I don't use just one AI. I use four. Not because I'm an unhinged technology enthusiast but because after months of testing, breaking things and discovering limits, I understood that each one has a different role in my process. Like a team where each member does what they do best.
This is my real workflow. Unfiltered, no romanticizing. What works, what doesn't, and why I keep using all four.
Claude: my coding partner
If you are reading this on this website, you are seeing Claude's work. I built this blog from scratch. No WordPress, no templates. Pure PHP, MySQL, HTML, CSS. A personal challenge that became something much bigger than I expected.
Claude is who I talk to when I need to solve technical problems. I don't tell it "make me a website." I explain the architecture, show how my database is organized, describe what I want to happen when a user does something. It's like having a senior developer next to you who never gets tired of your questions.
What I discovered with Claude is that the quality of its help is directly proportional to the quality of my explanation. If I give it a vague brief, I get vague code back. If I explain the full context with details, I get solutions I sometimes hadn't even thought of. Recently I asked for help creating a scheduled posts system and it didn't just give me the code but explained why certain technical decisions were better than others.
What I value most is that Claude doesn't treat me like I know nothing. It recognizes my experience and builds on it. When I mention I'm a designer with over ten years of career, it adapts its responses to that level. It doesn't explain what a div is. It talks to me about component architecture.
Midjourney: my infinite photo studio
Midjourney is where my designer mind feels the freest. Generating images with this tool is like having a photo studio with an unlimited budget where you can shoot a thousand times without anyone complaining.
But that freedom has a cost: you need to know exactly what to ask for. And that's where my years of visual design experience make the difference. I don't write "photo of a person." I think about lighting, camera angle, fabric texture, lens quality, emotional atmosphere.
My typical prompt has layers. The first describes the subject. The second the photographic style. The third the technical details. A real example I use often:
active lifestyle, casual colorful clothing, sunshine, neutral expressions, DSLR style --v 6 --ar 3:2
Then I iterate. I change "sunshine" to "golden hour" and the result is completely different. I change "DSLR style" to "The shot was taken with a Leica Q2" and the entire image texture transforms. Every word is a design decision.
What has been hardest to master is the low angle shot. It was a headache for weeks. Midjourney tends to interpret certain angles very literally and I ended up with distorted perspectives. But with patience and many iterations I found the word combinations that produce what I need.
ChatGPT: my idea editor
ChatGPT isn't my main tool but it fills a role no other does: it helps me think out loud. When I have a half-formed idea for a design project, I explain it as if talking to a colleague. And it returns questions or angles I hadn't considered.
I use it a lot to refine prompts before taking them to Midjourney. I describe what I want to achieve visually and ask for photographic or artistic terms I can use. Sometimes its suggestions are obvious. Other times it throws out a word that changes the entire result.
I also use it for quick research when designing for a sector I don't know well. If I have a project for a sportswear brand and need to understand visual trends in the sector, ChatGPT gives me an overview in minutes that used to take hours of searching Pinterest and Behance.
Where it doesn't serve me as well is code. I've tried it and its solutions tend to be correct on the surface but fragile in the details. For technical things I prefer Claude without a doubt.
Gemini: the pinpoint specialist
Gemini is the AI I use least but the one that surprises me most when I need it. I don't use it for long conversations or complex projects. I use it for very specific tasks where I need a quick second opinion.
For example, if Claude gives me a technical solution and I want to validate it, I pass the same problem to Gemini to compare approaches. Sometimes they agree. Other times Gemini proposes something different that makes me reconsider. It's like consulting two doctors before surgery.
I also use it when I need image analysis. Gemini has good ability to describe what it sees in an image and that helps me understand how an AI interprets visual compositions. It's an interesting exercise that has taught me a lot about how these machines think.
But I have to be honest: Gemini needs a short leash. If you give it too much freedom on a creative task, it drifts. It works better when the task is well-defined and the expected result is clear. Giving it unlimited space creates problems.
How the four flow together
A typical day working with AI looks like this: I start with an idea or a project. If I need to explore visually, I go straight to Midjourney. If I need to think through the idea before executing, I talk to ChatGPT. If the project involves code or something technical for the site, I work with Claude. And if at any point I need a quick second opinion or to analyze something specific, I turn to Gemini.
It's not a linear process. Sometimes I start in ChatGPT, move to Midjourney, go back to ChatGPT to adjust the prompt and end up in Claude implementing the result on my site. Other times I work only with Midjourney all morning and Claude all afternoon.
The important thing is that each AI has its place. I don't try to make one do another's job. That was a mistake I made at the beginning, wanting a single tool to do everything. It doesn't work. It's like wanting Photoshop to do what Figma does or vice versa. Each tool has its strength and your job as a designer is knowing when to use which.
What this workflow taught me about design
Working with four AIs every day forced me to become a better communicator. Each one needs you to speak to it differently. Claude needs detailed technical context. Midjourney needs precise visual vocabulary. ChatGPT needs an open conversation. Gemini needs bounded instructions.
It's exactly what we do with clients, with development teams, with art directors. We adapt our language depending on who is on the other side. The difference is that now on the other side there is a machine and the feedback is immediate.
As I wrote in my post about what a prompt is, this is not a new skill. It is an evolution of what we already know how to do. We just have four more tools on the belt now.
And the belt keeps growing.