Cody Schneider opened 10 Claude Code windows on a live podcast and ran a marketing department out of them.
In one 54 minute sitting with Greg Isenberg, he generates 100 Facebook ads from real customer pain points, publishes them through the API, builds a tracking dashboard, kills the losers, and has three more agents doing outreach in the background the whole time. The ad creation loop costs him about 1,000 tokens. Basically nothing.
Greg titled the episode "Claude Code & MCPs built my $145K marketing machine." The discipline behind it has a name: GTM engineering.
I've watched it twice. Now that anyone can vibe code a product in a weekend, every founder is stuck on the same question (fine, but how do I get customers?), and this is the most concrete answer I've seen.
Here's the episode, then the full breakdown of what he built and what you can steal.
What GTM Engineering Actually Means
Funny detail from the episode: the term is a buzzword invented by Clay. It originally described people building cascading data enrichment workflows for outbound sales. Scrape a list, enrich it, email it.
Cody's version is bigger. His definition: everything that used to be "middle work," anything he'd touch a keyboard for, gets handed to an agent use like Claude Code. His job shrinks to two things. Having ideas, and polishing what comes back.
And he has receipts. This is the guy who built Swell AI to $1.1M ARR with a 10 person team, spun Drafthorse AI out of its codebase in a weekend and hit $10K MRR in 30 days, and now co-founded Graphed, a data warehouse and dashboard tool. When he talks about marketing workflows, it's from doing this daily for his own companies.
He's been working this way for six weeks at the time of recording. His verdict: he's buying a new computer because he needs more RAM to run more agents.
The Setup: One Folder, One Env File, 20 Tools
The starting point is almost stupid.
Cody lives out of a single folder. Inside it sits an environment file holding the API keys for everything he touches daily:
- → Instantly (cold email) and MillionVerifier (email verification)
- → PhantomBuster (LinkedIn scraping) and Apollo (enrichment)
- → Rephonic (podcast database), Perplexity (research), HeyGen (UGC video)
- → Facebook Ads API, HubSpot, Intercom, SendGrid, cal.com
- → Railway (servers and databases) and Vercel (deploys)
Greg counted around 20 tools by the end of the episode. Every one of them accessible to Claude Code through an API key in that file.
Two additions complete the setup. A voice transcription app (Cody uses Superwhisper) so he can talk instead of type, and Claude Code's frontend design skill so generated UIs look decent.
The deeper shift is how he buys software now: by how good the API is. He tells a story about a friend choosing a CRM. Salesforce, historically the clunkier product, wins over HubSpot in his friend's eyes because the more strong API means agents can do more with it. Cody is about to churn from one tool purely because a feature exists in its UI but not its API.
Greg connects it to something Sam Altman has said: every company is going to be an API company. When your buyer is an agent, the UI is the nice-to-have.
100 Facebook Ads in 30 Minutes, for 1,000 Tokens
The centerpiece demo is the ad machine. Watch the loop:
- → Claude Code uses the Perplexity API to scrape Reddit for the pain points growth marketers actually complain about
- → It writes ad copy variations against each pain point
- → It renders each ad as a React component, exported to PNG with html2canvas. The creative is code, not a design file
- → It bulk uploads everything as drafts into a Facebook ad set through the Ads API
- → A dashboard tracks clicks, cost and CPC, plus impressions by age bracket
- → Claude pulls the live performance data, flags the ads with the worst CPM, and turns them off on command
Ideation, creation, publishing, analysis, and pruning. One person, about 30 minutes, roughly 1,000 tokens of generation cost.
Greg pushed back on the ugly ads: why not use Nano Banana Pro for scroll-stopping creative? Cody's answer is the most useful ad strategy in the episode. Cheap code-generated ads are for finding the message. Once one angle proves it can turn $1 of spend into $1.50, you remix that winner into polished formats and push it toward $3. Testing messaging with expensive creative is backwards.
Anyone who has manually uploaded 50 ad variations knows why this matters. Greg's reaction on the pod was basically PTSD.
The Agents Running While He Talks
The ad machine was the foreground task. During the same episode, Cody has other agents working in parallel:
A LinkedIn auto-responder. He runs giveaway posts ("comment for the asset"). An agent using the Claude Chrome extension opens the post, finds everyone who commented a keyword, and replies to each with the Notion document. It ran 15 minutes on its own while he built other things.
A podcast booking pipeline. Claude scrapes podcast host emails from Rephonic, verifies them through MillionVerifier, and loads them into an Instantly campaign. An agent answers the replies to get him booked. His calendar filled up faster than he expected.
A LinkedIn engagement scraper. Anyone on his team drops a relevant LinkedIn post into a Slack command. PhantomBuster pulls everyone who engaged with it, Apollo enriches the profiles, MillionVerifier checks the emails, Instantly emails them.
