OpenAI Doesn't Give a Crap (About Your Startup)

I was planning a small mixer. Real tiny. Humble.

And then Arjun was all like, “Hey, let’s just have a few people speak,” so OK, yeah. Then Paul said, “Wow, this looks cool. I want to be the sponsor.” I was like, “Sure.” We secured Stifel as a cosponsor, and the RSVPs started to roll in.

However, the venue I had initially booked was going to be way too small and had some issues, so I found a new one.

The bottom line is: We have more space. Apply now to attend the Vibe Your SaaS Founder/Investor Mixer at TechCrunch.

Nineteen tracks of Morrissey and The Smiths, plus a few unique cover versions, this list has been perfectly curated for October's grey skies and Halloween's gothic spirit. Enjoy the music of the patron saint of beautiful sadness this autumn.

If the answer is "not long" and that doesn't worry you, we need to talk.

Most founders under 35 grew up in a world where platforms just... work. AWS doesn't go down (okay, sometimes it does, but rarely). Stripe processes payments reliably (such a great product). Twilio sends your texts (perhaps too reliably). These platforms became core startup infrastructure, a boring, dependable commodity.

So when you build on OpenAI's API or Claude or Gemini, it's the same thing. Right? Just another piece of infrastructure.

Wrong.

Here's what's different about AI platforms in 2025:

They're still figuring it out. OpenAI keeps changing its pricing model. Anthropic is still experimenting with usage limits. Google will forever be launching and sunsetting not just AI products, but also all its products. These companies don't even know what their business will look like in 18 months.

Your use case might become their product. Remember when there were dozens of startups building AI chatbots for websites? Then ChatGPT launched custom GPTs and basically absorbed that entire category. If your startup is "AI that does X," OpenAI might just build X into their next release.

The technology is moving insanely fast. When a new model is released, your carefully tuned prompts may stop working, or they may improve. Who knows. Your cost structure might change overnight. Your competitive advantage might evaporate because the new model is just... better at everything.

There are no SLAs or agreements that protect you. Sure, there are uptime guarantees. But what happens when GPT-5 launches and deprecates GPT-4, and your entire product is built on GPT-4's specific behaviors? You're migrating whether you're ready or not. This is true of all businesses built on someone else’s API.

In 2009, Zynga’s FarmVille had 80 million monthly users. The company was worth billions, heading toward an IPO. Their entire business was built on Facebook's platform. They had mastered the social graph and used viral notifications and other techniques to construct a massive, low-cost distribution channel.

Then Facebook changed the rules.

They killed ALL the viral channels that helped FarmVille spread. They changed how apps could notify users. They modified the news feed algorithm. Zynga's growth engine died overnight. The company's value cratered, and they had to pivot hard into mobile apps, where distribution was still the wild, wild west, and low-cost arbitrage opportunities abounded. (Note, I know because I was deeply involved.)

They eventually went public, but as a shadow of what they could have been.

It’s important to note that Facebook didn't do this to hurt Zynga specifically. They were optimizing their own platform to improve their monetization as they prepared to go public. Zynga's success or failure wasn't even part of the calculation.

So how do you know if you're building the next Zynga, or something more defensible?

It comes down to how dependent you are on the platform. Here's the framework I use to assess AI startups:

🔴 The Wrapper (Highest Risk): Your product is essentially "ChatGPT but for [specific use case]." Your code is mostly prompt engineering. Your differentiation is UX and a narrow focus.

This is like building FarmVille on Facebook. You're completely dependent on the platform, and you're one feature update away from irrelevance. If OpenAI decides to build your exact feature into ChatGPT, you're done. And unlike Facebook in 2009, OpenAI is actively watching what people build on their API and absorbing the good ideas. This is what happened to many early iPhone apps. Remember Flashlight?

This doesn’t mean it can’t be a great business. Kayak seems to have done OK.

🟡 The Orchestrator (Medium Risk): You use AI as a component, but you've built systems around it. Maybe you're combining multiple models, adding your own business logic, and building proprietary datasets. The AI is important, but it's not the only thing.

This is like building Instagram in the iPhone era. Yes, you needed the iPhone's camera and distribution, but you built something genuinely new on top of it. You had your own network effects, your own data, your own experience. When Apple changed the rules, Instagram adapted because they weren't just a thin layer on iOS.

