find something painful, time-consuming, or boring. Lead enrichment, content repurposing, report generation, competitive monitoring. Start there.
Choose a coding agent. Claude Code, Codex, or Cursor. All three work. Pick whichever feels least intimidating and just open it.
Connect Zapier MCP. This gives your agent access to 8,000+ app integrations through a single connection point, with managed credentials and scoped permissions. If you don’t know how to set it up, check our feature guide. Or just ask the agent. It’ll walk you through it.
The goal isn’t to get the most ROI out of what you build. It’s to build enough confidence to keep you going.
Alex Lieberman
Three AI tools every marketer needs right now
Alex recommends these three tools to every marketer:
A coding agent (Claude Code, Codex, or Cursor) as the engine. It builds workflows, connects to APIs, generates content, and automates multi-step processes. Pick one and start building.
Zapier MCP as the integration layer. Instead of installing individual MCP servers and managing authentication for every app, Zapier MCP gives your agent access to 9,000+ integrations through a single, governed connection point.
A call recording tool (Granola, Gong, or similar) to help provide full context (for example, content ideas, customer insights, competitive intelligence, or training data) to your AI agents.
Between these three, you can build a surprisingly complete AI-powered marketing operation.
Where taste beats ideation
One of the more thought-provoking moments from the webinar was Alex’s take on creativity. The common narrative is that humans are still needed for ideation. Alex isn’t so sure.
“I don’t think any ideas are truly novel,” he said. “Every idea we have on earth has priors that we are combining in a new or different way. It’s like a DJ mixing music. The original soundtracks aren’t new, but remixing them creates novelty.”
If that’s the case, then LLMs can create novelty the same way humans do. What matters isn’t who generates the idea. It’s who has the taste to recognize which ideas will actually resonate with an audience and the judgment to steer the process in the right direction.
That’s the role marketers should be leaning into: not as content creators or even ideators, but as curators with the taste and context to guide AI systems toward outputs that actually land.
AI coding for marketers FAQ
Do I need to be technical to use Claude Code, Codex, or Cursor?
Nope! These tools accept plain English instructions. You describe what you want, and the agent figures out how to build it. If you encounter technical terms you don’t understand, you can ask the agent to explain them. As Alex put it, “You don’t have to know what it is. You just have to be able to follow directions.”
What’s the difference between Claude Code, Codex, and Cursor?
All three are AI coding agents that can build workflows, connect to services, and automate tasks. Claude Code is built by Anthropic and runs in a terminal or desktop app. Codex is OpenAI’s coding agent with a desktop app. Cursor is a standalone code editor with AI built in. For marketing use cases, all three are capable. Pick whichever interface feels most comfortable.
How long does it take to build a useful workflow?
The initial build typically takes about 30 minutes. The real time investment is in the refinement loop: providing feedback, adjusting prompts, and training the system on your preferences. Alex estimates he spent about 25 hours total on his content machine, with the vast majority of that time spent on prompt refinement rather than setup.
What is Zapier MCP and why does it matter?
The Model Context Protocol (MCP) is an open standard that lets AI agents connect to external tools and services. Zapier’s MCP server gives any MCP-compatible agent access to 9,000+ app integrations and 30,000+ actions through a single connection point. It handles authentication, rate limiting, and error handling. Actions that flow through Zapier also benefit from the platform’s scoped permissions and audit logging. So instead of managing API keys for every individual service, you manage one connection to Zapier.
Is it safe to connect AI agents to my business tools?
When you route actions through Zapier MCP, those actions run through Zapier’s security infrastructure, which can include managed credentials, scoped permissions, and audit logging. You control exactly what your agent can do in each app. The security of the agent itself (its plugins, its system access) is still your responsibility, though. Zapier provides a governed integration layer, not a blanket security solution.
What if my company hasn’t approved these tools yet?
If your company hasn’t approved a specific tool, that doesn’t mean the tool isn’t valuable. It often means the organization’s operations haven’t caught up yet. The best approach is to start with low-risk, high-visibility projects that demonstrate value, and use platforms like Zapier that provide the governance controls your IT team cares about.
How do I prevent AI-generated content from sounding generic?
Two things: context and feedback. Feed the model your existing content, brand voice guidelines, audience data, and specific examples of content you consider high quality. Then provide detailed feedback on every draft. The models are good enough now that generic output is almost always a context or feedback problem, not a technology limitation.
Will AI replace marketers?
Probably not—at least as of today! But marketers who use AI effectively will outperform those who don’t. The webinar made a clear case that the human role is shifting from execution to curation: providing taste, judgment, and strategic direction while AI handles the production. As Alex framed it, “taste sits above ideation” in the hierarchy of what makes humans valuable in an AI-powered workflow.
