ChatGPT is the Craigslist of AI (and Ripe for Unbundling)
Entrepreneurs can use ChatGPT as a proving ground to spawn specialized AI consumer apps
Craigslist was the go-to platform for everything from finding a job to selling your old couch. Then, savvy entrepreneurs turned trending Craigslist sections into billion-dollar companies like AirBnB, TaskRabbit, StubHub, and NextDoor.
I believe we're witnessing a similar phenomenon with ChatGPT. We used this fast MVP technique before creating OneCircle.ai (the personal secretary for superconnectors to manage their contacts)
In this post, I'll explore:
The parallels between ChatGPT and Craigslist
How to spot opportunities within ChatGPT
A real-world example (our OneCircle personal CRM)
Why generalized apps like ChatGPT aren’t enough
Jack of All Trades
In its heyday, Craigslist was the simple go-to platform for... well, almost everything. Need a job? Craigslist. Selling your old guitar? Craigslist. Looking for a date? (Yep, you guessed it.) Craigslist.
Similarly, ChatGPT has become the catch-all text-based tool for an incredibly wide range of tasks:
Writing essays
Debugging code
Brainstorming ideas
Answering trivia
Planning trips
And the list goes on...
This generalist nature is both their strength and their limitation. They're incredibly versatile, but they're not optimized for any specific use case.
The Innovation Springboard
Here's where it gets really interesting. Both Craigslist and ChatGPT serve(d) as unintentional incubators for innovation:
Identifying Pain Points: Users of both platforms often think, "This is great, but if only it could..."
Revealing Market Gaps: The popularity or high value of specific use cases highlights unmet needs in the market.
Low-Stakes Testing Ground: Entrepreneurs can observe user behavior and test ideas without significant investment. Every ChatGPT thread is a working prototype.
Unbundling a Contact Manager from ChatGPT to build OneCircle.ai
I used ChatGPT as a makeshift CRM (Client Relationship Management) system to track high-value people I met. I had a dedicated thread where I’d mention who I spoke with and some notes about our convo.
Every day, I'd update this thread with updates from my convos. Over time, ChatGPT started to get confused and forget which updates were about which contacts. There were too many disparate details in the same context thread.
So, we’re building OneCircle as a dedicated iOS app with stable data structures and a UI that appropriately displays various permutations of your contacts. The main interface is a voice-first conversation with your AI secretary, but there is a deep CRM-specific design built around it.
Check out OneCircle.ai for an AI secretary to help you stay connected to your network.
Won’t ChatGPT’s Generalized AI just do everything?
While many assume a generalized ChatGPT-like super AI will do everything flawlessly, I find that generalized AI hits usability limitations with memory, action, user interface (UI), and perceived brand. This may change in 3 years, but 3 years is a long time to build a high-impact consumer app.
Memory
A large language model’s context window memory is generalized and unstructured, which often leads to confusion.
For example, let’s say you used ChatGPT to plan a travel itinerary to Europe. You first wanted to start in London, but then you switched to Paris and then later shifted to Madrid as the first leg.
As you continue to work with ChatGPT, you will find that it sometimes gets confused where you’re going first. That’s because the decisions to go to each of the three cities are within the same thread, which the LLM treats as the same input context.
Instead, a human travel agent would first write down London, then cross out London before writing Paris. In database speak, the app should replace “City1 = London” with “City1 = Paris”. However, this requires a case-specific, persistent data structure that has fields for cities, dates, travelers, etc., and an AI agent that is designed to work with this data structure.
Action
After deciding where to stay, you’d want the AI to take action. You’d expect a human travel agent to check flight prices, make dinner reservations via Resy, book hotels on Hotels.com, etc.
In theory, one can imagine a generalized AI that defines and executes all possible actions across endless 3rd party tools in every domain. In practice, each of these actions needs to be provided with specific tools and battle-tested for edge cases. Plus, the compute required for a generalized AI to define, test, and validate tools dynamically for every use case (and do so within milliseconds) is unreasonable for the perceived future.
User Interface (UI)
While I believe conversational UI is the most natural way to interact with assistants, plain text is shallow. The best human assistants would supplement verbal conversation with charts, illustrations, and data tables.
Back to our travel analogy, you might want to see a color-coded table with your itinerary, maps showing you how to get to destinations, or iconography indicating which excursions have been successfully booked vs. which are still in progress.
Perceived Brand
Super apps exist in other parts of the world (e.g., WeChat, Gojek, Paytm), but consumers in the West are conditioned to turn to specialty brands. We might wear a Hugo Boss suit but a Lululemon yoga pant. Our phones are filled with dedicated apps for every task.
This holds for our human assistant analogy: we don’t expect the skillful mechanic fixing our car also to be the well-trained attorney writing our estate plan. While both are humans, and each person could potentially be retrained to handle the other’s tasks, we inherently believe that expertise is crafted from depth.
What specific use cases have you found?
Are you using ChatGPT in a unique way that could be transformed into a standalone product? Have you spotted any trends that might be the next big thing? I'd love to hear your thoughts and ideas. We’ve built an incubator specifically to create specialized AI applications.
DM or comment below.