I bought DankGeek at the tail end of 2023. It wasn’t a stranger to me, back when I ran a wholesale company called UPC Distribution, DankGeek was one of my customers. During COVID they were thriving, doing around five million dollars in annual revenue, which is genuinely strong for a retail site in this space. But by the time I acquired them, the business was failing, down to about eight thousand dollars a month.

My partner Sean Gang and I were colleagues at Smoke Cartel. We both later left to join major companies in the traditional technology and software development space; I went on to become a product manager, and Sean became an engineering manager. That combination of product and engineering leadership is exactly what we brought back to DankGeek. This was 2023, right when OpenAI had just opened up their APIs to the public, and we were essentially pioneering the use of AI to scale an e-commerce site fast.

And I mean at real scale. We leveraged AI to write tens of thousands of product descriptions, titles, highlights, and specifications, to handle pricing, product categorization, and the entire information architecture of the website. We used AI to scrape manufacturer sites, pull raw product specifications, and pipe them into our own AI pipeline, which then generated the descriptions and every other detail for DankGeek. This was all before these capabilities were baked into Shopify and other platforms, we were a year or more ahead of these tools becoming available anywhere in the market.

But the thing that really made our situation different is that Sean and I both live in this world. We’re users of the product, and I’m passionate about standing up for the rights and freedoms tied to this space. At the same time, we had the technology background, the e-commerce background, and in my case a background in wholesale and manufacturing. I already owned LA Pipes, so we had a warehouse. All the pieces came together.

We rebuilt DankGeek around something the industry sorely lacked: customer service. We focused on the right people, the right support, a strong user experience, high-quality product images and descriptions, and reliable shipping. We brought it back up to a quarter of a million dollars a month in revenue, with some of the best customer satisfaction out there and a great product selection, all backed by technology we keep improving.

What genuinely sets us apart is that the things which are table stakes in most industries good customer service, a quality website, strong user experience, product images and descriptions detailed enough for customers to actually make decisions barely exist among most sites in our space. We obsess over those things, and we use technology to enable them. And this isn’t a platform run by a bunch of suits. It’s run by people who are actually in the space, using the products, who understand the industry and are passionate about it. That evolution is why we now think of ourselves as a technology company that happens to be in cannabis e-commerce. We’re building a new product called ShipGeek, a marketplace connection tool that leverages AI to help other retailers scale the way we did. It’s genuinely revolutionary, and it’s where we’re headed.

Most of our business comes from organic search on Google. Given the space we’re in, we can’t advertise, and honestly, I wouldn’t want our sales to come from advertising anyway, because the moment you stop paying, the sales stop coming. It means a lot more to me to know that people genuinely find our website useful, that they come to us seeking information and products, and they find what they’re looking for. That’s what earns our rankings, and organic search is where we get essentially everything.

We put enormous effort into that: making sure the site has genuinely useful information, that it’s structured so Google can understand it, that we’re technically sound from an SEO perspective. We’ve also invested heavily in outreach, making sure our name gets mentioned out in the real world, because that signals Google to send more eyes our way and hopefully more clicks and purchases.

But here’s the challenge. You can do everything right, everything by the book, everything per their own documentation, and then Google ships a core update or a spam update, and overnight your site loses fifty, sixty, seventy-five percent of its traffic for no discernible reason. Meanwhile, a competitor doing pure black-hat SEO, paying for links, spamming everything, gets rewarded. We’ve watched it happen, to us and to competitors I know intimately, because they’re companies I’ve worked with, worked at, or whose owners I know personally. We play everything straight, we never touch black-hat tactics, and we still get nicked while the people cutting corners outrank us. Then the pendulum swings back, and you’re in the green again, climbing.

We scaled tremendously, and then took a seventy-five percent hit for no reason. We changed nothing. If anything, we just kept doing things better and better. It’s unfortunate that this is the reality: that all of us are essentially beholden to a single company for e-commerce success, and that company is Google.

What I hope changes is that other players enter the space and compete honestly, companies that don’t operate as a black box, publishing one set of rules while behaving completely differently behind the scenes with no transparency. I genuinely think AI and AI-driven search could be the future here: a more honest representation of what’s actually out there, without the gatekeeping Google does.

So if I could choose, I’d love to see more of our business come from AI-driven search. I think that’s ultimately where online purchasing is headed; a great many of these purchases are going to be made by AI agents.

The most recent thing we’ve been trying to hand off to AI is SEO work, technical SEO, content, link building, internal linking. Everything you need to do on-site for SEO, and everything you need to do off-site for it too.

