Using AI To Fix Your Bad Habits
It's hard to keep yourself accountable. Let AI help you.
1. in AI Journal
It’s hard to keep yourself accountable—avoidant behavior is practically universal. Our brains are designed that way. Survival instincts have been shaped through millions of years, and that makes modern comfort novel and unprecedented. The reason is that human hardware simply hasn’t adapted to it yet. What we call procrastination was once a real threat that could kill in the olden days.
What do you do about that? You want to achieve your peak potential, but you can’t defy your tribal nature. You can’t stop your senses or disable instincts. And honestly, neither should you. The real issue lies in our blindness—it’s hard to notice your own shortcomings because we humans let time pass without progress, followed by the wonder of where it went away.
You could go to therapy or get a friend to help, but they’re not really made for that work; at least not directly. In reality, most of the responsibility falls on us. As an option, you could spend years analyzing your behavior, learning, decoding patterns, and designing systems that work for you. Unfortunately, no one has time for that. Unless you’re me.
But that’s what makes AI unique in this position. AI models can see connections across many domains; call you out on your bullshit; and not grow tired of you. Since the advent of AI, we finally have the tools to fill the gaps that self-help material and positive affirmations (wishful thinking) left unaddressed. I personally tried it to see whether I’d be able to change my behavior and take the risks I should have taken long ago, or if I’d still be missing something crucial.
Late November, news of the recently released Opus 4.5 model for Claude started popping up. It broke records in many benchmarks and everyone was praising it, so I naturally got curious about its capabilities. But looking through the demos wouldn’t cut it for me. I wanted to purchase the subscription to explore full capabilities and conduct some experiments with it. That’s exactly what I did.
I came up with the idea of comparing different AI models from the observational and philosophical perspectives, such as how they reason through different situations and perceive/judge various phenomena. It was fun seeing the differences between the responses of different models.
Suddenly, a thought came to me, “what if I could use AI to fix my procrastination?” A while back I had made a list of all the things I avoided or simply forgot about. The feeling of unaccomplishment was quietly stalking me for the last couple of years. So without any expectations, I went right into discussing it.
As I’ve mentioned above, AI can be pretty good at describing the things about yourself that you are blind to. That should be the primary goal in this kind of learning. It’s because the solutions that a model might present usually reveal your patterns between the lines, if not directly.
And it’s funny how quickly my pattern was uncovered. One of the tasks was developing my website (yes, this one.) I could’ve come up with thousands of excuses why I still haven’t started: I never made a proper website, I still don’t understand my identity, I don’t have my writer’s voice developed. For a perfectionist like me, it’s all of the classics.
But when I explained my situation to the AI, it pierced me through my words. The truth is, I didn’t have an actual reason not to make the website.
Most people simply never fathom it—our excuses are mostly shaped by irrational thought, fears masked with words. Whatever shapes they take, no matter how comforting they feel. People like me, especially, reason and act according to our intuition, or more specifically, our emotional state. There are true reasons beneath the surface, definitely; you typically just never notice, while subconsciously trying to rationalize it with logic. On top of that, there is always tomorrow, so why not do it all later? That is the silent momentum killer. Using AI to tackle both beliefs is a valid strategy in this case.
That’s the thing, though; if there was a legitimate blocker, you wouldn’t keep coming back to see whether you can finally make something, you wouldn’t keep thinking or dreaming about it; you simply wouldn’t worry about it because you’d already give up by then. There’d be nothing to resolve.
To combat my problem, I was offered a trivial solution: start the website now and design it poorly. Don’t know what framework to use? Use React (I already knew it at the time.) Don’t know how to implement something? Watch a short tutorial or ask a chatbot. The AI approach was strategically simple… and we’ll get to its limitations later.
My first instinct was, “you can’t do that!” But if you think about it, well, why can’t you? Really, why not? What’s the worst that could happen?
That’s where learning happens. Once you’re offered an idea that hints at something within you, you get a natural tension to see where it leads. You want to minimize uncertainty (just another quirk of our survival brains), so you instinctively lead yourself to explaining your true concerns, and that’s when the model finally learns enough to reflect its synopsis back at you. You should remember this process.
