Mar 2, 2026 · 54m · startup-ideas
Claude Code & MCPs built my $145K marketing machine
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this masterclass, Greg Isenberg and Cody Schneider demonstrate how to build and orchestrate autonomous AI growth marketing engines using Claude Code, Model Context Protocols (MCPs), and API-first architectures. Through live technical builds, they showcase how lean teams can automate bulk ad generation, outbound prospecting pipelines, and continuous campaign optimization loops.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Greg holds 15.5% of the talking time here. How this is scored →
Greg as informed peer 3.2 Guest teaching 4.1 Guest disagreement 0.6 Greg pushing back 0.8
speaking balance: gold is Greg, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 19:34 Cody defends lightweight programmatic ads over heavy image models
When Greg questions why Cody does not use top-tier image models like Nano Banana Pro, Cody pushes back that code-generated creative costs virtually zero tokens and eliminates brand drift.
Hardest push from Greg ▶ 19:34 Greg challenges Cody on ad asset creation approachGreg directly challenges Cody's choice of building ad creatives in raw React code rather than using state-of-the-art AI image generation models.
Biggest teaching moment ▶ 34:05 Cody explains the hidden pagination and rate limit flaws of raw MCPsCody explains why directly hooking agents to raw ad APIs via MCP results in severe pagination data blindness, demonstrating the necessity of a dedicated data warehouse.
Greg holds their own ▶ 48:50 Greg articulates the strategic demise of traditional SaaS UIsGreg demonstrates high-level strategic expertise by articulating how terminal-based agent harnesses invert the SaaS business model, turning traditional web UIs into expendable features.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Greg as informed peer | Guest teaching | Guest disagreement | Greg pushing back | Why |
|---|---|---|---|---|---|---|
| Episode Learning Objectives and Marketing Stack Overview | 2 | 4 | 1 | 0 | Greg sets the stage by asking what listeners will learn and admitting he does not know what GTM engineering means. Cody explains the origins of GTM engineering from Clay.com and lays out the agent harness mental model. | |
| Setting Up the Growth Agent Development Environment | 1 | 5 | 0 | 0 | Cody delivers a detailed walkthrough of establishing an environment folder, managing API keys, and evaluating software like Salesforce vs HubSpot purely on API robustness. | |
| Automating Inbound Social Engagement and Lead Distribution | 0 | 4 | 0 | 0 | Cody demonstrates running an autonomous LinkedIn comment respondent script via Claude Code while Greg watches the live execution. | |
| Scaffolding a React-Based Bulk Facebook Ad Generator | 1 | 4 | 0 | 0 | Cody dictates the specifications for scaffolding a React-based bulk Facebook ad generator using HTML-to-Canvas and Claude in plan mode. | |
| Building an Automated Podcast Guest Outreach Pipeline | 2 | 4 | 0 | 1 | Cody explains his automated pipeline for scraping marketing podcasts and cold emailing them. Greg asks for clarification on what Instantly is, and Cody explains cold email tooling. | |
| Drafting Notion Documentation and Previewing Local Ad Generators | 3 | 3 | 1 | 1 | Greg playfully teases Cody for typing manually instead of voice transcribing. Cody runs local instances of the bulk ad generator built entirely from React components. | |
| Mining Social Pain Points for Programmatic Ad Generation | 7 | 5 | 2 | 4 | Greg pushes Cody on why he uses React code rather than Nano Banana Pro image generation models. Cody clarifies that code variations cost zero tokens and keep brand consistency, while Greg articulates the two strategic schools of thought in creative testing. | |
| Iterating Creative Messaging and Formatting for Multi-Channel Scaling | 2 | 4 | 0 | 0 | Cody explains porting winning messaging angles to UGC video using HeyGen and bulk publishing directly to Facebook. | |
| Building Slack-Triggered Outbound Scrapers and Agent Jockeying | 4 | 3 | 1 | 2 | Greg comments on the mental friction of context switching as an agent jockey across multiple desktops. Cody confirms his workflow demands buying a computer with more RAM. | |
| Bulk Uploading Ad Drafts and Generating Performance Dashboards | 2 | 4 | 0 | 0 | Cody demonstrates bulk uploading generated ads into Facebook draft sets and prompting Claude to create performance dashboards with custom metrics. | |
| Data Warehouse MCP Integration and Automated Ad Pruning | 3 | 6 | 1 | 0 | Greg reinforces that ad copy is sourced directly from public social pain points. Cody educates listeners on the pagination and rate limit flaws of direct API MCPs versus querying a synced data warehouse. | |
| Designing Autonomous Marketing Feedback Loops and Daily Briefs | 1 | 4 | 0 | 0 | Cody shows how cron jobs can autonomously manage ad budgets and how team members can query live GA4 marketing data via mobile Claude chats. | |
| Deploying Ephemeral Databases, Cloud Infrastructure, and Course Launch | 2 | 5 | 0 | 0 | Cody reveals spinning up ephemeral Postgres databases on Railway to clean and analyze data in 20 minutes before tearing them down, then announces his free GTM engineering course. | |
| The Macro Impact of Autonomous Agents on the Marketing Workforce | 7 | 3 | 1 | 2 | Greg synthesizes the demo into the concept of autonomous marketing, predicting massive workforce displacement and immense leverage for one-person businesses. Cody shares an anecdote of a founder planning to replace 70% of his staff with agent swarms. | |
| Domain Expertise and Vocabulary as the Ultimate Prompting Superpower | 6 | 4 | 1 | 2 | Cody argues that domain vocabulary is the true differentiator for agent output quality. Greg adds nuance, noting that even domain experts struggle because they do not yet know the specific tooling. | |
| The API-First Paradigm Shift and the Demise of Traditional SaaS UIs | 8 | 3 | 1 | 1 | Greg highlights the paradigm shift where APIs become the primary product while SaaS UIs become an optional nice-to-have, citing Sam Altman. Cody enthusiastically agrees, sharing that his company contemplated dropping web UIs entirely. |
Statements from this episode (20)
Schneider: Clay.com coined 'GTM engineering' for outbound data enrichment workflows
“So this is actually like made up by clay.com, which is hilarious. And they originally did it as like a way to explain somebody that like does basically like cascading workflows for like data enrichment to do outbound sales motions over email or slack, or, you …”
Schneider delegates manual keyboard execution to AI agents like Claude Code
“Basically everything that used to be the middle work that we would do, like all of anything that I would do to touch the keyboard, I'm now passing it on to some type of agent harness whether it's Claude code or it's codex or any of these tools. And so my job s…”
Schneider: API robustness determines how he buys all software now
“This is actually how I'm thinking about everything I do now and like how I buy software in particular is how robust the API is.”
