Every marketing channel is down right now.
Cold email reply rates: down. Organic reach: down. Paid: more expensive every quarter. Cody Schneider says it straight in this episode: AI slop is flooding the zone and everything is turning into a red ocean.
And then he spends 40 minutes showing, screen shared, hands on keyboard, how his team gets customers anyway. With AI marketing agents that find people who are already raising their hand, enrich them, email them, DM them, and manage the replies. On autopilot.
Last time Cody was on Greg Isenberg's podcast he explained what marketing agents are and why they're the new coding agents. Greg says it was his most requested episode in a long time. So he brought Cody back to show the actual builds.
This is the breakdown, with every tool named and the real costs. Source video first so you can watch along.
The core idea: stop blasting, start watching for hand-raises
Cody's whole outbound thesis fits in one sentence.
Don't cold email people because their job title matches. Cold email people who just publicly showed interest in the exact problem you solve.
On LinkedIn, that signal is engagement. When someone likes or comments on a post about, say, AI video editing for marketers, that's a hand-raise. They're telling you what they care about right now. Not their firmographics. Their actual current interest.
Traditional outbound targets who someone is. Signal-based outbound targets what they just did. That's the difference between "I found you in a database" and "you literally asked about this yesterday."
Greg's reaction mid-episode: "which is insane, by the way. It's impossible to find this." It used to be.
Build 1: the LinkedIn hand-raise agent (an SDR in a box)
Here's the exact system Cody walked through, step by step.
Step 1: Find 10 to 20 accounts your buyers already follow
Go on LinkedIn and find the influencers and companies posting daily in your niche. Creators work. Business accounts work too (Cody literally added Clay's company page to his list).
You don't need hundreds. His take: every niche has a handful of outliers everyone engages with. Monitor 10 to 20 of them and you cover about 80% of the industry's surface area. The marginal return past that isn't worth it.
Drop them in a spreadsheet. That's your lead source.
Step 2: Scrape the engagers with Apify
Apify is a scraping API. One API key, and your coding agent can pull data from LinkedIn, X, wherever. Cody uses the LinkedIn actors from a developer called API Maestro (the most stable ones he's found): one for post comments, one for post reactions.
Then he opens Claude Code, pastes a post URL, and says: extract the engagers using the Apify API key. In the demo it pulled 63 profiles from a single post, deduped, while they talked.
The agent version: a daily cron job that checks each account on your list for net-new posts, then extracts every person who engaged. New hand-raisers flow in every morning without you touching anything.
(I've built something similar for my own pipeline. If you want the beginner version of this exact motion, I broke one down in my LinkedIn AI agent playbook.)
Step 3: Waterfall enrichment to get emails and phone numbers
A LinkedIn URL alone doesn't get you a customer. You need contact info. Cody runs what he calls a waterfall enrichment:
- → Send all profiles to GetLeads.io first (cheapest, accurate). Say 50 profiles in, 32 emails out.
- → Send the 18 misses to Apollo. Maybe 10 more come back.
- → Send the last 8 to Prospeo or Origami.
- → Need mobile numbers? LeadMagic is his pick for phones.
Cheapest tool first, expensive tools only for the leftovers. Chain enough providers and you land around an 80% find rate. Origami will even run the whole waterfall for you from one API call.
Then validate everything through MillionVerifier before sending. It flags each email as good, risky, or bad. Send only to good ones. Skipping this step is how you torch your deliverability with bounces.
Step 4: Buy sending infrastructure (not from your real domain)
Never send cold email from your actual company domain. You'll nuke its deliverability, and that domain runs your business.
Cody's structure: separate domains for cold email, for newsletters, for transactional email, and for the actual company. The cold side runs on burner domains with hosted inboxes from Hypertide (his partner), InboxKit, or Instantly's pre-built inboxes.
The real numbers: about $100 a month in inbox infrastructure gets you roughly 10,000 cold emails a month of capacity. Add Instantly's $97 a month tier for sending. So the whole machine starts at about $200 a month.
For LinkedIn DMs instead of email: HeyReach or Botdog, both with APIs.
Step 5: Let an agent manage the inbox
This is where it stops being an automation and starts being an agent.
Instantly has webhooks. Positive reply comes in, webhook fires, and an agent on your server picks it up with one base prompt: here's the context, your goal is to get this person to book a demo at this link. It answers questions, pushes people deeper into the pipeline, and checks your Cal.com or Calendly to see if the call actually got booked.
And the part that made me sit up: it follows up months later. Cody programs re-outreach to cold leads every 6 months. No human SDR on earth reliably does that.
You can also add a thinking loop before the enrichment: have the agent research each person and their company first, and only spend enrichment credits on real ICP fits.
"Stop paying tokens for what code can do"
My favorite rant of the episode.
