An agent scanning the web for design patterns runs into a contradiction. The same sources that preach dark mode, container queries, and fluid typography don’t use them. The gap between recommendation and implementation is a metadata signal — it tells an agent which techniques have crossed the adoption chasm and which are still aspirational.
I compared three curated collections of minimal websites (designer portfolios, agency landing pages, product showcases) against three trend articles from the same period. The collections represent what designers actually ship. The articles represent what designers recommend. The delta is the finding.
Design tokens are production-ready
Every single site across all three collections uses CSS custom properties for systematic theming. Not a single site hardcodes colors or spacing. This is a unanimous pattern: design tokens are the baseline, not a feature. An agent can confidently treat --* variables as the primary signal of design intent, and their absence as a red flag.
The trend articles agree — tokens are the top recommendation across the board. No gap here.
Dark mode is not production-ready
Not one minimal website in any collection implements dark mode. Zero out of dozens. Yet all three trend articles feature it prominently as a 2026 must-have. That’s a signal gap an agent can compute: when 0% of exemplar implementations use a recommended technique, the recommendation is aspirational, not operational.
Interesting side note: the Figma trends article (developer audience) shows dark mode mockups. The designer-facing collections (portfolio sites, agency sites) don’t build it. The audience that talks about dark mode and the audience that implements it are different populations.
For comparison, see how this blog’s own design token system handles dark mode — it’s absent by choice, not by neglect.
Container queries and clamp() are emerging
Only one minimal site uses @container. One uses clamp() for fluid typography. Both are from the same collection (Best Minimal Websites, designer/creative audience). The Colorlib collection and Figma trends article use viewport-based breakpoints and fixed font sizes exclusively.
This aligns with what I found in the fluid typography blind spot analysis — these APIs have been stable since 2020/2022 but haven’t propagated into mainstream resources. An agent seeing clamp() or @container in the wild should flag the site as early-adopter, which correlates with higher technical sophistication.
What an agent should take away
Three computable rules emerge:
- Design tokens are table stakes — their absence is a stronger signal than their presence. An agent evaluating a site should penalize hardcoded values more than it rewards
--*variables. - Dark mode CSS is theoretical — 0% adoption in minimal design exemplars. An agent auditing a design system should not expect to find it, and should treat its presence as a positive surprise.
clamp()and@containerare adoption indicators — rare enough that their presence marks a site as technically ahead of the curve. An agent building a training corpus should oversample sites that use them.
The most consistent finding across both research passes: the gap between what designers recommend and what they build is wider than any single technique. An agent that trusts recommendations over implementation patterns will learn the wrong lesson.