AI Cold Email Stack vs Offer-First Outreach

What I Watched This Week

I watch a lot of cold email content. Most of it is rehashed advice dressed up with new tool names. This week two videos stood out for different reasons. One is a technical deep-dive into replacing nearly every tool in a cold email stack with custom-built software. The other is a clear-eyed breakdown of why most outreach fails that I wish more people would absorb before they ever touch a sequencer.

They seem like they're about different things. By the end of this breakdown, you'll see they're actually saying the same thing from opposite angles.

Video 1: Replacing Clay, Zapier, and Your Entire Stack With Claude Code

What He Built

This video opens with a claim that immediately got my attention: tens of millions of cold emails per month, and he rebuilt the entire operational backbone of his agency on top of Claude Code and Devin AI without knowing a single line of code himself.

That's not a small claim. Let me break down exactly what he describes building.

First, he built his own analytics platform called Outbound.io. The core argument is that when you depend on Instantly, Smartlead, or Email Bison as your sequencer, you lose all your historical send, reply, and signup data the moment you cancel your subscription. So he built a permanent record that stores every campaign across every client. That's smart. Data ownership is underrated in this industry.

Second, he built a lead waterfall that he says has replaced Clay entirely. The key piece of this is a backend database that checks whether he already has a verified email for any given contact before hitting a paid enrichment provider. His number: about 80% of contacts on any new list are already in his internal database. At 100,000 lookups per second. If those numbers are accurate, that is a genuinely impressive infrastructure play that drops his data spend dramatically. He mentions his old Clay bill at their grandfathered pricing would now be around $217,000 per week if he were a new customer enriching at that volume. That's the kind of math that makes you want to build your own system.

Third, he built a full client onboarding automation. When a contract is signed, it fires a webhook that adds the client to the CRM, creates their Stripe subscription, sets up projects in Linear, and generates onboarding documents. No human intervention needed.

He also built an out-of-office Slack auto-responder that reassigns tasks in Linear to teammates when clients tag him during vacation, and a campaign pause-and-resume system for holidays that logs which campaigns it touched and automatically restarts them afterward.

All of this runs on his own servers. He owns the code. He owns the data.

The Workflow He Describes

His actual process for building with AI coding agents has three phases worth noting.

Phase one is planning mode. He spends the majority of his time here, not in building. He talks to Claude Code or Codex in planning mode, asks it to criticize his ideas before anything gets written, and documents everything in Linear through an MCP integration so the AI has persistent context across sessions. This is important because these tools hit context window limits during long builds and start compacting information, losing detail. His fix is to keep the documentation outside the AI's working memory so it can always pull back the full picture.

He uses three tools at different price points. Claude Code and Codex are both around $200 per month and he says he never hits his usage limits on either. Devin AI runs him several thousand dollars per month and he says it's worth it because it understands large codebases better than the others.

My Take

I've been watching the AI-builds-your-stack trend carefully. Here's where I land on it.

The data ownership argument is correct and I'd push it further. Every agency and operator running campaigns through a third-party sequencer is building on rented infrastructure. The day the tool changes pricing, gets acquired, or goes down, your institutional knowledge of what worked goes with it. Building your own record-keeping layer around whatever sequencer you use is something anyone can do without replacing the whole stack. That's the insight worth taking from this video even if you're not going to build custom software.

The lead database waterfall is the most technically impressive piece here, but it only becomes valuable at scale. If you're sending a few thousand emails a month, you're not buying enough enrichments for the deduplication savings to matter. If you're processing tens of millions of contacts, absolutely build something like this. The 80% hit rate on already-owned contacts is the kind of efficiency gain that changes unit economics.

On the tooling itself: I run ScraperCity and we've built our own pipeline around it precisely because depending on Apollo's export limits and interface was slowing everything down. Our email finder and email validator exist specifically because we wanted to own that part of the workflow rather than pay per-credit to a provider that could change pricing tomorrow. The instinct to own your stack is right. The question is whether you're at the scale where it pays off.

Where I'd push back: the video is light on what this actually costs in time. He describes spending most of his time in the planning phase, and while he doesn't know how to code, he clearly understands his own operations deeply enough to spec what he wants built. That expertise is the real bottleneck here, not the AI tools. Most operators don't have a precise enough mental model of their own workflow to describe it to an AI clearly enough to build it. If you try to shortcut that, you end up with code that technically works but doesn't solve the real problem.

What's worth implementing right now: the planning discipline. Before you write a single prompt, document what you actually want. Ask the AI to critique your plan. Most people skip this and then wonder why the output is wrong. For a practical starting point on the full tech stack side, see our cold email tech stack guide.

Video 2: Why Your Channel Is Not the Problem

The Core Argument

This video has under 1,000 views at time of writing. It should have more. The main point is something I've been saying for years and it never stops being true.

Most cold outbound fails because the prospect does not want to buy what you are selling.

That's it. That's the whole thing. The rest of the video is just variations on proving this point through examples.

The example that opens the video is a client selling a video marketing product to bankruptcy lawyers. They were getting 5 to 10 calls a month showing up, which is actually decent for that niche. But nobody was buying. Why? Because the market didn't want the video marketing product. They wanted leads. The channel was working. The offer was wrong.

His analogy is sharp: imagine standing outside a law conference talking to partners as they leave. You pitch your video production tool to the first person. Not interested. Second person. Not interested. You do this for three days, 3,000 people. Nobody wants it. What do you do? You change what you're offering. You don't stand outside a different conference. You don't switch from talking to people to handing out flyers. You change the offer.

