Why I Sell Shovels in the AI Gold Rush

Every few weeks somebody asks me the same question. Sometimes it's a friend, sometimes a client, once it was a stranger at a dinner table. You live and breathe this AI stuff, they say. You're testing new models the week they drop. So why are you running an agency and fixing other people's businesses? Why don't you build your own startup and keep the upside?

Fair question. My honest answer is a story from 1849.

When gold was discovered in California, a merchant named Sam Brannan did something clever. Before the news spread, he quietly bought up every pick, shovel, and pan in San Francisco. Then he ran through the streets waving a bottle of gold dust, shouting that there was gold on the American River. The rush he kicked off made him California's first millionaire. Twenty-cent pans were suddenly going for fifteen dollars. He never mined an ounce.

Here's what stuck with me about that story. In the gold rush, there were people who made money and people who were chasing money. The miners were betting on a specific spot on the map, and almost all of them guessed wrong. The merchants never had to guess. They weren't betting on where the gold would turn up. They were betting on the rush itself: a flood of money in motion, thousands of people showing up who'd need tools, food, and help, no matter what happened out on the rivers. The trend was the bet, and the trend was real.

That's my answer. I don't know which startup I could build that wouldn't get blown out of the water by Google, Microsoft, Apple, or Anthropic within a couple of quarters. And I follow this space closer than almost anyone I know. I'm reading release notes at two in the morning. I still can't call it. Nobody can. The ground moves too fast.

But here's what I do know with complete certainty: right now, today, I can bring real value to my clients. I can save a team hundreds of hours. I can build systems that work this month, on this month's models, and rebuild them when better ones ship, because better ones always ship. The map is blank, but the trend is very real. That's the shovel business. I'm not betting on where the gold lands. I'm betting on the rush.

Which brings me to a phone call I had last week.

A good friend of mine just got back from one of his industry's big conferences. He runs a massive company with his partners, the kind of operation you measure in billions, and the conference was buzzing with what every conference is buzzing with right now. AI in real estate, AI in healthcare, AI in everything. He came home with a clear instinct: hire a big firm. Find a serious name that will follow the latest models and come up with solutions for the business.

I get that instinct completely. At his scale, the oldest name feels like the safest bet. That's the whole point of a track record. It turns uncertainty into a brand you can point to when the board starts asking questions.

And that's exactly where I stopped him. Because here's the uncomfortable part: in AI, right now, nobody actually has a track record.

Look at the numbers. BCG says roughly a quarter of its revenue, about $3.6 billion, now comes from AI work, and McKinsey says 40% of its projects touch it. Serious money, serious names. Meanwhile Gartner expects over 40% of agentic AI projects to get canceled by the end of 2027. And those aren't weekend hackathons getting canceled. They're projects run by the most credentialed teams money can hire.

Both things are true at once, because a track record needs a stable game to pile up in, and this game keeps resetting. This spring, in one three-week stretch, one lab shipped two flagship models and another announced its next. Every strategy deck written before those three weeks aged in real time. A hundred years of consulting experience is experience in a world that held still long enough to learn it. Nobody's got that here. The big firms are prospecting like everyone else, just in better suits.

So who actually knows this stuff? The people who touch it every day. The field moves so fast that the real knowledge lives on YouTube, Reddit, and X, in the release threads, with the people who tested something last night and posted what broke. People like me, admittedly, so discount accordingly. But that's the honest answer. A consultant reading reports from three or four months ago isn't a quarter behind. They're two model generations behind. I keep a running list of the AI tools that survive my own workflow, and it turns over faster than any procurement cycle could track.

Here's the part that sounds extreme but isn't. If I had to choose, I'd rather hand a serious company's AI strategy to someone who's never heard of their industry but deeply understands the machine, than to a domain veteran with a stale AI playbook. That sounds backwards until you notice what each one is missing. The outsider is missing industry knowledge, and the company is drowning in industry knowledge. There are hundreds of people in the building who have it. What nobody in the building has is fluency with the machine. Hire for the scarce thing. You already own the abundant one.

And that fluency is made of little things that never show up in a deck. Half the craft of prompting isn't what you ask the AI to do. It's what you tell it not to do, and what you show it versus what you hold back. These models are agreeable by default. Tell one your theory and it'll find a way to agree with you. So you blind it, like a good study: when I want an honest audit, I never reveal what I'm hoping to hear. When I want a claim checked, I ask a fresh session to tear it apart, not confirm it. The systems that survive production are the ones built to argue with themselves. One process does the work, another attacks it, and nothing ships until the work survives the fight. None of that comes from a framework. All of it comes from testing.

So here's the punchline my whole answer has been building to: nobody knows what the heck is going on. Not the hundred-year firms, not the labs shipping the models, not the boutiques running open source, not me. It's the Wild West, and it's new to everybody.

In 1849, that sentence wouldn't have scared anyone. Of course nobody knew where the gold was. That was the whole nature of a rush, and everyone priced it in. Our mistake today is pretending otherwise, and paying billions to people who are willing to pretend along with us.

So don't buy predictions. Buy small experiments with fast feedback, from someone who's visibly out on the terrain. Ask them what they tested last week with their own hands. Ask what they've changed their mind about this month. If the answer is nothing, they're not paying attention.

I still don't know where the gold is going to land. Neither does anyone offering to sell you the map. That's exactly why I sell shovels. And it's why I sharpen mine every single day.