The AI tools actually worth adding to a design workflow

Every few months there's a new list of "the best AI tools for designers," and it's usually twenty-plus items padded with anything that mentions design in the marketing copy. I've tried most of them. This is a shorter list of what I actually kept.

Claude for almost everything text-shaped

Claude is where I spend the most AI time. Writing, editing, thinking through decisions, drafting documentation, reviewing copy, structuring arguments. It's good at holding a lot of context and useful for the specific kind of work designers do: articulating decisions, writing rationale, explaining tradeoffs, writing feedback on other people's work.

The versions matter. Claude Sonnet is fast enough for most tasks and good enough for nearly everything. Opus is slower and noticeably better at nuanced reasoning. I use it when I need to think through something complicated. For quick tasks I want to stay in flow for, the lighter models are usually fine.

The main thing that took me time to learn: quality of output scales with quality of input. A vague prompt produces a fluent response that doesn't help. Clear framing, real context, and a specific request produce something worth editing. This is a skill, and it develops with use.

Claude Code for implementation work

If you have any involvement in how your designs get built, Claude Code is worth learning. I use it to prototype interactions, build components, explore implementation options before writing specs, and maintain this site. The workflow is: describe what you want, read what it produces, evaluate whether it made the right choices in the places you didn't specify, redirect.

It's not a replacement for engineering judgment. It makes choices you didn't ask it to make, some of them wrong in ways that matter. But the speed of iteration between intent and running code is fast enough that it's changed what's feasible in the time I have.

For designers who aren't comfortable in a terminal: the barrier is lower than it looks, and the most relevant things to learn aren't about code. They're about how to describe what you're building clearly enough that something can build it. That's a design skill.

Perplexity for research

For factual research and gathering background before working on something new, Perplexity has replaced most Google searches in my workflow. It pulls from current sources, cites them, and produces an answer rather than a list of links to evaluate.

Where it's not useful: anything where you need judgment, synthesis across a long body of work, or expert-level nuance on a specialized topic. The answers are accurate more often than not but sometimes confidently wrong in ways that aren't obvious. It works as a starting point, not an endpoint.

What I've stopped using

Most image generation tools for design work. Midjourney is good at what it's good at. A certain kind of photographic aesthetic, styled illustration, mood-board exploration. For UI design, for anything that needs to communicate information, or for anything where the design decisions are the point, I've never found image generation useful enough to keep in my regular workflow. The results are impressive and directable to a point, then they hit a ceiling.

AI-powered Figma features: mostly not useful. Auto-layout suggestions, the "Make design" input. I've tried each one multiple times and they produce output I then have to undo or heavily modify. The gap between "Figma component with reasonable structure" and "something worth shipping" is still the work. Nothing I've used in Figma has narrowed that gap enough to matter.

Tools that summarize things I should just read: I've noticed a tendency to reach for "summarize this for me" on things that are short enough to read in five minutes. That's not a time-saver, it's a way to consume information at lower resolution while feeling productive. The tool isn't the problem, the habit is.

A note on the churn

New AI tools are releasing faster than it's possible to evaluate them properly. The pattern I see is: most tools that are genuinely useful for design work are improvements on an existing workflow, not new workflows. They do something you were already doing but faster or with less friction. The ones that claim to replace the workflow wholesale are usually the ones with the most marketing and the least utility.

The evaluation process I've settled into is to use a tool for a real piece of work, not a demo or a trial task. Most tools look useful in a demo. Demos are optimized for looking useful. Running something through a real project with real constraints and real time pressure is the only way I've found to tell if a tool is going to stay in the workflow or get uninstalled after two weeks.

The useful question isn't whether a tool is impressive (many of them are), but whether you'd notice if it disappeared from your workflow tomorrow. The ones on my short list I would notice.