Schneider: Salesforce's robust API makes it a better AI foundation than HubSpot
“If you're looking at Salesforce versus HubSpot right now, Salesforce, even though it's like historically a more clunk like clunky CRM, it's actually the better product for this AI foundation because it has a more robust API. So you can do more with it basicall…”
Schneider built an AI agent pipeline to automate podcast guest bookings
“Basically scraped all of the podcasts that were within the marketing category and then built a workflow that goes in cold emails them. And then an agent that responds back to book me on that podcast. This ends up turning into way better performing than I expec…”
Schneider: Code-based React templates beat generative AI for rapid ad iteration
“The only thing I've found with, like, Nano Banana is that I sometimes have trouble, like, getting it to stay on brand, and if I'm trying to just, like, figure out the messaging variations that I'm trying to go after, This can be a faster way to do that.”
Schneider: Infinite ad generation shifts the marketing bottleneck to identifying winners
“The, anybody can go and generate as many of these as they want, right? Like it's literally infinite, but identifying those winners, like you're talking about now becomes the challenge that you're going to face with all of this.”
Schneider is building autonomous agents for end-to-end ad creation and optimization
“Where I'm seeing this head personally is like, I'm going to build these tools. That an agent is going to have, and then it's going to be able to run this process in the background where it's basically has the ability to make new creative. It can publish that c…”
Schneider deploys validated agent workflows onto Railway for perpetual execution
“This is literally how I'm working now. This is like, I'm just jockeying agents across, and then if I can automate them and get them to do like if I can figure out, okay, this is the specific lane that you can focus on, then I'm spinning that up onto a server o…”
Schneider manages 15 simultaneous AI agent windows after six weeks of practice
“Like, and again, this is just how I've been working for the last like six weeks. And it was like, maybe I could have like Two or three of them in the beginning, and now it's like, I'm comfortable with, like, we could have 15 windows open.”
Schneider automates ad testing with daily cron jobs that promote winning creatives
“So like, for example, how I would run this is I would have a test campaign where I'm basically testing new creative constantly. I would have a cron job that's on a daily basis, basically going and turning off the low performers. And then the high performers, t…”
Schneider: AI and Ephemeral Postgres Cut Data Cleaning From Hours to Minutes
“What have used it would have taken me probably five hours historically to like clean the data appropriately. And I smashed that out in like probably 20 to 30 minutes.”
Schneider: Ephemeral Databases and On-the-Fly Software Will Become Standard
“The epiphany I had was like on the fly UIs on the fly databases, like on the fly software is going to become the standard for these people that are like, you know, working at the forefront of this.”
Isenberg predicts autonomous AI marketing agents will cause widespread real job losses
“So I think and then the unfortunate thing is I think a lot of these jobs to be done, and this is where I disagree with a lot of people, is I think that there are, is going to be a lot of job loss. Real job loss.”
Schneider runs an autonomous AI agent for LinkedIn prospecting and cold outreach
“So like I have one that's just like crawling LinkedIn, like as we speak and it's like looking for like ICP and then it enriches them. It writes a personalized email and a cold emails.”
Isenberg: SaaS UIs are becoming nice-to-haves compared to agentic LLM workflows
“But now when you're living in a terminal, for example, and you're using MCPs to talk to LLMs you kinda, you know, the nice to have is actually the UI. The nice to have is the SaaS. The nice to have is going to this website and look pretty. Ultimately what you …”
Schneider is churning from SaaS tools that lack API and UI parity
“There's a thing you can do in their UI. I can't do in their API. And I'm literally about the churn because I'm just like, this is critical for me. And now it feels archaic for me to go and interact with your fucking, like UI to do this outcome, like output tha…”
Schneider: AI tools must fit into any agent harness seamlessly
“You're basically making your agent so that, or whatever it is your tooling is so that it fits into any harness. So whether you're working from cloud iOS or chat GPT desktop or cloud code on, you know, in your terminal or like cursor or, you know, even the UI i…”