Design Principles That Outlast the Tools

In the last fifteen years, a working product designer might have moved from Photoshop to Sketch to Figma, picked up a prototyping tool or three, and now spends part of each week prompting a model to generate first-draft screens. Each switch felt significant at the time. Very little of what made their work good changed with any of them.

What carried over were principles: compact statements about what makes a designed thing work for the people who use it. The oldest ones in common use are more than forty years old. They have outlived every tool listed above, and they are likely to outlive whatever comes next.

Four bodies of principle that have held up

Dieter Rams: ten principles (late 1970s)

Rams, head of design at Braun, wrote his ten principles in the late 1970s as an answer to a question he kept asking himself: is my design good design? Vitsœ, the furniture company he designed for, keeps the list online. A few of them:

Good design makes a product useful.
Good design makes a product understandable.
Good design is honest.
Good design is long-lasting.
Good design is as little design as possible.

None of these mention a material, a screen size or a tool. That is why they still work for a mobile settings page, an industrial control panel or a robot's teach pendant. "Honest" in particular has become more pointed, not less. A dark pattern that tricks users into a subscription fails it as clearly as a fake chrome trim on a radio did in 1978.

Don Norman: The Design of Everyday Things (1988, revised 2013)

Norman's book gave designers a vocabulary for why people struggle with objects: affordances, mappings, feedback, conceptual models, and the gulfs of execution and evaluation. The 2013 revised edition kept the science essentially unchanged, with one important addition: signifiers, the perceivable cues that tell you where and how to act.

Norman's examples were doors, stoves and telephones. The ideas transferred directly to software, then to touchscreens, and now to voice and agent interfaces. When a chat-based assistant silently takes an action and gives no sign that it did, that is a feedback failure Norman described decades ago. When an icon-only toolbar leaves users guessing, that is a missing signifier.

Jakob Nielsen: 10 usability heuristics (1994)

Nielsen's ten heuristics were published in 1994, and NN/g still uses them largely as written. The list includes visibility of system status, match between the system and the real world, user control and freedom, error prevention, and recognition rather than recall.

They endure because they describe people, not interfaces. Short-term memory did not get bigger when we moved from desktop to mobile. People still lose track of where they are, still make slips, and still need a way to undo. A heuristic review takes an afternoon with no special tooling, and it catches problems in a 2026 interface as reliably as in a 1994 one.

WCAG 2.2 (2023)

The Web Content Accessibility Guidelines are the most concrete of the four, and the most testable. WCAG 2.2 became a W3C Recommendation on 5 October 2023. According to W3C's summary of what is new, it added nine success criteria, including:

2.4.11 Focus Not Obscured (Minimum), so a sticky header cannot hide the element that has keyboard focus.
2.5.7 Dragging Movements, requiring a single-pointer alternative to drag-only interactions.
2.5.8 Target Size (Minimum), with targets of at least 24 by 24 CSS pixels unless an exception applies.
3.3.8 Accessible Authentication (Minimum), which limits login steps that depend on memory or puzzles.

Older criteria still carry most of the weight. The 4.5:1 contrast ratio for normal text in 1.4.3 has been in place since WCAG 2.0 in 2008, and it applies equally to a hand-built page and to a layout a model produced in four seconds.

Why principles outlast tools

Tools encode a particular moment: a screen resolution, a rendering engine, a workflow, a business model. Principles encode things that change slowly:

Human perception and cognition. Attention, memory, motor control and reading behavior have not changed in the time it took the industry to go through four design tools.
Human values. People still want products that are honest, that respect their time, and that do not exclude them.
Physical constraints. Fingers have a size. Screens have glare. Factory floors are loud and operators wear gloves.

A tool can make it easier to follow a principle. Design tokens make it easier to apply consistent contrast. Auto layout makes it easier to keep spacing systematic. But a tool cannot decide which principle matters most in a specific situation, and that is where most design work actually happens.

What AI changes, and what it does not

Generative tools can now produce plausible screens from a prompt. Figma introduced Figma Make in May 2025, a prompt-to-app tool that turns existing frames into working prototypes. Similar features exist across most design and front-end tools. For first drafts, variations and boilerplate, they save real time.

They also tend toward the average. A model trained on millions of interfaces produces something close to the median interface. That is often fine for a login page. It is not enough for a dialysis machine's alarm screen, a security console used during an active incident, or a robot cell's emergency stop flow.

Three kinds of work remain human, and principles are the language for doing them.

Taste: knowing what to remove

"As little design as possible" is a judgment about what this product, for these users, can do without. A model will happily add a sixth card to a dashboard. Deciding that the dashboard needs two numbers and a trend line requires knowing what the operator actually looks at during a shift.

Judgment: weighing principles against each other

Principles conflict. Error prevention argues for a confirmation dialog. Efficiency of use argues against it. On a consumer app, you might skip the dialog and offer undo. On a press brake controller, you confirm, every time, because undo does not exist for a bent sheet of steel or a hand. A generator does not know which world it is in unless a person tells it, and that person needs the vocabulary to explain why.

Empathy: noticing who is missing

Accessibility failures are rarely visible to the people who create them. Someone has to ask how a screen reader user gets through the flow, whether a colorblind technician can read the status lights, and whether a 24-pixel target is enough for a gloved hand on a vibrating machine. WCAG gives the minimum. Going beyond it starts with imagining someone unlike you, then testing with them.

Practical use

If you want principles to shape real work instead of sitting in a slide deck:

Review AI-generated drafts against Nielsen's heuristics before refining them. It is fast and catches the common failures.
Make WCAG 2.2 AA part of your definition of done, and test it with a keyboard and a screen reader, not just an automated checker.
Write down the principle behind each significant decision in your design notes. "We kept the confirmation step because error prevention outweighs speed here" is more useful to the next designer than a screenshot.
Re-read the originals. Rams' list fits on a card. Norman's book rewards a second reading every few years. Each time, you will notice something you have been getting wrong.

The tools will keep changing. The person deciding what good looks like, and for whom, is still the part that matters.

Sources

Dieter Rams: ten principles for good design, Vitsœ — the ten principles and their late-1970s origin.
The Design of Everyday Things, Revised and Expanded Edition, Don Norman — 2013 edition; science unchanged except the addition of signifiers.
10 Usability Heuristics for User Interface Design, NN/g — Nielsen's heuristics, first published 1994.
Web Content Accessibility Guidelines (WCAG) 2.2, W3C — the standard, including the 4.5:1 contrast minimum in 1.4.3.
What's New in WCAG 2.2, W3C WAI — Recommendation date (5 October 2023) and the nine new success criteria.
Introducing Figma Make, Figma — May 2025 launch of Figma's prompt-to-app prototyping tool.