Ok, time for a rant. 90% of "AI Startups" aren't even AI. I've seen founders that use LLMs or have a chatbot somewhere on their site and call it an "AI startup". I've seen people who vibe code or use Cursor/Lovable and then say they have an "AI startup". I've seen people continue to build ChatGPT/Claude/Llama wrappers and call it an "AI startup". If you haven't build a custom model or used meaningful proprietary data then you aren't an AI startup, you're AI Enabled. AND THAT'S FINE. You don't have to be an AI company to solve a problem. So many companies and VCs are going to have poor performance because they have little differentiation and have claimed to build/invest in "AI Companies" and we will soon learn they are not really "AI".
Robelen Ryan
2mo
For peeps like me building with low code, the label fight is a distraction. Users don’t care if it’s AI native or AI enabled. They care if it works on a Tuesday, plugs into the stack, and doesn’t torch ops. Not sure wrappers are a sin, more like a wedge. The moat isn’t the model, it’s whether you ship value fast and it sticks. Great discussion.
Paul O'Brien
2mo
It's the same hype cycle we see every few years as founders try to capitalize on the next buzz: we're blockchain, a community, in the metaverse.... Advised a startup last night to drop saying their solution is thanks to AI; "how" is irrelevant to whether or not they solve the problem.
Dhruv Patel
2mo
Using LLMs or APIs doesn’t automatically make you an AI company. The real moat lies in proprietary data, custom models, or unique problem-solving not just wrapping ChatGPT with a UI.
Jeremy Burton
2mo
Amen. Leveraging AI is now assumed. But how is AI materially changing the solution and/or the economics is the question.
This hits a nerve. The hype cycle blurs the line between AI-native and AI-enabled. Wrappers and LLM plug-ins solve problems, sure, but defensibility lies in proprietary data, models, and system design. The market correction will expose who’s truly building versus branding.
Andrei Yuranau
2mo
The real moat isn't the API call - it's the proprietary data flywheel. Without it, you're a feature, not a company. 'AI-enabled' is a perfectly valid and often smarter strategy.
Chris Oates
2mo
My definition is what is the startup's business model / comparative advantage. I try to be very clear when talking about mine that the core is network effects and aggregation, with AI sprinkled about where necessary, but there's such a temptation to really lean into the AI aspect of it to get attention
Arno Burnuk
2mo
There's a lot of work that goes into proper prompt engineering, building custom ai tools and curating context for your agent. So called "thin wrappers" aren't always that "thin". You could've said lovable and cursor were just "thin" wrappers when they started. Now they probably fine tune their models on their user data, but that comes afterwards when you get a lot of traction.
Prem Sharma
2mo
You’re right. We built the whole stack ourselves at Tandem and it’s the only way the AI actually delivers real value. Wrappers and labels don’t solve adoption—owning the workflow does.
Reminds me of 5 years ago when everyone wanted to be a "blockchain startup" Ironically customers don't even care. Does the product or service actually solve their problem? If X ACTUALLY solves for Y then it could be chimps banging on keyboards in the background for all I care...