AI makes you move fast, but you have to slow down and think clearly enough. If you don’t, you get stuck in a loop and go nowhere, and AI stops feeling as fast as it did when you started the project.
Coding with AI is changing so quickly, and it’s affecting a lot of people. For some it’s a lifeboat, for others it’s the opposite.
What Is Coding With AI Really Like for Me?#
I love coding by hand. I still enjoy it, and I still learn something new every single day.
At the same time, I’ve been using AI tools and AI agents a lot. They cut down the time I spend on research, on searching for tools, and even on helping clients build websites and other projects. I can offload the boring tasks to it, which means I can handle a lot more work than before, and that’s the good side for me.
There’s a downside too. Some tasks got smaller, because clients now use AI tools themselves. They build a website, then hand it to me to clean up, check, and deploy.
What Happened When My Startup Project Started to Feel Slow?#
On a startup project, I was responsible for the frontend. I got it up and running, and the main layout came together fast. Then it started to feel slow.
I was using OpenAI’s Astra model, and it didn’t go as fast as when I first got started. Then I realized why. I didn’t have a clear picture of what the user flow looks like, what options should be available for users, or how the UI should be structured.
My prompts were generic, and I hoped the AI would figure it out for me. I started to become lazy. Then I started to plan and think more before prompting, and things moved forward fast again.
AI makes you move so fast, but you have to slow down and really think.
So Who Is Actually Driving?#
The AI agent is only a tool. You are the driver. AI is a good servant, but a terrible master.
You still need expertise, a clear vision, and a clear design of the project, or else you can only go as far as your vision.
What Happens When the Picture in Your Head Is Clear (and When It Isn’t)?#
You can go as fast as you want with AI. But the picture in your head has to be clear.
It doesn’t have to be clear on the pixel level. You just need the big picture, the design, and the flow roughly mapped out.
When you can describe what you really want in as much detail as you can, that’s when the two of you move fast together. I don’t even need the latest, smartest model. A lot of the time I just use a free model, and I get good results from it.
When it isn’t clear, my prompt is general and I hope the AI will figure it out for me. Many times it doesn’t, and honestly, of course it doesn’t. You don’t even know what it is, so how can you describe it?
You get whatever the model found. If it happens to be what you wanted, great. If not, you’re prompting, re-prompting, or going back to sit down and think.
Even the latest, smartest model struggles when you don’t have a clear picture of what you’re going to build.
Why Do the Details Make Your Product Stand Out?#
Think about it. If everyone can use AI to build anything, then what makes you stand out?
It’s the detail. It’s the design, the experience, the UI. It’s the time you put in to build your product with care, like an artist or a craftsman.
Everyone can type “build a fitness app for me.” Everyone gets the same general thing back, not much different from the next person.
But if you build as an expert, you can say more. You can tell it which algorithm to use, which pattern or design you want, how the UI should look, and so on. That expertise is what separates you, and the product ends up reflecting you.
Some people treat AI like a slot machine. They put in a coin and hope for the best, and what comes out is plain and general. Nothing new, nothing creative, nothing unique.
When you put in real effort, the product still feels like a craft. That’s the difference.
How Do You Prompt Like an Expert and Not Like a Lazy Person?#
I’m not avoiding AI. I still use it, but I use it the way an expert uses it, not the way a lazy person does.
No matter how good your prompting technique is, if you have no idea what you’re doing, your prompt will only be as good as what you know. In the end, a prompt is just telling it what you want, and the more detail you give, the better the result.
So it starts before you type anything. Here’s what my pre-prompt plan looks like:
- Know the big picture of what you’re building before you prompt
- Map out the design and the flow in your head first, or on paper
- Write down the user flow, so you know what it looks like
- List the options that should be available for users
- Decide how the UI should be structured before you ask for it
- Describe what you want in as much detail as you can
- Use your expertise. Tell it the algorithm, the pattern, and how the UI should look, and put care into the details and the experience
On my startup project, the user flow, the options, and the UI structure were exactly what I was missing.
Here’s what that looks like in a real prompt. This is the lazy version:
Improve the UI according to the best practices as you see fit.
I’m basically asking the AI to decide everything for me. Here’s the version I wrote after I stopped and thought:
Update the content editor page, where users write in markdown and get help from AI.
Change the content editor section into two columns with a 2:1 width ratio.
Left column (two thirds): the markdown editor.
Right column (one third): an AI chat box, with the message history on top and the input at the bottom.
User flow:
- The user writes in the markdown editor.
- The user highlights some text.
- The highlighted text appears in the chat box as a quoted reference.
- The user sends a message, and the AI reads the highlighted text as a reference when it answers.
Use the existing components and match the current styling.
Same tool. The only difference is that I knew what I wanted.
How Do You Keep Learning in the Age of AI?#
For me, learning in this age of AI means going deep and becoming an expert. That’s why I still code by hand, and why I’m not going to stop learning.
We need to learn more than ever, and go deep. I wrote more about this in Stop Finding Shortcuts and Start Learning the Hard Way.
- Go deep on what you’re working on or interested in
- Explore unknown territory just enough to get an idea, then keep going
- Keep coding, drawing, painting, or whatever your craft is, by hand, so your skills stay yours
You don’t have to go deep on everything. Just get a feel for the new stuff, enough that you feel okay, and keep going deep on whatever you’re really into. That helps in the long run.
If you want a solid place to go deep on web fundamentals, the MDN Web Docs are a great start. And before you ship anything an AI built, a website checklist will save you from a lot of headaches.
What Happens If AI Gets Taken Away?#
We don’t know how good AI will get, or how it will be in the future. But investing in yourself gives you the best return.
Think about it this way. If one day AI is taken away from you, who do you become? Is your identity gone with it? Are your skills gone too?
Or can you still operate normally, with as little turbulence as possible, as you move back to a world without AI? That’s the question I keep asking myself.
Final Thoughts#
Coding with AI is a good thing. It helps me do more, and I’m grateful for it.
But AI makes you move fast, and you have to slow down and think clearly enough. The tool only goes as far as the driver can see, so keep thinking, keep learning, and keep building yourself. That way, with AI or without it, you’re still you.