Building an original dataset is a defensible core asset. Aim for that.

🟢 The AI-Enhanced Product (Lower Risk): Who are we kidding? There are no low-risk startups. However, if AI makes your product better, but you'd still have a viable business without it. Your core value is something else: your data, your network effects, your workflow automation. AI is a feature, not the foundation.

This is similar to Spotify adding AI-generated playlists. Nice feature, but Spotify's real value lies in its music catalog, the relationships with labels, user data, and listening habits. If AI were to disappear tomorrow, Spotify would be fine.

What’s unknown is what business AI will destroy, like Chegg, which is struggling.

Once you’ve assessed your risk level. Here's what to do at each stage, from "just trying to survive" to "raising a Series A":

Take the risk. Seriously. Use whatever AI platform gets you to market fastest. Platform risk doesn't matter if you never find product-market fit.

Right now, you're trying to prove people want what you're building. Speed matters more than sustainability. Ship fast, learn fast, pivot fast. Worry about platform dependence after you know people will pay you.

However, and this is crucial, keep a list of every assumption you're making about your platform. Which specific model behaviors are you depending on? What pricing are you assuming? What rate limits? You'll need this list later when things change. And they will change.

Congratulations, you're past the "does anyone want this" phase. Now it's time to stop pretending platform risk will solve itself.

This week: List every AI service you depend on. For each one, write down what breaks if it goes away or changes significantly. Be honest. This isn't for investors, it's for you to ensure your survival.

This month: Pick your biggest dependency and spend one day testing an alternative. Don't overthink it. Literally try to swap in Claude for GPT, or vice versa. See what breaks. You'll discover where you're tightly coupled real fast.

This quarter: Build one layer of abstraction. Not a complex orchestration system. Not some over-engineered solution. Just make it possible to swap providers without rewriting your entire codebase.

Prepare to get asked uncomfortable questions. If investors aren’t all over this, enterprise customers definitely will. You need better answers than "we'll figure it out if it becomes a problem.” Here's what they want to know:

If you don't have good answers, work on that before your next pitch.

The best defense against platform risk isn't clever engineering. It's building something the platforms can't replicate, such as:

Use AI platforms to move fast. But build these moats underneath.

Here's what keeps me up at night for founders: it's not that you might need to switch from GPT to Claude. That's annoying but solvable.

The real risk is building a company where a platform update can commoditize your entire value proposition, where you're renting your differentiation instead of owning it. Ask yourself: If AI models become 10x better and 10x cheaper next year (which they might), does your business get stronger or weaker?

If the answer is "weaker," you're not building a platform play. You're building on someone else's platform. And that AI platform doesn't care about your startup.

On the Gregory and Paul Show, we break down the latest in startups, SaaS, AI, memes, and whatever the internet is debating this week.

This episode welcomes designer Stanley from Be Curious Studio to riff with Gregory and Paul on OpenAI’s Sora launch, the rise of AI “slop tools,” fraud fallout in startup land, Claude’s coding edge, Gemini 3.0, and why everyone is suddenly building “AI friends.” Plus: Reddit stock drama, Canva outages, and the meme of the week.

Perhaps you’ve been following me on social media and reading this newsletter, but you’re hesitant to reach out. I can promise you this: VYS is a fraction of the cost of Y Combinator, and you will get way more value.

Ninety-five percent of startups fail at marketing. Not because they don't have great products. They fail because marketing has become complicated. Expensive. Slow.

What if marketing could be simple?

This is why I created Vibe Your SaaS, an AI-first, micro-fractional CMO service.

But here's what makes VYS different: You get me, a Silicon Valley veteran and three-time startup head of marketing. Every week for one hour. Plus, phone, email, and chat access. No gatekeepers.

You get C-level help without the C-level price tag.

I'm Gregory Kennedy, former creative director, 3x head of marketing, and founder of Vibe Your SaaS. I help early-stage startups build real momentum with strategic clarity, AI-driven execution, and zero BS.

How am I doing? What do you want more of, or maybe less of? Write me back and let me know your thoughts. I will write you back. It’s a promise.