OpenAI Didn't Just Kill Your Startup

OpenAI just did founders building AI tools a massive favor but not everyone realizes it yet:

This week OpenAI announced GPTs , custom “versions” of ChatGPT that can be created entirely without code or integrations, work with custom datasets, and can be shared with (and soon sold to) others via a marketplace.

Immediately there were a ton of posts on X about OpenAI “killing” startups:

However this isn’t what’s going to happen. Ironically the exact opposite will happen instead.

This week I’m sharing:

• Why GPTs are a good thing for founders (but are making them anxious)

• How founders can leverage GPTs for their startup

• What we can learn from Apple’s App Store

No, OpenAI Didn’t Just Kill Your Startup

Why GPTs are Making Founders Anxious

You hear this more often from non-founders but even founders can fall into the trap of thinking their product is the same thing as their startup.

This couldn’t be further from the truth, especially at the early stages of a new company.

I haven’t seen exact statistics on this but after seeing thousands of startups throughout my own career I’d estimate that over 50% pivot at least once, regardless of whether they succeed or fail.

The most famous example is probably Slack, which started as a game studio called Tiny Spark and built Slack as their in-game communication tool for their first game. I shared lessons from Slack and 7 more famous pivots earlier this year.

In reality startups are aggregated bets on a market, team, and product.

People on X act like product is the most important whenever a big company releases something new.

And often, founders think the team is the most important:

While Marc Andreessen thinks the market is the most important.

The lesson here? The people shouting loudest that needing to make changes to your product are the ones with the least experience building startups.

A good rule of thumb: If anyone ever "kills” your startup’s product all that means is that you need to consider adjusting your product strategy. It potentially has zero impact on your longer term vision, the size of the market, or your passion to solve the problem.

Additionally, sometimes large product launches simply bring more attention to niche products that are better equipped to solve user problems.

Evaluate your startup honestly. Are there defensible advantages either around data, usability, network effects, or performance that a more generalized solution can’t replicate? If so, stay the course.

GPTs are Good… Actually

What’s going to happen instead of mass startup death is a Cambrian explosion of new experiments, ideas, and companies.

By reducing the barrier to entry for creating LLM-powered apps, OpenAI is sending up a bat signal to people who build products with no-code tools or even find those too intimidating.

Now they can “code” in their natively spoken language. Here’s a quick tutorial on how to create one yourself.

And to top it off, OpenAI is providing (limited) built-in distribution via discovery in their store.

Don’t believe me that this will be a big deal? We’ve seen this movie before Apple’s App Store is a perfect proxy in this case:

How to Engage with the GPT Store as a Founder

There will be some key differences between the GPT Store and the early days of Apple’s App Store, but there’s also learnings founders can take from what happened on the App Store is 2008 - 2009, when there were what now seems like a tiny amount of apps available.

Interestingly, there’s already an unofficial third-party directory of live GPTs.

First, what types of Apps were immediately successful?

Three categories that did well right off the bat were games (Angry Birds), productivity tools (Evernote), and social networks (Facebook, Instagram).

I’d argue part of the success of social networking apps was due to the macro rise of social networks that was still emerging (Facebook was only 4 years old when the App Store launched).

Since ChatGPT doesn’t have real-time information the opportunities for social apps and experiences seem initially limited. Maybe that will change in the future (or maybe Grok will launch custom groks before ChatGPT launches real-time data availability).

Instead let’s briefly look at games and productivity tools:

• Games → The UI may not make it look like it, but creating both single and multi-player games with GPTs will be possible and some will go viral much like Fruit Ninja and Angry Birds did 15 years ago. I’d expect mostly text-based games to start, but eventually multi-modal games as well.

• Productivity → LLMs present so many opportunities to build tools that improve knowledge and productivity that it seems self-evident at this point. Many of the startups that have built “GPT wrappers” in the last year fall into this category. Everything from PDF.ai to jenni.ai and many more. Since ChatGPT operates in natural language effortlessly “productivity” isn’t limited to note taking, it’s much more dynamic.

Next, what attributes did the successful early apps have?

All GPTs will have a tailwind of scarcity, there just won’t be that many GPTs early on (though it may feel like it). And each will have access to discovery via the GPT Store (and maybe within ChatGPT itself, like plugins?).

But the factors that will differentiate winners and losers will be similar to what we saw with the App Store. The winners will:

• Leaning into novelty / new functionality → What experience can you deliver to users with LLMs + GPTs that you simply could not before?

• Making it easy to understand → Since everything else about the experience will be new, the GPTs that click for people the fastest will win.

• Solve one specific problem → Ties into the last one, don’t boil the ocean with your GPTs.

• Generate eyeballs via existing distribution → Don’t rely on the GPT Store to push users to your GPT like everyone else. Send people there via other channels.

Should you go all-in on your GPT?

The big question facing founders now is whether to focus entirely on building the best GPT(s), or use them more as a marketing channel.

This is where I think larger differences with the App Store start to emerge.

Both the App Store and the GPT Store are/will be aggregators.

In general, founders should avoid being aggregated whenever possible. Stratechery’s Ben Thompson outlines aggregation theory and why it’s better to avoid being aggregated if you can.

An aggregator (like Google Hotels or Booking.com) inserts themselves as a middleman between the consumer and the vendor. The consumer benefits from an easier search experience (all options are in one place) and the vendor is pitched on getting more traffic and revenue by joining the aggregator:

When I worked at Airbnb we obsessed over this. Google, along with travel sites like Booking.com, were launching products around short-term rentals and wanted to feature our inventory. They offered us great terms, prime placement, and were amenable to requests.

However our mandate at the time (and I believe this is still true) was to resist.

We turned down the traffic because we wanted users to come to airbnb.com, we believed our brand identity and awareness were strong enough to not need aggregators, and that joining them would diminish that. There’s no substitute for having a direct relationship with your users.

Getting back to the App Store, Apple’s strong position as an aggregator has allowed them to take up to 30% of every single app’s revenue. It’s a de facto tax that generates over $70 billion in annual revenue for them.

I assume the same will become true for the GPT Store, if it takes off, but the difference is that it was worth it to go all in on the App Store due to it being the only way iPhone users could download and use apps.

With the GPT Store, unless the revenue split is very favorable for founders you may still want to treat GPTs like marketing channels (or stunts) and drive users back to your main product so that you can easily own the relationship with them and avoid platform risk.