The GPT wrappers are coming

In search of moats and long-term defensibility.

Here’s 2025 prediction: we’re going to see the first really successful, AI-native, “GPT wrapper” companies.

Just a few months ago there was nothing worse for an up-and-coming AI company than being labeled a ChatGPT wrapper. If you didn’t build your own model from scratch, how could you say you had a moat relative to Gemini? Wouldn’t OpenAI eventually just eat your business? Couldn’t customers just do whatever you’re doing themselves with Claude?

This was always a bit like saying in 2010 that companies built on AWS had no moat because they didn’t run their own data centers, but that didn’t stop it from being the dominant narrative among the tech investor class for the last two years. In the world of infinitely reproducible software, there are very few companies that have true technical moats. Sustainability for startups is much more likely to come in the form of a thoughtfully-designed product with high retention and founders that have the domain expertise needed to sell to a specific audience better than anyone else.

That’s more true than ever with AI. Whether you believe progress at AI labs is slowing down or not (and FWIW, it’s not), we’re just beginning to figure out how to apply the technical breakthroughs of the last few years to real-life use cases.

Despite the productivity gains of the SaaS era, most companies still have wildly inefficient workflows – and we now have a powerful new toolkit to address the fuzziest, most non-deterministic bottlenecks. There is staggering opportunity for purpose-built AI agents everywhere you look, and there is no replacement for doing the hard work of deep user and market research.

For what it’s worth, that’s the approach we’ve taken with Norby. Sam, Steven, and I know the world of e-commerce and digital marketing really, really well because we’ve spent years working in the space. Last year we embedded directly with Shopify brands and performance marketing agencies, studying workflows and figuring out painpoints, and then taking our learnings and building thoughtful product experiments. We focused specifically on the parts of marketing and ops workflows that take the most time, have the most friction, and that no one particularly likes doing – crunching numbers, pulling data, building spreadsheets and the like. We figured that if we could streamline everyone’s least favorite parts of the job we’d have something really valuable on our hands.

What we’ve built today feels like it shouldn’t be possible, but it is. Norby is like an always-on, 24/7 data analyst and marketing expert rolled into one. It’s insanely good and there’s nothing else on the market like it – and there won’t be until someone else rolls up their sleeves to do the same hard work we did.

That’s where the moat is – the same place it’s always been – and I have a feeling it’s going to be a really big year for companies that figure this out sooner rather than later.