people will look back at 2025 and laugh.

it’s funny we’re using a machine learning research portal and calling it product design.

the biggest opportunity for applications in ai right now is better design to match the level of intelligence available by new foundation models to specific workflows.

400M+ people text or talk with chat gpt every. chat is easy to use and adaptable to different contexts. this is leading to a ubiquity that is misleading. this explosion of growth has infected the experience design of every single product integrating large language models.

• the internet? you can chat with that with perplexity. the chat pressure has even pushed google for their new ai mode to do chat.
• your code base? cursor or windsurf allows you to chat with that.
• pdfs, files, or documents? notebook lm is one big sandbox for personalized chat.
• youtube videos? sure, tag it @youtube in gemini and chat with it.
• edit an image? chat with open ai’s new image gen capabilities.

there is almost nothing you can’t chat with in 2025.

and i have nothing against chat as a bridge to what’s next, but it’s important to realize chat it is just that: a bridge to something that needs to get better asap because chat falls short in major ways.

chat windows fall short to the human experience

three big gaps in great product experience in the chat with everything paradigm:

1. chat is cognitively draining as a never ending guessing game.
2. chat is a linear experience with consistently missing context.
3. chat sets the wrong expectation for real time use, making users impatient.

1. chat is a cognitively draining as a neverending guessing game.

o3 making a cartoon about itself.

the need to actively write or speak a specific prompt to get a particular output feels like a continous guessing game with the multilayer perceptrons.

the current paradigm of talking to a large language model (ie. prompt engineering) is it’s own kind of language. rather than the product adapt to us, we’re adapting to it.