How We Evaluate AI Tools for Design Work
Toolify lists nearly 30,000 AI tools across 459 categories and the number grows daily. For design teams, that scale has made selection a real problem. Since every team is working from the same list, the ones doing the best work are those with a clear process for deciding what belongs in their stack.
At Graybox, we measure every tool against one question: does it make the work genuinely better?
Start With The Job, Not the Tool
“We should use AI for this" is not a strategy.
Creative work spans strategy, design, content, prototyping, motion, development, and delivery. Before any tool enters our stack, we ask what problem it solves inside one of those areas. Every platform takes time to learn, every output needs review, and every workflow change carries a cost. Tools that do not justify that cost do not belong.
Where AI Earns Its Place in a Design Stack
AI is most useful at specific gaps in the creative process:
| Situation | What AI provides |
| Two strong directions, neither quite right | Additional options to react against; sharpens the original thinking |
| Unfamiliar category, no existing instincts | Pattern recognition and landscape summary; context and point of view still comes from the team |
| The right image exists only in your head | Visual direction closer to the idea before production; makes taste more important, not less |
| An idea needs to be experienced, not just shown | A working prototype to test before the team has to commit |
What it Looks Like When There is No Map
A partner came to us with a product unlike anything we had worked on before: a new industry, an app with no real comparable in the market, and no category playbook to build instincts from.
Before we designed a screen or wrote a line, we used AI to understand the category. We surfaced patterns in unfamiliar territory, mapped category conventions, found what users cared about, and identified where competitors were falling short.
By the time we sat down with our partner to discuss direction, we had enough command of the category to lead the conversation.
How We Decide What Belongs
Before any tool enters our stack, it has to answer these questions.
- Does it simplify the work or add something new to manage? A tool that requires someone to own the prompt workflow, manage outputs in a separate system, and review every result before it enters the project is not saving time. The best tools integrate seamlessly into the workflow.
- A tool's output also has to be editable. An AI-generated image that arrives as a flattened file with no way to adjust the composition, swap the background, or layers gives you something to react to but nothing to work with. We need outputs we can shape, refine, and hand off.
- Does it improve the quality of decisions? Generating 40 headline variations is not useful if the team still has to review all 40 to find the two worth considering. The question is not how much a tool produces but whether its output helps the team decide faster and with more confidence.
- The real test is whether we return to it when the stakes are real. Any tool looks useful on a low-pressure brief with room to experiment. The ones worth keeping are the tools we consistently reach for when the deadline is tight, and the brief is demanding.
The Five Tools That Made the Cut
- Figma Make is a part of the Figma ecosystem, which was the deciding factor. When a design is ready to become interactive, Make lets us move directly from design to code without rebuilding from scratch.
- Figma Weave is our primary AI generation tool. Rather than locking us into one AI engine, Weave lets us choose the right model for the task and build node-based workflows that can be saved, refined, and reused. Once you know how to construct those workflows, nothing else comes close at the price.
- The Claude ecosystem covers three distinct phases of the work. We use Claude.ai early in the process, helping the team research, pressure-test ideas, and think through problems before a pixel gets placed. Claude Code handles development and prototyping. Claude Cowork handles operations, automating repetitive tasks that would otherwise consume time better spent on detailed design work.
- Replit stayed because it moved us from describing ideas to testing them live. Spinning up a working prototype and sharing it in a partner meeting changed how early we could get a real reaction.
- Jitter solved a recurring problem. Because animation takes time to produce, it rarely makes it into early creative reviews and usually gets cut or deferred. Jitter makes it fast enough to include in the first round, which changes the quality of decisions made before production begins.
A Tool is Only as Good as the Team Using It
Every design team has access to the same 30,000 tools. Choosing deliberately within that universe is what separates the teams doing strong work.
Without human creative judgement, AI-assisted work tends to converge and produce recognizable results. AI does not know what a brand should feel like, when something is technically correct but emotionally flat, or when an idea is clever but not true.
At Graybox, that judgment is built into how we work, in how we evaluate tools, how we run projects, and how we decide when AI belongs in the process and when it does not.
If you are building out a design stack or working through a project that needs to move fast without losing quality, get in touch. We have the process and the team to help.
