Stop Buying Cold Lists. Build an Agent That Watches for the Signal.

Every cold list you buy is a guess. A hand raise is not.

It is 9 p.m. On a Tuesday. Priya is closing her laptop after a twelve hour day running a six person brand and web design studio in Toronto. The work is good. The team is good. The problem is the same problem it has been for two years: some months the phone rings, some months it does not, and there is no pattern she can point to.

She has tried the usual things. A cold list from a data broker. A few hundred dollars on ads that got clicks and no calls. A referral program nobody remembers to ask for. Each one is a guess dressed up as a strategy.

Here is what changed this week. A video crossed her feed, and inside forty four minutes it explained something Priya had never actually seen named out loud: the businesses pulling ahead right now are not emailing more people. They are watching for the people who already raised their hand, and reaching out while the hand is still up.

What does it mean to watch for a signal instead of buying a list?

A cold list is a guess about who might want what you sell. A signal is proof that someone already cares, today, about the exact problem you solve.

On Greg Isenberg's podcast this week, marketer Cody Schneider walked through exactly this shift, live, building two working agents on screen with Claude Code. His opening line is the one worth sitting with: reply rates are falling across every channel because AI generated volume has flooded every inbox. The fix is not writing a better cold email. It is not sending a cold email at all, until you know someone is already listening.

The signal he uses is embarrassingly simple. When someone likes or comments on a post from a well known voice in your industry, they are telling the internet, for free, that they care about that exact topic right now. That is a warmer lead than any list you could buy, because a list tells you who a person is. A like tells you what a person wants, this week.

You are not short on prospects. You are short on timing.

Most small businesses do not have a prospect problem. They have a timing problem.

The plumber has a full list of every homeowner in the metro area. The bookkeeper has a full list of every small business in the state. Neither list tells them who needs help this week versus who needed help eight months ago and already hired someone else.

An agent that watches engagement solves the timing problem, not the prospect problem. It does not find more people. It tells you which of the people who already exist are paying attention right now.

How can I use AI to find new customers?

Here is the build, scaled down from Schneider's version to something a five or six person business can actually run.

1. Pick your ten to fifteen source accounts. Not everyone in your industry, just the creators, company pages, and communities your actual buyers already follow and engage with. Schneider's rule of thumb: ten to twenty well chosen accounts capture roughly 80 percent of the audience that matters in most niches. Chasing the other 20 percent costs more than it returns.
2. Watch who engages, not who posts. A scraping layer like Apify pulls the likes, comments, and reactions on each post from your source accounts. Claude Code or Codex runs a short script that takes a post URL and returns a clean, deduplicated list of everyone who engaged. In Schneider's demo, one post yielded 63 raw engagers in under a minute.
3. Enrich cheapest tool first, not most expensive tool first. A name and a profile link are not a contact. Run the list through a waterfall: a cheap, high volume matcher first, then a mid tier tool for whoever is left, then a specialist tool as a last resort, with a verification step at the end so you never send to a dead address. Schneider's own numbers: fifty profiles yielded 32 verified contacts at the first, cheapest step alone, with the rest resolved further down the chain.
4. Reach out like a person, reviewed by a person. An agent drafts the first message using the actual post the person engaged with as context, not a generic template. A human approves before anything sends. Schneider keeps cold sending on separate inboxes from the main business domain, so a bad reply never touches the address customers already trust.
5. Turn the exhaust into content. This is the second agent from the episode, and it is the one most small businesses skip. Record your own sales calls and team conversations. Let an agent pull out the real insights and draft posts from them. Publish. Watch what actually gets engagement. Recycle the winners. Schneider's team runs this on a 90 day cycle, tool name Ordinal, feeding performance data back into what gets written next.

Isn't this just scraping people without permission?

Ask it plainly, because your customers will ask it too.

The engagement data is public. Someone who likes or comments on a post has already made that action visible to anyone who can see the post, which includes you. What matters is what happens after that. Schneider is direct about the boundary: the enrichment tools should only surface information already available through legitimate business data providers, never personal data scraped from private sources, and every message that goes out should be reviewed by a person before it sends, not blasted automatically at volume.

Keep the same rule OpenAI gives in its own guidance on building agents like this: test on a small batch first, grade the results, and keep a human approving the final message. An agent that finds a warm lead is an assistant. An agent that emails a thousand strangers unsupervised is spam with better handwriting.

An agent is not the same thing as an automation

Schneider draws this line clearly and it is worth stealing. A plain automation runs the same steps every time, no matter what it finds. An agent is code on a schedule with a language model attached at the one or two points where judgment actually matters, deciding whether a profile is worth enriching, or which post is worth replying to first.

You do not need an agent for every step. You need one exactly where a human would otherwise have to sit and think. Everything else can be a script.

If you have not built a first outreach agent yet, that piece walks through the basics. This one is the upgrade: replace the guesswork of who to email with the certainty of who is already listening.

What you have after this

A week in, you have a short list of source accounts and a script that turns one post into a clean list of engaged prospects.

A month in, you have a working waterfall, a small number of verified, warm conversations started, and a sense of which posts in your industry actually attract the people who buy.

Three months in, you have two systems running quietly in the background: one that finds the people already paying attention, and one that turns your own team's real conversations into the content that earns attention in the first place. Neither one needs you to write a single cold email again.

Same team. Same offer. A completely different way of finding out who is ready to hear about it.