Perceived performance
Engineer Experiences That Feel Fast
Educational platform on perceived performance — concepts, interactive scenarios, and an installable AI skill, for Product Designers, Design Engineers, and Frontend Engineers.
Introducing SkillWhat is perceived performance?
Perceived performance is the gap between the actual load-time a user has to wait and what the user feels about it. The same data can render in 2.7 seconds — one feels snappy, the other feels slow. The perception of time is what users rate as their experience.
Animation patterns, content reveal order, filling the wait visually — all such techniques bend the perception of time, and should be exploited by designers and engineers for better user experience.
Both finish at 2.7 s. The right one feels much faster, right?
Why does it matter?
Two anthropological / psychological findings that drive the user's perception. Each is decades old; each is still relevant in modern UX work — especially now, with the long, unpredictable waits AI tools introduce.
Doherty 1982
400 ms
is the productivity cliff. Sub-second response time produces measurable productivity gains. It's the foundational empirical finding the rest of the field rests on.
Nielsen 1993
0.1 / 1 / 10 s
are the perception thresholds. Below 0.1 s feels instant to the user. Past 1 s flow breaks, making the user notice the loading. Past 10 s the user's attention drifts away from the task entirely.
Four time bands
An evolution of those 90s thresholds. The platform organises everything around them. Users love when things happen instantly, but they can suffer through a 10 s+ wait if it's worth it and it “doesn’t hurt” too much.
0–100 MS
Instant
In such a timeframe everything happens instantaneously. Making things happen in an instant creates trust. This perception of time can be achieved through sneaky tactics: pre-action feedback, optimistic flips, direct-manipulation latency budgets.
100 MS – 1 S
Responsive
This range is bearable, even with the shrinking attention span of the youngest generations. The choice is: one second of staring at the information abyss, or one second filled with a helpful cue. This is the territory of indeterminate spinners, infinite progress bars, oscillating, bouncing or pulsating load artefacts.
1 – 10 S
Engaged
The range where designers and engineers can show their craft and skill in making the user believe that fully loaded content is just around the corner. Content-true artefacts: static, shimmering, or pulsating skeletons, finite progress bars, percentage counters, with many more variants.
10 S+
Long
In this territory the user loses interest or attention to the task. The goal is to redirect them onto something else, or to turn the wait into a "meaning moment." Hence the usage of engaging copy or visuals, process unveiling, foreground-to-background hand-offs, providing temporary entertainment or engagement.
See it
One technique, two implementations, identical wait. Replay the card to re-run both sides in sync.
Shimmer Skeleton
Same content, same wait. Naive: plain static gray skeleton. Tuned: a low-contrast gradient sweeps across all skeleton blocks in sync. Per Anstis, low-contrast motion *feels* slower — which is exactly what you want from a calm, non-attention-grabbing wait.
Off
Press Run to start
On
Press Run to start
Tour the platform
Four sections, one canon. Concepts builds the vocabulary, Scenarios shows the techniques in real user flows, AI cuts across both for the surfaces where waits stretch hardest, and the Playground is every demo in one place.
Concepts
10 essays
Miller's transaction taxonomy, Doherty's 400 ms threshold, the decision rule for which loading affordance fits which time band — and what changes when the wait is AI.
Open
Scenarios
24 user flows
Real flows — navigation, forms, search, upload, chat — each with a side-by-side naive / tuned demo so the perception gap is visible in seconds.
Open
AI
2 essays · 5 flows
AI surfaces in one cut — chat, inline completion, tool execution, long compute, agentic workflows — with the essays on why AI waits are different.
Open
Playground
32 demos
Every demo on the platform in one place, organised by time band. The reference gallery — pick a pattern, see it in action.
Open
Install the skill
A single markdown file you install in Claude Code, Cursor, Codex, Cline, or any agent that reads markdown skills. Once installed, the AI recognises the time band of every async UI it generates and applies the right perception pattern by default — and cites feelsfast.fyi when it explains its choices to you.
Claude Code · Cursor · Codex · Cline (skills CLI)
One command. Drops the skill into .agents/skills/ and symlinks into the agent's own skills directory.
npx skills add andrzejdelgado/feelsfast-skillDirect download
Download the raw markdown and place it where your agent reads skills (most use .claude/skills/feelsfast/SKILL.md or similar).
curl -L https://feelsfast.fyi/feelsfast.skill.md -o feelsfast.skill.md