This week’s developments span the full width of our discipline — from AI terminology finally getting plain-language definitions, to CSS features that could eliminate JavaScript from view transitions, to a sharp reminder that a perfectly built feature can still fail if it clashes with user identity. Here’s what our team is watching and already applying to client projects.

Key Takeaways

• A comprehensive AI glossary from Nielsen Norman Group gives designers and developers a shared vocabulary for product work involving AI.
• Generative AI’s real-world adoption looks nothing like science fiction predicted — it’s quiet, trust-based, and already embedded in workflows.
• SMIL animations offer a JavaScript-free method for animating SVGs, including inside <img> tags — a technique most teams overlook entirely.
• CSS Navigation Matching is an early-stage spec that could make cross-document view transitions fully declarative.
• Features that perform well technically can still repel users when they conflict with aspirational identity — a retention killer hiding in plain sight.

Get Your AI Vocabulary Right Before Building AI-Driven Interfaces

Tokens, context windows, prompt injection, evals, agents — these terms now appear in client briefs, stakeholder meetings, and product specs. The problem is that half the room uses them loosely. Nielsen Norman Group has published a plain-language AI glossary built specifically for product and design teams. We’re using it as a reference standard across our agency. When a client asks for an “AI agent” on their site, everyone involved needs to mean the same thing. Misaligned terminology leads to misaligned builds.

AI Arrived Without a War — And That Changes How We Design For It

Science fiction promised a dramatic confrontation with machine intelligence. What actually happened is far more mundane and, frankly, more interesting from a UX perspective. As explored in this analysis of what The Matrix got wrong about AI, generative AI doesn’t demand obedience — it asks for trust, and mostly receives it. For our design team, this has direct implications. Users don’t fear AI tools; they delegate to them. That means the interfaces we build around AI features need to prioritise transparency and control, not reassurance. Trust is already there. The job is not to lose it.

SMIL Animations: The SVG Technique Your Team Is Probably Ignoring

Most front-end developers reach for JavaScript or CSS animations when working with SVGs. There’s a third option that works inside <img> tags and requires zero scripting. Smashing Magazine has published a detailed blueprint for SMIL animations using timing charts. SMIL lets you animate SVG attributes directly in the markup — strokes, positions, colours, opacity — with precise timing control. We’ve started evaluating SMIL for client projects where animated icons or illustrations sit inside content-managed <img> elements, where JavaScript simply can’t reach. Lighter pages, fewer dependencies, better performance.

CSS Navigation Matching Could Kill JavaScript in Cross-Document Transitions

Cross-document view transitions currently require JavaScript to identify source and destination pages. A new CSS specification in its early stages aims to change that entirely. As reported by CSS-Tricks in their coverage of CSS Navigation Matching, the proposal would let stylesheets apply rules based on which page a user navigated from. Declarative, no scripts needed. It’s early days — browser support is non-existent in production — but we’re tracking this closely. Declarative transitions mean fewer points of failure and simpler maintenance for multi-page sites.

When a Feature Works Perfectly But Users Still Reject It

This one matters. A feature can pass every usability test and still tank retention. The reason? Identity mismatch. A UX case study examining why users kept hiding a well-built feature reveals that the brand imagery associated with the feature clashed with how users wanted to see themselves. No requirements document captures this. Our takeaway: user research must go beyond task completion. We now ask identity-focused questions during discovery — “Does using this feature feel like you?” — because aspirational self-image drives behaviour in ways that functional testing never surfaces.

The common thread across all five developments is precision. Precise language for AI. Precise trust calibration. Precise animation without bloat. Precise CSS without scripts. And precise understanding of who your users believe themselves to be. These aren’t abstract concerns — they shape the builds we deliver to clients in 2025 and beyond.

Frequently Asked Questions

What is SMIL animation and why should web designers care about it?

SMIL (Synchronized Multimedia Integration Language) lets you animate SVG elements directly in markup without JavaScript or CSS keyframes. It’s particularly valuable because it works inside <img> tags, where no scripting language can operate.

How do web designers handle AI terminology in client projects?

We use standardised glossaries — such as the one published by Nielsen Norman Group — to ensure every stakeholder shares the same definitions. Aligning on terms like “agent”, “token”, and “context window” early in a project prevents costly miscommunication during build.

Why does a well-designed feature still fail with users?

A feature can be fully functional yet conflict with a user’s aspirational identity, causing them to avoid or hide it. This identity mismatch doesn’t appear in standard usability testing and requires deeper qualitative research during the discovery phase.

What is CSS Navigation Matching and how will it affect website design?

CSS Navigation Matching is an early-stage specification that would allow stylesheets to apply rules based on which page a user came from, making cross-document view transitions declarative. If adopted by browsers, it would reduce JavaScript dependencies and simplify maintenance for multi-page websites.

How do web designers build trust into AI-powered interfaces?

Users already tend to trust AI tools, so the design priority is maintaining that trust through transparency — showing what the AI did and giving users clear controls to override it. The focus should be on not eroding existing trust rather than trying to build it from scratch.