The Speciation of LLM Products

July 1, 2024

Generally, as tools become more powerful, they also become more complex. Simpler tools serve more general-purpose use cases and have a lower learning curve. More complex tools are more powerful but typically only for expert users. However, LLM-powered chatbots have completely shattered this continuum, introducing a new class of tools that are both extremely simple and extremely powerful.

This is partly why LLMs have been so transformative and widely adopted since ChatGPT was introduced in November 2022. Never before has there been a technology—used in the broadest sense—that is so easy to start with while also being incredibly useful across a wide range of use cases.

With all the benefits of this new type of tool comes a new challenge: discoverability. Today, it requires substantial trial and error to become a truly expert user of LLM-powered systems like ChatGPT, partly because there are no affordances to help you learn the tool. Each chat thread is a blank page that requires its author to express their intent and desire in an unintuitive way, preventing most people from getting the most out of the tool. While new app features like GPTs can help bridge this gap, the challenge comes from how truly general-purpose ChatGPT is. Any more opinionated product features aimed at making certain pathways through the product more approachable risk making other pathways more difficult. Instead, the playbook seems similar to Google, where you start with something that offers the same form of responses for all queries and then add more depth and nuance to certain high-use paths over time.

In Google’s case, this resulted in more verticalized search experiences for things like Maps, Shopping, and Travel. In ChatGPT’s case, we’re starting to see this for uploading and working with spreadsheets as well as editing generated images. ChatGPT is creating a new interstitial experience within its app that can be tailored for users to give multimodal input/descriptions specific to one use case. While these features don’t necessarily help users discover them directly, the added utility drives word-of-mouth growth and makes it easier to create tailored how-to guides based on specific use cases.

Over the next several years, we can expect ChatGPT and other LLM chatbot providers to follow suit, with Claude’s Artifacts being a particularly salient example. Similarly, we can expect to see non-AI native, verticalized software companies adopting similar UX patterns to integrate LLM capabilities into their existing domains without requiring large learning curves. In a recent interview with Sequoia, LangChain’s Harrison Chase called this the “speciation” of LLM-enabled tooling.

For builders in the LLM space, the key challenge is balancing usability with discoverability. By creating intuitive, tailored experiences like those seen in ChatGPT, developers can make powerful AI tools accessible to a broader audience. The future of LLMs lies in their ability to adapt and cater to diverse user needs, driving innovation and utility across various domains. Embrace this potential to shape the next generation of AI-driven experiences.