Jonathon's AI Wiki

The LinkedIn-Signal Outbound Agent (Cody Schneider)

7 min read

Source: raw/Marketing_Agents_Masterclass_GROW_your_startup.md — Greg Isenberg with Cody Schneider, youtube.com/watch?v=mD7JpNHLT70. Follow-up to the episode behind Marketing Agents That Run Facebook Ads; this one builds a different system.

Transcription normalization

The auto-transcript renders Apify as “ampify” throughout. Corrected here, consistent with the wiki’s existing Apify coverage (Cowork + Apify).

The build: monitor LinkedIn posts from ~10–20 accounts in your niche → extract the people who engaged → waterfall-enrich to emails and phone numbers → run cold email and LinkedIn DMs → hand replies to an inbox agent that pushes toward a booked demo. The strategic claim underneath it is the part worth keeping even if you never build the pipeline: with AI slop saturating every channel, intent signals beat firmographics.

Key Takeaways

  • The premise is channel decay. “Cold email is getting decimated. Reply rates are down… every marketing channel is down right now” — because “AI slop is flooding the zone and it’s becoming just red ocean everywhere.”
  • The proposed differentiator is the targeting input, not the copy. Instead of firmographics, demographics or psychographics, target on a hand-raise: someone engaging with content on your topic has demonstrated propensity and interest, which traditional outbound filters cannot see.
  • Source selection is deliberately manual, and that is defended. You need only 10–20 accounts whose engagers are your target customer — and “typically the company knows who their target customer is interacting with.” Asked directly why not automate this step, Schneider’s answer is that the volume doesn’t justify it.
  • Business accounts count, not just influencers — his example is scraping engagement on Clay’s posts. This is the non-obvious half of source selection.
  • Search quality test: “is the content that’s being served what your target customer would be interacting with?” His own worked example rejects “AI for WordPress” as too weak a signal and switches to “AI marketing.” He uses the For You feed as the discovery surface, on the grounds that the algorithm already knows what’s relevant to him.
  • Apify is the scraping layer — one API key across LinkedIn, X, and other channels, feeding data “into the context for my agent.” The stated difficulty is actor maintenance: “the challenge with Apify is finding good ones that are actually being monitored and being maintained.” He names the apimaestro LinkedIn actor set as the stable choice.
  • Waterfall enrichment is a cascade of paid databases, each catching what the last missed: getleads.io → Apollo → Prospeo / Origami. His worked numbers: 50 LinkedIn URLs → 32 emails from getleads → the remaining 18 to Apollo → 10 more → the last 8 to a third provider.
  • Never send from your core domain. Sending ~10,000 cold emails from your real domain “will nuke the deliverability of the business URL.” Buy separate domains and host inboxes on them.
  • **Startup cost is small and specific: ~97/mo for the sending platform (Instantly; Inbox Kit priced similarly and “run sales all the time”), plus domains and hosted inboxes.

Where the agent actually is

Schneider is unusually direct that most of this is plumbing. His working definition: an agent is “something that’s doing a job to be done” — and more concretely, “it’s code, maybe some thinking loop, and a live data stream.”

Two places the loop earns its keep:

  1. ICP-fit gating before enrichment. “Okay, agent, research this person and the company that they’re at. How many employees do they have?” — an LLM judges fit against the customer profile, and only those that pass proceed to enrichment and outbound. This is the cost-control step: enrichment is metered per lookup, so filtering before the waterfall is what stops the bill scaling with scrape volume rather than with qualified leads.
  2. The inbox agent. Wired to both the email and LinkedIn inboxes, answering questions and driving toward a booked demo.

Everything else — the daily cron job, the scraping, the enrichment cascade — is deterministic pipeline. That framing is consistent with the workflow-vs-agent distinction this wiki tracks: put the model where judgment is needed, not where a script suffices.

Compliance — carried faithfully, including the hedges

Asked point-blank whether harvesting these emails is legitimate, Schneider says yes, then immediately qualifies it, and the qualifications matter more than the answer:

  • Acquiring the data is “buying data from a data broker, which is legal.” He is candid that the broker layer itself — companies buying and aggregating lists from each other — is “a whole other shady network.”
  • What you do with it is where compliance changes. US cold email is permitted subject to a CAN-SPAM checklist.
  • “The EU has totally different compliance pieces.”
  • Both participants explicitly disclaim expertise: “I don’t think anyone would mistake you for a lawyer” / “take this with a grain of salt… your own research.”

This wiki’s context makes that non-optional. WEO Marketly’s clients are dental practices — a regulated vertical where HIPAA and health-advertising rules recur as design constraints throughout the production stack. A playbook validated for US B2B SaaS outbound does not transfer to patient-adjacent outreach without independent legal review. Do not treat “it’s white hat” as a finding. It is one practitioner’s characterization of his own practice.

Try It

  • Steal the signal thesis first, cheaply. Before building anything, list 10–20 accounts whose engagers are your ICP. If you can’t name them, the pipeline has no input and the rest is wasted effort.
  • Apply the search-quality test to each candidate: would your target customer actually engage with this content? Schneider rejects his own first example on this basis — do the same rather than accepting a broad category.
  • Put the ICP-fit check before enrichment, always. It is the single most consequential ordering decision in the build, because it is what decouples spend from scrape volume.
  • Set up sending domains separately from day one. Deliverability damage to your primary domain is not reversible on a useful timescale, and this is the cheapest mistake to avoid.
  • Budget the enrichment cascade, not just the software. The ~$200 figure covers sending infrastructure only; getleads/Apollo/Prospeo are per-lookup costs on top and scale with volume.
  • Audit Apify actor maintenance before depending on one. Schneider names actor rot as the main fragility; a stale LinkedIn actor breaks the pipeline silently.
  • For regulated verticals, stop and get legal review. See above.

Open Questions

  • What are the actual reply rates? The premise is that this outperforms conventional cold email, but no numbers are given — no reply rate, no meeting rate, no comparison against a firmographic-targeted control. The core claim of the episode is unquantified.
  • What are the enrichment costs per lead across getleads → Apollo → Prospeo? Only the sending infrastructure is priced.
  • What is LinkedIn’s position on scraping engagers? Platform terms are not discussed at all — and this is exactly the permission-not-capability ceiling the wiki documents elsewhere. Technical feasibility here says nothing about authorization.
  • How does the inbox agent avoid becoming the slop it is reacting to? The strategy is premised on channels drowning in AI-generated outreach; the proposed solution automates the reply side too. The tension is unaddressed.
  • Does the signal decay as the tactic spreads? If engagers of the same 10–20 accounts become everyone’s list, the hand-raise stops being differentiating — the same commoditization that produced the red ocean it is escaping.

Graph View

Backlinks