At their annual MAX conference, Adobe unveiled new versions of most of their apps. One of the key elements in Photoshop, for instance, is a full embrace of MCP (Model Context Protocol), which allows the application to be in full communication and control of an outside app.
MCP is basically a conduit of communication. The MCP-enabled application broadcasts what commands it understands, while you use an AI chat engine to do the commanding. This is not generative AI (creation of something new), it is agentic agent (control of something). A lot of people think that generative AI is bad, but I’m betting those same people will see agentic AI as good if they give it a chance.
One common complaint about feature rich products such as Photoshop is that they represent a learning cliff to a new user. You’ve got to learn what all the icons and menu items mean, where they are, and then deal with all the sub-panels and dialogs they lead to. You know what you want to do (e.g. “brighten the shadows”), but where is the function for that? Moreover, in Photoshop, you’ll find that generic requests like that can be done in many ways.
What MCP allows is for you to run an AI chat engine in one window and Photoshop in another. You can then type what you want done (“lift the deep shadows a bit, shift the color slightly warmer for the subject, crop so that the fire hydrant on the right isn’t in the frame”). With voice input, you don’t even need to type that, you can just say it and have it put in the chat prompt box.
Most photographers have a general idea of the things they want to do with an image, but don’t know exactly which widget or tool they should be using or even if they know which one, where it is and how to use it properly for what they want done. Using an AI chat box as your agent, the dirty details go away and you just concentrate on the effect you’re looking for. Don’t worry, Photoshop has multiple levels of Undo, so should the AI not quite do what you asked for, you can back that out.
I’ve been using agentic AI with my Web site tool, my word processor, my spreadsheet, my presentation program, my app development tools, and more for quite some time now. And I’m much more productive because of that. AI is not creating any content for me, it’s just doing tasks I ask it to without me having to move my cursor all over the screen and step through every last little check box, slider, or other control.
One interesting thing about doing that is that I’ve learned a few new things about using those applications, too, as I can watch what the AI chat does in the application as it’s doing it. So now when you say “Create a new layer and apply a bit of extra saturation to the subject” you can see what the AI actually does to make that happen (create a new layer, select the subject, apply an effect). That’s a really simple example, but consider “duplicate the layer and apply a slight darkening effect only to pixels in the underlying layer that are brighter than 245.” That’s blending. Do you know how to blend in Photoshop? Your AI chat does. It might be able to teach you a few things along the way.
I’m sure Photoshop YouTubers are 100% on top of this and this week will be giving you infinite demonstrations of how all this works. It’s actually pretty cool and useful when applied well.
There’s just one teeny problem: every AI gets set up for MCP slightly differently. And it’s not always obvious how to do that. Adobe is publishing steps to use for common AI models, but if you’re using one of the older local AI choices, it might not be obvious at first how to set up the connection. You may also first need to install node.js, depending upon which AI you’re using. (Help is an AI query away, or at least a Web search.)