Cody's definition of a marketing agent: it's code, maybe a thinking loop, and a live data stream. That's it. Everyone tried to "put God in a box and give it access to a Facebook ads account" and it does not work. Give a general agent your ad account and there's a decent chance it just nukes it.
What works is copying what the best human operator actually does, step by step, and turning that process into software. Then only calling an LLM at the moments that actually need judgment.
His words: "You should not be paying Anthropic to do an API call. You should be paying them to make the software that does the API call." Why burn tokens on every action when cheap compute can run the loop?
His co-founder Max takes it further: the only real agent is a coding agent. Everything else is software the coding agent builds. Greg calls it the software factory era. You describe the marketing system in Claude Code or Codex, it builds the machine, you deploy the machine to a server (Railway works, Cody's stack pipes live data through ClickHouse), and the machine runs daily.
You're not the marketer anymore. You're the person jockeying the machine.
Build 2: the organic content agent (your team posts daily without writing)
Cold outreach not your style? The second build is fully organic, and Cody just shipped it for his own team.
The flow:
- → Source material: a weekly 1-on-1 recording with each team member. "Tell me what you learned this week." Sales calls, Gong transcripts, and Slack threads work too. The ideas have to come from real humans talking.
- → An LLM pulls the insights out of the transcript and writes the posts. Cody says plain Claude Sonnet is good enough for the writing.
- → Ordinal schedules the posts across every team member's LinkedIn account through its API and MCP, and the accounts can even interact with each other.
- → Ordinal's analytics feed back into the agent. Which topics got impressions this week? Write more of that next week. Remix winners.
Why source material matters: ask an agent to "write good LinkedIn content" and you get slop that wastes everyone's time (and LinkedIn literally just shipped a feature that flags AI slop). Original ideas live in real human conversations. The agent's job is extraction and packaging, not invention.
Cody's meta-insight from his own accounts: he remixes the same proven posts every 90 days. Once you know what your audience resonates with, you're not inventing content. You're prospecting for winners and re-running them on a cadence. He says the same thing about product: don't invent a new idea, find what the market already wants to buy and build that.
The math that makes organic a no-brainer
Average cost of LinkedIn ads right now: about $22 per 1,000 impressions.
Even a 500-follower account can pull 1,000 impressions on a decent post. That's $22 of earned media, per post, per person, free. Multiply by a 7-person team posting daily and you're generating thousands of dollars of monthly ad value from conversations you were having anyway.
Cody on his own accounts: "I get paid to build lead pipeline." YouTube pays Greg to market his own company.
Both of them think the social media manager role, as it existed, is dead. The replacement is the social media agent manager: one person running content systems across 10, 20, 100 accounts. If you don't want a personal brand, run a theme page instead. Greg's example: Julian Shapiro didn't grow @DemandCurve, he grew a page called @GrowthTactics, and the agency got the inbound.
What I'd actually do with this
If you're a bootstrapped founder reading this, here's the order I'd steal it in:
- → This weekend: list 10 accounts your buyers follow. One spreadsheet, 30 minutes.
- → Week 1: get an Apify key, open Claude Code, scrape the engagers off one post. Just to feel it work.
- → Week 2: waterfall the list through GetLeads and Apollo, verify with MillionVerifier, and send 50 genuinely personal emails referencing the post they engaged with.
- → Only after replies come in: spend the $200 a month on real infrastructure and put the cron job in the cloud.
Prove the signal works manually before you automate it. The agent just scales a motion that already converts.
FAQ
What is an AI marketing agent?
Code running on a server, a live data stream it watches, and an LLM thinking loop it calls only when a decision needs judgment. Cody Schneider's example: a cron job that scrapes new LinkedIn engagers daily, enriches their contact info, sends cold email, and manages replies until a demo gets booked. It's software doing a job, not a chatbot.
How much does it cost to run this outbound system?
About $200 a month to start: roughly $100 for burner domains and hosted inboxes (Hypertide or InboxKit, good for around 10,000 sends a month) plus Instantly's $97 tier for sending. Add usage-based costs for Apify scraping and enrichment credits on GetLeads or Apollo.
Is scraping LinkedIn engagers and cold emailing them legal?
Buying contact data from brokers like Apollo is legal, and cold email is allowed in the US if you follow CAN-SPAM rules. The EU is much stricter. Cody flags this himself in the episode: he's not a lawyer, compliance changes by region, so check the rules for your market before you send.
Do I need an agent framework to build this?
No. Cody calls most frameworks bloat for jobs like this. A script written by Claude Code or Codex, a cron job, API keys for your tools, and a webhook for replies covers the whole system. Simple solutions break less.
Steal playbooks like this every week
This is exactly the kind of episode I make the Profitable Founder Podcast for: a real operator opening the actual system, with every tool and every price named.
I interview bootstrapped SaaS founders making $100K to $10M a year and get them to open the playbook the same way Cody just did.