A Notion doc generator. Trained on his existing giveaway documents, it writes the next one to match.
By the end, Greg jokes that you finish this podcast with a hundred desktops open. Cody doesn't disagree. He calls the skill "jockeying agents," and admits the context switching was hard until roughly week four.
From One-Off Tasks to a 24/7 Machine
This is the part that changes the game for a bootstrapped founder.
Every workflow above starts as a local script Cody co-writes with Claude. The moment one proves itself, he tells Claude to deploy it to a server on Railway, and it becomes software his whole team uses, running around the clock.
His target end state for ads: a test campaign constantly trying new creative, a daily cron job killing the losers, winners promoted into their own ad sets with dedicated budgets. Autonomous marketing on a loop. (His earlier episode on marketing agents goes deep on that architecture. This one shows the actual keystrokes.)
The throwaway example might be the best one. Cody had a data analysis job that historically meant 5 hours of Excel and pivot tables. Instead, he had Claude spin up a Postgres database on Railway on the fly, pump the data in, analyze it together, push the outputs where they needed to go, and then delete the database. 20 to 30 minutes total.
Cody's own phrase for where this is heading: on-the-fly UIs, on-the-fly databases, on-the-fly software. Built for one job, then thrown away.
And each morning he wakes up, opens Claude on his phone, and asks his data warehouse how many new users hit the homepage yesterday. A conversation with live business data before getting out of bed.
Who Wins, and Who Should Be Nervous
Greg's read on the winners: one-person businesses and small teams. If you're a solo founder competing against a company that needs a meeting to change an ad budget, this stack is your unfair advantage.
A $100K a year head of marketing who masters this could honestly ask to triple their salary, in Greg's words. One person doing the output of ten is a straightforward value case. Greg has a whole episode on that job title; I broke it down in marketing engineer.
The dark side is just as concrete. Cody tells a story about a founder friend who texted him: "I think I'm going to fire 50 people." That was 70% of his team, replaceable in his estimation by agent swarms. An agent per task, a manager agent per pillar.
So what's left to defend? Vocabulary, weirdly.
Cody spent ages failing to describe a texture he wanted on an ad background. Then he found how a designer would say it ("a TV type texture") and Claude one-shotted it. His co-founder Max gets top 1% output from coding agents because he can describe problems with a precise technical vocabulary Cody says he'll never match.
The tools are the same for everyone. The words you bring to them aren't. If you've spent 20 years in your niche, that vocabulary is exactly what makes an agent's output good.
The Starter Version for a SaaS Founder
You don't need 50 desktops on day one. The on-ramp from the episode:
- → Create one folder. Add an env file with API keys for the 5 tools you already use daily
- → Install Claude Code, a voice transcription app, and the frontend design skill
- → Build one workflow, the one you hate most. For most founders that's outbound: scrape, verify, email
- → Babysit it the first run. Let it run in the background the second
- → When it works twice, deploy it to Railway and give the team access
Cody is also giving his entire process away at gtmengineeringcourse.com, free. It had a 100 person waitlist before he'd finished building it.
One honest warning from the episode: agents inherit your accounts' reputations. Cody had an agent running an Etsy shop. It got banned two days before recording. Platforms are still deciding how they feel about this.
FAQ
What is GTM engineering?
GTM engineering is using code and AI agents to run go-to-market work (ads, outreach, content, analytics) instead of doing it manually. The term started at Clay as a label for data enrichment workflows in outbound sales. In Cody Schneider's version, an agent use like Claude Code connects to your marketing stack through APIs and does the middle work while you direct it.
Do I need to be a developer to do this?
No, and Cody isn't one in the classic sense. He describes what he wants by voice, Claude Code writes and runs the code, and he corrects the output. What matters more than coding skill is domain vocabulary: knowing precisely what good marketing looks like so you can describe it. The setup is a folder, an environment file with your API keys, and a transcription app.
What does the stack cost?
The generation itself is nearly free. Cody's 100-ad creative run cost about 1,000 tokens. The real costs are the tool subscriptions you likely already pay for (Instantly, PhantomBuster, Apollo, an ads budget) plus a Claude subscription and a few dollars a month for Railway hosting. Compare that against the $145K in the episode title, which is what hiring this output would run you.
Will AI agents replace marketing teams?
Some of them, and faster than most people think. Cody's founder friend was planning to cut 50 of 70 people in favor of agent swarms. But the episode's other conclusion cuts the opposite way: a marketer who masters agents becomes a one-person team worth multiples of their old salary. The job disappearing is the execution work, not the judgment.
Steal More Playbooks Like This
This was one episode's worth of tactics.
Every week on the Profitable Founder Podcast I sit down with bootstrapped SaaS founders doing $100K to $10M a year, and they walk me through exactly how they get customers. No theory, just what worked with numbers attached.
