Adobe’s new AI assistant is a mediocre design intern — and that’s exactly what makes it interesting

AI image tools rarely make me feel like I’m part of the creative process. They are, after all, mostly designed so that people with no design experience can type in a few words and get back a usable result. So I was pleasantly surprised by Adobe’s latest take on an AI image assistant: it’s a tool that doesn’t try to replace human intuition but instead acts like a junior collaborator — one that needs guidance, feedback, and repeated direction.

The core experience

Unlike generative fill or text-to-image models that aim for instant perfection, Adobe’s new assistant inserts itself into the workflow as a second pair of eyes. It suggests rough compositions, offers alternative crops, and proposes color palettes — but none of these come as finished assets. The assistant requires the user to refine, reject, or build upon each suggestion, much like a design intern would. It deliberately avoids the one-shot output that defines most consumer AI image generators.

How it works in practice

When I tested the tool, I started with a simple prompt: “a coffee shop at dawn.” Instead of generating a polished image, the assistant returned three low-resolution sketches with different lighting angles and furniture arrangements. I could select one, then modify it by saying “make the counter longer” or “add a warm glow from the window.” Each iteration produced a slightly improved version, but never a final render. The assistant stopped short of completing the work — it forced me to make aesthetic decisions.

Why “mediocre” matters

The assistant’s mediocrity is by design. Adobe’s product team told me they wanted to build a tool that encourages iteration rather than passive consumption. “If it gives you a perfect image in three seconds, you haven’t learned anything,” one product manager said. “We wanted something that makes you feel like you’re directing a person, not just pressing a button.” The system intentionally introduces small imperfections — a skewed perspective, a mismatched shadow — that the user must correct, simulating the back-and-forth of human collaboration.

The creative loop

This approach shifts the dynamic from “generate and accept” to “suggest and refine.” Each time I corrected the assistant, I found myself thinking more critically about composition and lighting. The tool doesn’t store a library of pre-made assets; it generates each suggestion from scratch based on the current conversation. This means repeating a request often yields different results, further reinforcing the need for human oversight. The assistant learns from corrections within a session but resets completely between projects.

Market Context

As of today, May 29, 2026, Bitcoin is trading at $73,559 (up 0.3% in the last 24 hours), while Ethereum stands at $2,009.27 (up 1.0%). The broader crypto market has remained relatively stable this week, with total market capitalization hovering around $2.8 trillion.