There are a lot of companies offering solutions, honestly, too many to even sift through. Semrush, for example, has an impressive suite of tools, but every single one is an upcharge, and the data feeding their AI recommendations feels limited. I’m not convinced it’s truly triangulating between its own platform, Google Search Console, Shopify, and so on. So we’ve been building our own system, an agent with access to Shopify, Google Search Console, Microsoft Clarity, GA4, Semrush, Ahrefs, and Help Scout customer service data. It combines and synthesizes all of that, then generates a strategy: changes to make to the site, content to publish, where competitors have links we should pursue and how to approach them, what kinds of articles we should be featured in.

Building that has been far harder than I anticipated. People make these things look simple online, and they are, if you’re happy with an out-of-the-box setup and mediocre output. The MCPs give you too limited an access to the data; the one-megabyte file-size cap means pulling SEO data from Google Search Console can only reach back an hour or two before the file blows past the limit. So we’ve been porting that data into our own databases so our systems can access and make sense of it efficiently. I haven’t gotten rid of my marketing team, the system has simply taken that long to stand up. We’ll see if we can eventually change that architecture.

Day to day, though, the team makes heavy use of AI skills we share across the company in a shared repo, skills for reordering products, communication, mailbox cleanup, productivity, marketing, outreach, monitoring reviews for patterns, customer service in Help Scout, product recommendations, and more. We’re constantly rolling out and improving agents in the e-commerce space. And ShipGeek, our B2B platform, connects e-commerce stores to dropshippers, syncing orders and tracking, and using AI to write product descriptions, image alt text, and internal linking. Those are agents we actually built three and a half years ago and are only now refactoring. The space evolves so fast that what you did yesterday is obsolete tomorrow. Sometimes it almost feels smarter to sit on the sidelines, because whatever you build now you’ll likely have to redo in six months.

As for what we handed back to humans: on the consulting side of my company, Seed to Sold, we had a project where management wanted to scan the code bases of a variety of data consumers, to understand how each was consuming data, feed that into another platform, and act on it. But there was so much variety between those repos, in their data schemas and how they consumed and used data, that we simply couldn’t build a one-size-fits-all agent. That work has gone back to a human touch point within a graphic interface.

And that points to the real limitation. The context window may say one million tokens, but AI is genuinely intelligent for maybe the first 150,000 to 300,000. Beyond that, the attention mechanism falls apart and you have to start a new session. It’s still very hard to build systems that leverage memory efficiently, that get smarter over time and handle complex tasks in one shot, rather than an agent with amnesia restarting everything over and over. That paradigm still exists today, so a lot of work still can’t be passed off to AI.

I think a lot of what people believe is possible actually is possible, but it’s not efficiently possible. And until it’s efficiently possible, it’s just hype. Right now, there’s a lot of hype on the internet.

Definitely more people are finding us through ChatGPT and other LLM-driven search, but it’s still under one percent of our total traffic. Google still completely owns the space. None of us really search anymore; we Google. That’s unfortunate, and as I said, I think it’ll be genuinely helpful in the future when AI can not only surface products and brands but actually make the purchases. I hope ChatGPT, or some other LLM-powered search engine, becomes a place people go rather than always defaulting to Google. But for now, it’s still north of ninety-five percent from Google. That’s just how it goes.

In terms of behavior, customers are still doing research before they settle on a platform and then look for products. There’s still that discovery phase at the beginning, and now some of it might happen in AI before they make a purchase decision. But the data is hard to follow. You might ask ChatGPT which online store is best in a given category, then switch to a new tab and enter the site they want directly into the browser, and that traffic shows up as direct, not as coming from AI. We are seeing our direct traffic rise compared to before, and I suspect some of that is exactly this behavior.

As for what they already know about us, the traditional channels still matter, social media, YouTube, blogs, Reddit, Quora. Those all still exist. But at the end of the day, most people still run a Google search, look at the top three results, click one of them, and treat a high ranking as a signal of authority before they make a purchase decision.

I’d say the first thing is: don’t feel like you’re falling behind.

There’s a lot you can do with today’s systems, but the field is evolving so fast that mechanisms or techniques we built even six months ago have already been replaced, or simply baked into the models, by the time you’re rolling into production. So it’s important to figure out how you derive value now, rather than getting caught up chasing some perfect system built on the latest understanding. That understanding is changing far too rapidly. What matters isn’t having the fanciest, most complex system using the newest techniques, it’s that you’re actually solving a real user problem, bringing value to your users and your business, and creating better outcomes.

The foundation of product management still applies: identify the most important problem, the one that solves the most severe user pain point while delivering the largest business value, and solve it. Whether that’s a combination of code and AI, or just AI, or just code, that’s still the most important thing.

But make no mistake: AI has reached an inflection point where its value is finally being fully realized. If you’re not tuned in and taking advantage of it, you’re going to fall behind. Companies whose employees don’t understand how to use these systems will fall behind too. So it’s critical to train and equip your staff to leverage them. None of your staff are as smart as these LLMs, so the goal is to get them partnering with AI to deliver value. If they’re not doing that, your company will be replaced by one that is.