In 2024, my job at Copy.ai stopped being about traffic. We had run an SEO sprint up to roughly 350,000 organic visits a month, and at the peak of it the mandate changed: traffic was never the goal, pipeline was, and we needed enterprise pipeline now that we’d moved from a B2C product to a B2B one. So we rebuilt the system, deliberately let a huge chunk of that traffic die off, and over the next eighteen months drove millions in pipeline off the content engine without a human drafting anything on the website and without an ad budget. A human stayed in the loop on strategy and judgment, but the production ran on systems.
Watching that traffic fall off a cliff was genuinely scary, and the only reason it worked is that I had leadership who cared about the right number. That experience is the whole reason I started this podcast, because it taught me the difference between using AI and building with AI. Most teams right now are doing the first thing and calling it the second.
The license is not the plan
Here’s what “using AI” looks like in 2026. Leadership reads a few breathless LinkedIn posts, catches the fear, and hands everyone on the team a ChatGPT license. That’s it. That’s the strategy.
“A lot of teams ended up getting a ChatGPT license and called it a strategy.”
What happens next is rarely what leadership pictured. Some of those employees use the license to produce disconnected, off-brand messaging faster than they ever could before, which is worse than slow because now there’s more of it. Others quietly use their subscription for AI therapy, which is funnier and honestly probably better for their wellbeing. Either way, the company bought a tool and mistook the purchase for a system. Meanwhile that same team is running somewhere between 25 and 60 marketing tools, most of them older platforms that bolted an AI feature on top and started calling themselves AI-native, none of which actually talk to each other.
The numbers around this are not subtle. Marketing budgets have flattened to under 8% of revenue, 59% of B2B marketers are being asked to do more with less, and 42% have been through a real restructuring, which in plain terms usually means the team got smaller and the survivors were told to not just cover the old workload but beat it. Under that kind of pressure the instinct is to go buy another tool, and that instinct is exactly backwards.
Incremental growth and structural growth are not the same thing
Using AI tools gets you incremental growth, and building AI systems gets you structural growth. The two look similar for about a quarter and then they split apart completely.
A prompt is a task. You ask, it answers, you copy the result, you move on. Shave 20% off the time it took to write a first draft and you have a faster version of the same job, which is fine, except that everyone has the same tool and gets the same 20%, so it stops being an advantage almost immediately. A system is a chain: one input, ten outputs, and no blank page anywhere in the process. Research feeds drafting, drafting feeds editing, editing flags the weak spots, and a human spends their time on strategy at the front and judgment at the end rather than staring at a cursor. That’s structural, because it changes the shape of what a team can produce, not just the speed at which they produce it.
This is the whole idea behind the book I wrote, Pipes Before Chocolate. The name comes from realizing that a Ford factory and Willy Wonka’s factory are both factories, and the only real difference is the creativity of the inputs. Ford automated for sameness and Wonka automated for wonder, but to get the chocolate river flowing, somebody still had to lay the pipes first. The chocolate is the magical output everyone wants to talk about. The pipes are the boring infrastructure nobody brags about, and they are the only reason any of it flows.
The fastest way to spot a team optimizing the wrong thing
The clearest sign that a team is using AI rather than building with it is that they’re still worshipping traffic.
Traffic is the easiest metric in the world to grow and the least useful when it’s the wrong traffic. My favorite proof of this is what happened to a couple of the big SEO tools, which I won’t name but which you can probably guess, that spent years generating millions of pageviews off keywords that had nothing to do with their product. They’d run an automated SEO audit of an adult website, the resulting page would rank for that site’s name, and a flood of people searching for something would land on a B2B SEO platform. Hundreds of thousands of visits from that one keyword alone, and I have a very hard time believing it produced a single paying customer.
That is pageviews as a vanity metric in its purest form, and it’s the same trap a ChatGPT license walks a team into: more content, more traffic, more visible activity, and no more pipeline than they started with. Pipeline over pageviews is the principle that matters, because the point of a system is to attract the right people and convert them, not to win a traffic contest nobody is actually scoring.
How to tell whether you’re using AI or building with it
If you want to know which side of this line your own team is on, three questions sort it out quickly.
Does one input produce one output, or ten?
This is the cleanest test. If a person prompts the model, takes the single answer, and pastes it somewhere, you’re using AI to do a task faster. If one input (a transcript, a brief, a customer call) flows through a chain and comes out the other side as an article, a set of social posts, a sales talking point, and a newsletter draft, you’re building with it. Count the outputs per input and you’ll know immediately where you stand.
Are you shaving time off tasks, or removing the blank page?
Speeding up a task you already do is incremental and easy to copy. The structural shift is when the blank page disappears entirely, when nobody on your team starts from zero because the system always hands them a strong draft to react to. If your people are still opening empty documents every morning, you’ve bought tools, not built a system.
Does the work compound, or reset every Monday?
A system gets smarter as it runs, because you’re feeding it the two things a competitor can’t copy: your team’s lived experience and your company’s proprietary data. Code those into the system (the thought leadership, the weekly learnings, the way real customers actually use your product) and every new piece of content gets more specific and harder to replicate. If your output is just as generic this month as it was last month, nothing is compounding, and you’re renting capability instead of building a moat.
The honest part
I’ll tell you where this breaks, because the people selling AI as a magic button never will. The first version of my system was ugly. It still breaks. I’m optimizing something almost every week, and there are pieces I haven’t solved yet. My two real strengths here are stubbornness and optimism, which mostly means I don’t give up on a problem I believe AI can solve until it’s actually solved. If you decide to build, expect it to be messy, expect some give and take, and give yourself a clear measurement and an end date, because it’s always safer to leave something as an experiment than to ship it, and that safety is exactly how a project quietly becomes your whole year with nothing to show for it.
There isn’t much middle ground left, though.
“You can either build or you can drown.”
You can build your own systems, you can hire someone to build them with you, or you can keep collecting tools and drowning in experiments that never make it to the finish line. The teams that pull ahead over the next few years won’t be the ones still debating which tool to buy next. They’ll be the ones who put their lived experience and their proprietary data into a queryable system that connects every go-to-market function under one roof. I chose to build, and this podcast is just me showing the work.
This is the foundation episode of Barely Shipping. The full episode is on YouTube, and the free book it’s based on, Pipes Before Chocolate, is here. If you’d rather have the system built with you than build it alone, that’s what I do.