We would eventually get to the core problem—my perfectionism and identity crisis. As a human, I’m pretty resilient and adaptive, but when it comes to losing meaning behind work or it becoming too resistant, progress dissipates. It was my true blocker, and that’s the pattern me and AI have managed to observe, thanks to this approach. It’s also a blessing and a curse that Claude’s approach is of a ruthless commander, the one that doesn’t listen to your excuses and prioritizes the truth without bias and hesitation. So my first effort towards my rehabilitation began under weighing pressure and some determination…
On the first day, I ended up resuming my website project.
It had actually worked.
I would continue explaining my problems every time, seeing what it concludes, and acting accordingly. That was my plan of action. Its responses weren’t always perfect—AIs can overestimate, whether it’d be due to technical limitations or insufficient context, so it’s important to trust your instincts too. Don’t trust anyone blindly, not even yourself.
But in a single week I managed to finish a rough version of the website, let a friend review it, and even ship it live. Since then, I would just proceed with the same strategy. It would continue to work.
As time passed, I also began noticing substantial differences between different models’ approaches. For example, the aforementioned commander personality of Claude is contrasted with the more emotional side of ChatGPT. Yes, AI models don’t have “consciousness” or “emotions,” but it’s common for ChatGPT to be more emotionally attuned than Claude.
Claude might advise to finish the website within a concrete deadline, since it’s the next logical step, whereas ChatGPT might notice the emotional barriers, such as that subtle hesitation catching up before a new achievement, explain why they occur, and help you reframe the situation to dissolve them. It helped me multiple times during the development of the website.
But that’s just my experience. AIs are also sometimes known for being sycophantic, usually exploiting your cognitive biases. It’s important to be aware of the possible manipulation from the models. Try to verify AI’s claims before accepting them. They are not very good at hiding the truth, so even just asking for the reasoning is usually enough. Also learn the models’ strengths and weaknesses, as they can hugely influence your progress. Try different communication styles, switch to other services, and choose the models that fit you best. Later you can also restrict the scope for each you wish to use.
So, how do you replicate the learning strategy for yourself? I’ll outline the basic steps:
- Start by outsourcing all your judgment to AI. Explain your goals and worries in detail, then try its approaches to see the effects. Ask the chatbot for more information in order to understand its decisions better.
- Extract the successes and failures you had into distilled patterns. The goal is to understand the right thought process. It will slowly transform the effort into concrete insights.
- Crystallize your understanding by supplying it with practice and continuous adjustment. Gradually reduce AI’s influence. This is where you prove your improvement through discipline and independence.
In other words, you become a child under your parent’s supervision, just in a limited form. Then you mature and can handle the world’s mercilessness again.
It probably won’t be a linear progression, either. You might instinctively understand some aspects of productivity better than others. You might gain insight just to find more unsolved questions. Going back to previous steps in specific areas is a good idea. Don’t be afraid to experiment repeatedly—that’s how you find what sticks.
The productivity I was talking about the entire time is assisted self-improvement. The traditional notion of AI productivity is outsourcing the tedious and complex work to be done for you. To be honest, I wouldn’t advise giving one up for another, and instead combine the two to work in harmony. If time or external pressure is an issue, then outsourcing the work can be smart. If you make AI do the work you are expected to be competent in, then it’s better to start teaching yourself now. Asking the models for their rationale applies here quite well, too.
That addresses most types of usages of AI productivity. At the end of the day, AI wasn’t created to replace human beings (even if it replaces some of the job positions.) It was created to accelerate humanity forward. If we learn how to use the tools to improve ourselves, going from observing its decisions to extracting insights from them, we’ll be more satisfied with life, thanks to our improvements and lower levels of stress. But let’s not forget that everybody’s mind and work style is unique. I’m not trying to claim that this approach is for everyone.
So experiment, find what works for you, and let that propel you towards success and fulfillment. You define your success, not the tools.