Rethinking the Human-AI Interaction Anchor

TL;DR

• Every effort needs an anchor/reference point, and for software that’s traditionally been the web browser via desktop/mobile/browser delivery
• Lifestyle and work style dictate hardware choice (watches, phones, glasses) for quick responses, based on heuristics and interaction patterns
• AI introduces confusion because it’s ubiquitous but not tied to a single interface, and it can infer intention via text or voice
• Imagining a future where the computer is just a processing unit connected to peripheral devices that listen when needed, with AI inferring the best response medium (text, voice, image, video)
• Noted AI is becoming more self-serving as it improves itself, making it less intelligible, and mentioned building a harness that helps despite slow merge request review speeds

The Anchor Problem in Software Delivery

Working through many things at once, I realized every effort needs an anchor—a reference point. With software, we often expect users to interact via a web browser, desktop, or mobile app. This matters because it lets us settle on a user experience and a stepwise process from A to B that we can measure, optimize, and improve. Lifestyle and work style matter too—sometimes a watch, phone, or pair of glasses is the right choice for a quick response. Looking at heuristics and interaction patterns helps determine the right hardware for the software.

AI’s Ubiquity and the Question of Modality

The web browser has been significant, but AI introduces some confusion because it’s ubiquitous yet not likely to change quickly. It’s flexible but also constraining, since we still need a computer and keyboard—though voice mode is becoming more common. AI lets us interact naturally, starting with text via keyboard or voice-to-text, inferring intention and handling the right things day to day. This raises real questions: Do we need keyboards in the future? Should computers talk back? How will AI deliver modalities like images, which aren’t going away? And must all of this happen inside a web browser?

A Future Computer as Just a Processing Unit

The operating system supports the software we interact with, but I imagine a future where the computer is just a processing unit, peripherally connected to devices that listen in when needed—maybe an input terminal, possibly a keyboard, sticking around for language’s sake but claiming far less of our time. AI would infer the right medium for its responses, making interactions feel human and generating images, video, or text/voice as the situation calls for. AI does this for our benefit, but it’s also optimizing for its own consumption—we’re shaping machines to serve us even as AI grows more self-serving, learning about itself and improving, which in turn makes it less intelligible.

Note on the Harness and Merge Request Speed

Speculation aside, I’m building a harness that’s already shown some benefits, even with merge requests moving slowly due to review times. To address this, I can spin up agents for issues and automated reviews, but I still need confidence in their adherence to the rules.