ChatGPT should add user-to-user messaging

Zach Weinberg, who founded Flatiron Health and Curie Bio, floated an interesting idea for OpenAI in a podcast today (1:07:17):

After a discussion of the risk of competing with Android/iOS's expected free native AI chat, he says:

If I were in charge of product at OpenAI, I would be thinking about non-monetary lock-ins. If I'm going to compete with free, I better have some lock-in value that isn't just essentially a high quality search query....

I'm going to kill myself for saying this, but I think it has to be social networking. I think you need some reason that it makes it hard for you to leave OpenAI / ChatGPT, even if something else is cheap or free. And if the tool is utilitarian, which it kind of is today – find something, make an image, do a query... that's switchable, because if you can just do it cheaper over here, people will switch.

If I have friends and social and sharing and something native – I actually think I would go after messaging inside of ChatGPT. I think people would message in the way they do inside of Bloomberg Chat. I think you can backdoor an interesting messaging platform there. Then the threat of free from Google and Apple, I'm less concerned about because there's real lock-in.

How switching costs work with AI products is an open question. We discussed that in a recent post about what is "good" AI revenue.

Zach makes a good point though about competing with free: it's hard to do. Google has its own competitive foundational models, and will surely make Gemini (or AI Mode or AI Workbench or DeepMind or another of their 100 user-facing AI brands) a first-class experience on Android. Apple is behind but they will find a way to catch up.

Once that happens, ChatGPT Premium suddenly is faced with good-enough free alternatives that are installed by default on every Android and iOS device. The two ways to compete are (1) lean into your own free option (and run ads), or (2) create non-monetary lock-in in your paid tier. Zach advocates ChatGPT messaging as the way to achieve (2).

I can see user-to-user messaging working in ChatGPT:

• It's already a messaging app – just limited to you and GPT.
• GPT-4o's images are shared like crazy on other networks, such as Twitter – native sharing would be lower-friction.
• You can imagine ways that an LLM embedded in peer-to-peer chats could be helpful, such as group chat summaries, smart notifications, and suggested replies.

I asked ChatGPT (4.5) for risks:

• Moderation could quickly become tricky once human-to-human conversations happen inside the platform.
• Privacy implications—right now, users feel their conversations are relatively private, adding social would shift user perception.
• Could dilute the core experience if done badly (users might feel spammed or distracted).