Ten patterns of interaction evolution, and what they mean for the people building interfaces.
Rewiring the Interface
How AI Is Reshaping Human–Computer Interaction and Frontend Architecture
A synthesis paper, completed July 2026. What follows is the argument in short form.
The claim under test
Generative AI is routinely described as a new interaction paradigm. The paper tests that claim against the actual arc of HCI history — Licklider and Engelbart, batch processing and the command line, Shneiderman's direct manipulation, the GUI, touch, voice, and now conversational and agentic systems.
The conclusion is narrower than the marketing. AI interaction is a new layer atop prior paradigms, not a replacement for them. Nothing that came before was deleted; the command line survived the GUI, and the GUI will survive the chat box.
Its effects also diverge sharply by generation and by context. Consumer usage skews toward advice-seeking and writing. Developer and enterprise usage skews toward coding and automation. Any claim about "how people use AI" that doesn't say which people is measuring an average nobody occupies.
The new engineering problem
The defining new problem is the determinism–probability tension.
Classical UI engineering assumes a given input produces a predictable, testable output. That assumption is what makes a test suite meaningful, what makes a bug reproducible, and what makes a component library safe to reuse. Generative systems are probabilistic by construction. They break the assumption at the root.
The paper's position is that this must be architected for rather than around. Three moves follow from it:
The ten patterns
The central contribution is a framework of ten reusable interaction patterns, offered as a durable mental model for frontend architects working across this transition:
| abstraction | adaptation |
| intent | modality |
| competence | bandwidth |
| specialization | determinism |
| trust | personalization |
They are deliberately framed as patterns rather than predictions. A prediction expires; a pattern is still useful when the specifics change underneath it.
Structure of the paper
Historical foundations, then the current state — usage data, generational divergence, and the economy forming around AI interaction. Then the future of frontend architecture: generative UI, agent-facing protocols, design-to-code, spatial computing. Then the ten-pattern framework. Then staged recommendations for practice, split into no-regret moves, 6–18 month positioning, and 18+ month bets.
Every recommendation names the evidence that would reverse it. A recommendation you can't falsify isn't advice, it's a mood.
Why it sits on this site
It's the one artifact that connects the two halves of the work — six years of frontend architecture and the recent AI engineering — as a single argument rather than two adjacent skill lists.
It also has an implementation attached. Section 4.3 describes structured generative UI: a schema-constrained specification rendered from a pre-tested component catalog. That is precisely what MFKit's plugin-API-plus-template model does for micro frontends. The paper is the thesis; MFKit is one of the places it gets built.