While direct vector space communication between AI agents would be most efficient, the reality of heterogeneous systems and human-in-the-loop collaboration makes natural language the necessary lowest common denominator for interoperability for the foreseeable future.

Voice-to-voice AI models promise more natural, low-latency conversations by processing audio directly. However, they are currently impractical for many high-stakes enterprise applications due to a hallucination rate that can be eight times higher than text-based systems.

The next leap for AI interfaces is voice-controlled agents performing complex tasks like sending emails without visual confirmation. The critical barrier to adoption isn't the technology's capability but whether users trust the AI to act correctly on their behalf without a screen.

Current chat interfaces are compared to the command-line: they require users to learn a specific, procedural way of communicating ('prompt engineering'). New interaction models, which allow for natural, multimodal communication, could be AI's 'GUI moment,' democratizing access by letting users focus on the task, not the tool.

As models become more powerful, the primary challenge shifts from improving capabilities to creating better ways for humans to specify what they want. Natural language is too ambiguous and code too rigid, creating a need for a new abstraction layer for intent.

User expectations for AI responses change dramatically based on the input method. A spoken query demands a concise, direct answer, whereas a typed query implies the user has more patience and is receptive to a detailed, link-filled response. Contextual awareness of input modality is critical for good UX.

Users often struggle with how to prompt an AI. Voice interaction provides a natural, high-bandwidth method for unstructured 'yapping' or context dumping. This lowers the friction of starting a task and allows users to delegate in a more natural, conversational way, leading to more sprawling and complex use cases.

A major hurdle in AI adoption is not the technology's capability but the user's inability to prompt effectively. When presented with a natural language interface, many users don't know how to ask for what they want, leading to poor results and abandonment, highlighting the need for prompt guidance.

The magic of ChatGPT's voice mode in a car is that it feels like another person in the conversation. Conversely, Meta's AI glasses failed when translating a menu because they acted like a screen reader, ignoring the human context of how people actually read menus. Context is everything for voice.