{"id":1109,"date":"2026-05-29T11:53:25","date_gmt":"2026-05-29T11:53:25","guid":{"rendered":"https:\/\/silvybrand.com\/?p=1109"},"modified":"2026-05-29T11:53:25","modified_gmt":"2026-05-29T11:53:25","slug":"adobes-conversational-ai-agent-is-a-mediocre-design-intern","status":"publish","type":"post","link":"https:\/\/silvybrand.com\/?p=1109","title":{"rendered":"Adobe\u2019s conversational AI agent is a mediocre design intern"},"content":{"rendered":"


\n<\/p>\n

\n
\n

AI image tools rarely make me feel like I\u2019m 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\u2019s latest take on an AI image assistant: It\u2019s a bot designed to take away some busywork, while still granting you creative control.<\/p>\n<\/div>\n

\n

Unlike AI generators that are specifically designed to make and edit images or video, Adobe\u2019s Firefly AI Assistant<\/a>, which I\u2019ve been testing in beta, is more like a multitasking middleman that can operate Adobe\u2019s design apps for you. On its website, Adobe says that you can just \u201ctell Firefly AI Assistant (beta) what you need, and it will use tools from apps like Photoshop, Illustrator, and more to complete multistep projects in moments.\u201d<\/p>\n<\/div>\n

\n

The user interface looks like a typical chatbot. There\u2019s a text box you can type prompts into, and a plus symbol for uploading media files. It doesn\u2019t use the actual Adobe apps on your computer, but it has access to common capabilities like masking, object detection, and image generation. The AI assistant is designed to be conversational, so you can ask the chatbot to \u201cmake this photo more colorful,\u201d and it\u2019ll do so while explaining its actions.<\/p>\n<\/div>\n

\n
\n
\n
\n

1\/3<\/strong>

Here\u2019s an unedited photo of myself that I used for testing. I intentionally chose a shot with unusual lighting.<\/em><\/figcaption>Image: Jess Weatherbed \/ The Verge<\/cite><\/span><\/p>\n<\/section>\n<\/div>\n<\/div>\n<\/div>\n
\n

Photo edits and illustrations completed by Adobe\u2019s AI are convincing at a glance. It changed my hair color in one photo and then the background location and lighting in another. The results aren\u2019t perfect: some had colors that were too vivid or alterations that hadn\u2019t been properly blended into their surrounding environments. But I suspect the average person wouldn\u2019t assume that my results were made or manipulated with AI \u2014 it just looks like the work of a novice designer.<\/p>\n<\/div>\n

\n

What actually makes the Firefly AI Assistant intriguing is how it interacts with you. I gave it a picture of my cat by a window and asked it to make the sky cloudless and sunny. It didn\u2019t just go \u201csure\u201d and give me the edited image. The chatbot described the scene in the pre-edited image with surprising detail (it correctly identified that my cat is a Maine Coon despite the photo mostly just showing his ass), and then explained how<\/em> it\u2019s going to achieve the results I\u2019d requested. It mentioned specific tools from Photoshop and Lightroom using established editing terminology, explaining the process step by step. You don\u2019t actually get to see the image being edited in real time, but the chatbot will tell you which features it\u2019s using to achieve each result.<\/p>\n<\/div>\n

\n
\n
\n
\n
\n
\n
\n
\"\"<\/a><\/div>\n<\/div>\n<\/div>\n

Here\u2019s an unedited shot of my cat, Trevor, observing his kingdom\u2026<\/em><\/figcaption>Image: Jess Weatherbed \/ The Verge<\/cite><\/p>\n<\/div>\n<\/div>\n
\n
\n
\n
\n
\"\"<\/a><\/div>\n<\/div>\n<\/div>\n

\u2026and the results of me asking the chatbot to \u201cremove the clouds in this image and make the sky look sunny and bright.\u201d<\/em><\/figcaption>Image: Jess Weatherbed \/ The Verge<\/cite><\/p>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n
\n

The Firefly AI Assistant is also surprisingly forthcoming about its limitations. When I asked it to separate objects from a JPG file into separate layers, Firefly said it couldn\u2019t do so, but offered two different courses of action for splitting the image into separate elements, explaining the pros and cons of both. After I chose one, the bot then described its editing process, including the fact that what it was doing wasn\u2019t working. \u201cI notice the gaussian blur approach isn\u2019t giving me true transparent cutouts \u2014 it\u2019s outputting full-image PNGs,\u201d it wrote. The chatbot redirected itself and used masks and Adobe\u2019s image cropping and resizing tool instead.<\/p>\n<\/div>\n

\n
\n
\n
\n
\n
\"<em>It<\/a><\/div>\n
\"<em>It<\/a><\/div>\n