Interaction Design Before AI Ate the Interface

Looking back through years of saved links, one pattern is hard to miss: before AI became the center of every product conversation, the useful work was already happening in interaction design. The saved trail runs through HCI, accessibility, product craft, Figma, design systems, usability, data visualization, and the recurring question of how people understand what a system is doing.

That matters because the arrival of generative interfaces did not remove the older problems. It made them more expensive to ignore. A model can produce text, images, plans, code, and recommendations, but the product still has to answer familiar questions: Where am I? What just happened? What can I change? What is uncertain? How do I recover? Who is affected by this output?

The strongest archive cluster is not just about UX links. It is about a habit of noticing the interface layer before the market has a fashionable name for it.

The links around accessibility stand out because they show a better standard for product work. Accessibility is not compliance performed at the end. It is a way of asking whether the system can be understood, controlled, corrected, and trusted by more than one kind of body, device, context, or attention span.

This becomes even more important with AI. A chatbot that writes fluent answers can still be inaccessible if uncertainty is hidden, if generated structure is impossible to scan, if keyboard paths are broken, if motion or dense output overwhelms people, or if the product has no graceful way to say that it does not know.

AI teams often talk about intelligence as if the difficult part ends at generation. Interaction design says the difficult part begins when a person has to use, judge, edit, share, or reject what was generated.

Older HCI language can sound narrow if it is imagined as a person clicking a screen. The archive suggests the opposite. Human-computer interaction has expanded into human-system negotiation. People are not only operating tools. They are negotiating with ranking systems, recommendation loops, automation, moderation, personalization, and now generative agents.

That negotiation needs visible structure. When a system adapts, the user needs to know what changed and why. When a model makes a suggestion, the user needs to understand whether it is a draft, a decision, a probability, or a hallucination with nice formatting. When an interface asks for trust, it needs to earn that trust with controls, context, and correction paths.

The lesson from the pre-AI interaction design archive is simple: the more powerful the system becomes, the more carefully the product must expose its boundaries.

It is easy to discuss AI ethics at the level of principles: transparency, agency, fairness, privacy, accountability. Those words are necessary, but they do not become real until someone makes interface decisions. A label, a warning, a review step, an undo action, a comparison view, or a source trail is where a principle becomes usable.

This is where product craft matters. Typography changes whether people can scan a generated answer. A diff view changes whether they can review an edit. A disabled button can explain a boundary or silently punish confusion. A confidence note can help judgment or become a meaningless decoration. A reset control can give someone the right to change direction.

Good interaction design does not make complex systems simple by pretending complexity is gone. It makes complexity negotiable.

The current AI wave sometimes treats the prompt box as the final interface. That is too small. A prompt box is only one entrance into a system. The real interface includes memory, context, citations, history, permissions, defaults, output formats, feedback loops, model limits, handoff to humans, and the social consequences of whatever the system helps create.

This is why the old archive is useful for HAAM. It shows that AI work should not be separated from accessibility, HCI, product design, content strategy, and service design. The best AI products will not be the ones with the most magical demo. They will be the ones where people can understand the system well enough to do responsible work with it.

Interaction design before AI was already asking the right question: how should a person participate in a system they cannot fully see? AI has made that question central.

A useful AI interface should make state visible, not only output visible. It should show what context is being used, what has changed, what can be undone, where sources came from, what confidence is appropriate, and when human review is required.

It should support different modes of attention. Some users need a summary. Some need a comparison. Some need the full trace. Some need keyboard-first control, reduced motion, larger type, plain language, or a slower interaction that makes room for judgment.

Most importantly, it should preserve agency. The user should be able to correct assumptions, refuse automation, inspect recommendations, reset personalization, and leave without losing the thread. That is not old-fashioned UX. That is the foundation of humane AI.