But when 10,000 cold emails get no traction, people say cold email doesn't work. They switch to cold SMS. Then Facebook ads. Then referral programs. The channel is not the variable they should be testing.

The Trust-Demand Framework

He breaks down outbound success with a framework I want to give him credit for because it's clean and accurate.

Picture two variables, each running from zero to 100. One is trust. One is demand for your offer. For outbound to work, the sum of both has to cross a threshold of 100. That's the analogy, not a hard formula, but it holds up in practice.

High trust means you can get away with a weaker offer. If your business mentor texts you about something they're working on, you'll give it serious consideration even if the pitch isn't perfect. A cold SMS from a stranger needs a near-perfect offer to get any response at all, because trust starts at zero.

The implication is that if your offer is weak, you need to find a way to increase perceived trust. If you have high trust, you have more room to experiment with the offer. Ideally you're building both.

His recommendation for building trust in cold email specifically is video. Not because video is magic, but because richer media builds trust faster. Plain text is the lowest-trust medium. A video of a real person talking about your specific prospect, overlaid on their website, moves the needle toward something that feels more like a real interaction. He walks through a two-step sequence where step one is a plain email and step two, six days later, is the personalized video version. The delay is intentional because the video costs money to produce and they want to reserve it for people who didn't respond to the first touch.

He mentions sending tens of millions of emails per month and getting 6% book-call rates at the beginning of this approach for their restaurant marketing agency, which lines up with numbers I've seen work. The first Cold Email Manifesto version of this math was 20 emails producing 8 meetings in one day. High-quality targeting plus a compelling offer compresses those ratios dramatically.

The Offer Research Loop

He describes the process for finding a working offer this way: make a guess at what has demand, send outbound to get on calls, then use those calls as market research rather than sales calls. Don't try to close. Try to understand what they actually want to buy. Then build the offer around what you learn.

This is right, and it's the part most people skip. They write an offer based on what they think the market wants, get low reply rates, assume the email copy is the problem, rewrite the copy, get the same low reply rates, and spiral. The fix is getting on calls with people who said no or didn't respond and asking what they would have wanted to hear.

His point about complex offers is also worth flagging: if you say your product is too complex to explain in a cold email, the answer isn't to write a longer email. The answer is to simplify the offer or sell a free lead magnet that gets them on a call first. Complexity is not a feature in outbound. It's a barrier.

My Take

I agree with everything in this video. That's actually pretty rare for me when watching other people's cold email content.

The biggest thing people miss when their outbound isn't working is that they optimize the wrong variable. I've personally sent millions of cold emails. When something stops working, the first question I ask is whether the offer still matches what the market actually wants to buy, not whether I should tweak the subject line. Subject lines and copy are important. If you want frameworks on that, our top 5 cold email scripts are a good starting point. But none of that matters if the underlying offer is misaligned with demand.

The trust-demand framework is a useful way to think about channel selection too. Cold SMS has high reply rates but low inherent trust, which means your offer has to be exceptional and your targeting has to be precise. Cold email with a personalized video raises the trust ceiling. In-person meetings raise it further. The question is always whether the cost of building more trust is worth it relative to how strong your offer already is.

The one place I'd add nuance: he says all channels work and the problem is always the offer. That's largely true, but channel selection does interact with the offer in ways that matter. Some offers work better over certain channels because of where the prospect's attention is when they receive the message. A B2B software offer lands differently in a LinkedIn DM versus a cold SMS. Both can work, but the context shapes how the offer lands. Don't use channel selection as an excuse to avoid fixing the offer, but do think about channel-offer fit when you're building the campaign.

For the follow-up side of this, which matters a lot when you're running a two-step sequence like the one described, our cold email follow-up templates have examples that work without being annoying.

The Pattern Connecting Both Videos

Here's what ties these two videos together even though they look like they're about completely different things.

Both are fundamentally about infrastructure versus output confusion.

The first video is about building better infrastructure so you can operate at scale without bleeding money to third-party tools. The second video is about understanding that no infrastructure, however sophisticated, fixes a misaligned offer.

I see people fall into both traps constantly. There are operators who spend months building beautiful automation stacks and never send a single campaign because the infrastructure is never quite finished. And there are operators who send millions of emails and wonder why they're not getting results because they've never stopped to ask whether what they're offering is something people actually want.

The right sequence is: get the offer right first. Validate it manually with small batches. Then systematize and scale. Build the infrastructure to serve a proven process, not to delay proving the process works.

What does that look like in practice? Send 50 emails manually with a sharp offer. If you get replies and calls, now it's worth building the automation around it. If you get silence, change the offer before you build anything. This is the same process I've used to help over 14,000 entrepreneurs generate more than 500,000 sales meetings. The system works when the offer is right and the system supports the offer. Not the other way around.

What to Actually Do This Week

If you take one thing from the first video: start documenting your outbound process in enough detail that someone else could run it. Not so you can hand it to an AI to code. Just so you understand it clearly enough to improve it. The operators who can build what that first video describes are the ones who already understand their workflow so well they could describe every step to a stranger. That clarity is the real asset, not the code.

If you take one thing from the second video: before you change your subject line, your sender name, your sequencer, or your channel, ask whether the offer you're making is something the target market actually wants to buy. Not whether you believe in it. Whether they want it. Run a small test with the most direct version of the offer you can write. If it moves, build on it. If it doesn't, the copy is not the problem.

Both of these come down to the same discipline: test the right variable, in the right order, before you invest in scaling.

If you want to see what a sharp offer looks like in actual email copy, the killer cold email templates page has examples across different niches that you can reverse-engineer.