Skip to content
How to use coding agents without losing your mind?

How to use coding agents without losing your mind?

When they say AI is coming for your jobs, it doesn’t mean it’s gonna replace all programmers directly. It means using generative AI tools aggressively for coding reduces the problem-solving skills of engineers. That’s where coders lose their edge. I was going through the recent MIT study and the findings are interesting and alarming at the same time.

In the study involving around 54 participants aged around 18-39 are divided into three groups

1.) Can use AI tools (mainly ChatGPT)

2.) Can use search engines like Google

3.) No access to ChatGPT or search engines

Here are the key findings:

  • Participants who used ChatGPT showed significantly lower brain connectivity in regions linked to memory and cognitive engagement compared to other groups.
  • Users relying heavily on ChatGPT showed less critical thinking, reduced ability to assess their own work, and exhibited dependence on AI-generated outputs.

Here's the nice thread about the study for those who are interested:

Cheap dopamine

If the effect on brain for writing essays is such damaging, think of using it for coding for longer term without any mindfulness. The reason for the tools like Claude code or other coding agents being so addictive is the good old problem: Cheap dopamine

We used to get it from Doom scrolling social media apps, now we also get it from Vibe Coding. Dopamine is a pleasure/motivation hormone. Whenever we accomplish something the brain releases it to give a sense of pleasure. For example, implementing a nice animation for a button will give you a sense of accomplishment via the reward hormone so you feel good.

With current coding tools, instead of building component by component, we started building fullstack apps. By building each screen and when they see the output, they get a sense of accomplishment which would have took weeks to complete in just a few hours. This is a massive shortcut, compromising cognitive effort and problem solving skills. The more you do that, the more the brains wants it. And this cycle is addictive. This should stop.

I'm not completely against using AI assistants for coding, infact I highly recommend them. But we could do so much good for our brains if we use them as the name states - "assistants". In lot of cases, the AI tools become the master. Some even run their claude code sessions with "--dangerously-skip-permissions" mode for production apps which is quite crazy. Its acceptable for experimentation but for production apps, its a big NO for me. Pretty soon, you’re not really solving problems anymore. You’re just churning out features as fast as possible. Security? Testing? Performance optimization? Who cares - as long as the next dopamine hit comes along.

One of my favourite books "Dopamine Nation" describes this perfectly. The author neatly explains how easy access to quick dopamine hits like scrolling social media, fast-food deliveries, etc., can rewire your brain and leave you addicted to constant stimulation. And addiction is exactly what’s happening to many coders today.

So how do you use AI tools without losing your mind or your skills?

Simple: Always do the hard part yourself - which is planning!!

Your biggest value as a coder isn’t typing. It’s problem-solving, critical thinking, designing system, and making smart architectural decisions based on your experience.

AndrewNg said the same as well:

10x engineers don’t write code 10 times faster. Instead, they make technical architecture decisions that result in dramatically better downstream impact, they spot problems and prioritize tasks more effectively, and instead of rewriting 10,000 lines of code (or labeling 10,000 training examples) they might figure out how to write just 100 lines (or collect 100 examples) to get the job done.

Here’s how you can actually do it:

Master the planning phase yourself

Don’t jump straight into code with AI agents. Start with a pen and paper or a blank whiteboard. Plan your architecture, user stories, optimization strategy, performance considerations, and UX carefully. Ask yourself:

  • What’s the goal of this app?
  • How will it scale?
  • Where could it break?
  • Is it secure?

Force your brain to do this first. Once you have the plan, then use AI to speed up repetitive tasks or boilerplate code.

Review and question every line generated by the agents

Never blindly copy-paste or accept edits from Claude or ChatGPT. You should read, understand, and question every single line. Ask yourself:

  • Why did the agent choose this solution?
  • Is there a simpler or more efficient way?
  • Is there a security issue here?

Engage yourself in the process mindfully by actively reviewing and optimizing AI-generated code If you're a senior engineer, consider AI agents as Junior devs - look for gaps, provide feedback and iteratively improve the code. If you're a beginner or entry level engineer - treat AI as tutor to learn best practices and system designs via hands on exercises and tasks.

Set boundaries and limits on AI usage

Consider AI agents as a tool to increase productivity, not as a replacement for effort. Limit AI use to specific coding tasks:

  • Writing basic boilerplate code.
  • Auto-completing repetitive snippets.
  • Suggesting alternatives when you’re stuck.

Don’t let AI do all your thinking. Your brain needs the challenge to stay sharp and grow stronger - as any other muscle.

Consciously slow down occasionally

Speed can be addictive. Speed without any control can be dangerous too. Deliberately slowing down to review code, write tests, refactor, or learn fundamentals deeply is crucial. It strengthens your brain’s reward circuits differently- slowly, and safely.

Never stop learning or exploring

Always learn from the experts. For example, I love Martin Fowler’s blog posts - they’re a gold mine built from years of hands-on experience. Reading high quality content like his regularly shapes how you think and plan your code.

Also read actual opensource code of popular codebases. Personally, I regularly dive into smart contracts and audit reports from top protocols to understand the patterns and design decisions that subconsciously improve my planning processes.

Consider including the above practices in your routine and you'll be unstoppable.

I took programming as an example, but the same applies for other domains as well.

Remember: quality > quantity.

© 2025 PraneshASP. All Rights Reserved.
GitHubTwitterEmail