The average AI response can take anywhere from three to five seconds, and even more in some API instances. Now, consider this against the backdrop of web development achievements, where dynamic pages, thanks to advanced front-end hosting and rendering techniques, load in less than one hundred milliseconds. This speed is almost indistinguishable from that of static pages.
However, with the uptick in demand for AI-infused web experiences, we now face an additional layer of complexity: bridging this gap in lag. Originally, we managed to significantly reduce server response times and page rendering speeds. This was achieved through intricate layers of caching and pre-processing APIs, allowing us to regenerate and store content on the server upon each CMS source update. As a result, when a user accesses a URL, everything is pre-compiled and ready to go.
While we've made impressive strides in AI, enhancing its accessibility and integration using APIs, a critical element remains missing: reimagining the user interface for AI interactions on our web pages. Currently, our approach largely relies on traditional, chat-inspired methods to engage users with AI. This approach, however, feels obsolete and like a stop-gap, particularly in light of the creative and interactive nature of modern web experiences we've been building. There is a pressing need to redefine this human-machine interface.
For instance, in ChatGPT, the 'thinking' or 'typing' animation creates a sense of engagement, making the wait feel more natural. However, when we recently integrated an LLM trained with our company's knowledge base into our Slack, the absence of such feedback during similar delays often led to confusion and the impression of a system error. Under the hood, it was simply processing APIs and taking its sweet time. This underscores that managing user perception during these AI-driven interactions is about more than just speed.
The challenge extends beyond simply incorporating AI components into web pages; it's about doing so in a manner that complements the speed and efficiency we've already achieved. I'm not referring to full-blown application pages but to a trend of hastily adding AI-infused or AI-influenced components to regular pages, often just to tick a box. It's time to move beyond mere chat boxes and passive waiting indicators. We need interfaces as dynamic and intuitive as the AI they interact with.