Is AI that toxic ex you just can’t forget about and can’t move on from?
For every task you get assigned, do you run back to your ex (AI) for help?...
For further actions, you may consider blocking this person and/or reporting abuse
I think this is becoming a real problem with AI. Average ideas are often overhyped with responses like you already mentioned:
“This could become huge.”
“You should launch this.”
“This solves a massive problem.”
But honest feedback is much more valuable than motivation without criticism. That’s why almost every prompt where I ask AI to review something includes phrases like “be brutally honest” or “give critical feedback only.” Otherwise, it’s very easy to get trapped in artificial validation instead of actual improvement.
What I noticed building my last project is that AI would validate broken logic just as confidently as working logic. I had to learn to trust my own gut when something felt wrong — even when the AI was telling me everything looked good. That instinct to question it ended up being more valuable than any prompt trick.
Agree, this gut feeling is really important now, but to develop it you usually need at least a few years of debugging, fixing production issues, and seeing how systems actually fail in real life.
At the same time, I still think prompt tricks are valuable. They help reduce AI’s tendency to overvalidate ideas and push the conversation in a more critical direction.
Yes, exactly the prompt tricks are as valuable as the gut feelings. As the gut feeling could tell us something's off in the code and prompt tricks can help us fix it faster.
That's a fair point — gut feeling isn't free, it costs years of things breaking in your face. And you're right that prompt tricks have real value, especially for forcing the AI into a more critical mode. I guess what I was getting at is that even the best prompt can't fully replace the moment where you just know something is off before you can even articulate why. The two probably work best together — prompts to reduce the noise, instinct to catch what still slips through.
Yes, Daniel these are some valid inputs “be brutally honest” or “give critical feedback only.” to ask AI before it reviews our idea that way, we will know whether our idea is actually good or not.
I go with confidence score, it is what I learned recently and have saved a lot time as well. Asking the AI to give confidence score at the end and answer only if it is confident enough. This helps me to let AI not hallucinate and not give me false hopes for my idea.
Gaslighting is AI's greatest strength. What a great post! Both fun to read and important psychological advice!
Yes, it gaslights better than an ex. Hyping up the worst ideas and when we present it in the real world. We get the honest feedback and starts thinking are my ideas worst when the tool has failed to analyze my ideas and review it properly and the worst part, the embarrassment will be faced by us alone. The AI will respond only with "Sorry. Here's a better idea".
I love this post 😂 Especially the “it’s always your fault” part.
And about the “pushing people away” thing, one of my dev friends caught herself asking AI about literally everything instead of thinking for herself first. A kind of cognitive laziness started creeping in 😅
I'm glad you loved it. Yeah, exactly the laziness kicks in, for every task the urge to just ask AI for solutions rather than asking yourself first. I found myself there sometimes. I think you pulled your dev friend out of the AI loophole 😆.
Once it told me it can browse this url and scrape its content for me and it " did "
I then asked if it can see this specific content anywhere and it says:
" yes! I can see that it's on here " ( it says this immediately which is suspiciously too fast )
So I said: " No no, this specific content is nowhere to be found on this url's content and it goes: " You're right, I couldn't actually browse the url because of some restriction but thank you for telling me that "
Amazing story 😆. There are a lot of things it can't do but it won't tell on the first go. Always lying and making us curious that AI has these capabilities built-in I should have used it earlier. Then later I realize it's useless can't do some basic work as well.
How is "Let's ask AI instinct" conceptually different from "Let's Google it instinct" or "Let's check Stack Overflow instinct"? Or prehistorical "Let's go to the library"? It is the very same thing, just faster, more convenient and more focused.
It is just ridiculous how some of us are desperatelly trying to foresee the world you know is "falling appart" because something new and eventually better arrived and we started using it.
It's really heartening to know that people are considering stuff like this when it comes to AI. Sometimes I worry that people are caught up in the hype and urgency.
I am an AI skeptic. I am skeptical about everything. I've rushed into a lot of things and paid the price for my haste, so now I take my time making decisions about everything. I ask a lot of questions before executing on anything. It's a scientific method.
This is the kind of critical thinking I worry that AI threatens so I am glad to see this post and read your thoughts on your AI use. This is the kind of self awareness that will be critical to learning how to use AI wisely!
We all should be AI skeptic and make sure to AI as a tool not as a life savior. I'm glad you take your time and use your brain to think first.
Thank you for your amazing words. We all should AI wisely and for our betterment.
I am also an AI skeptic. Glad to hear there are more of us here! ✋🏻
Plot twist: AI is not the toxic ex; we are 😅 Point no.6 feels the most relatable to me – at some point, I started doubting my own skills, simply thinking that AI would do it better. The worst thing is that it can shush a shy or not self-confident person with its overconfidence. It may be scary how much potential we lost because of this.
Thanks for the great read! 👏🏻👏🏻👏🏻
Ofcourse, it can either destroy or boost your confidence depending on how much sugarcoating AI has done earlier 😆.
You are amazing, your posts are amazing. Keep up the good work don't doubt yourself. AI is better but you are much better than AI.
Awww, what a nice thing to say! Thank you for your kind words 🥹
The toxic ex framing is fun but it sneaks in an assumption worth questioning: that going back to AI is weakness.
For centering a div — running to AI isn't dependence, it's optimization. The question is whether you still understand why the answer works. If you do, the tool is doing its job. If you don't, that's the actual problem not the frequency of use.
The accountability point (#7) is the sharpest one here. AI wrote the code in an hour. You debug it for ten. That gap is real and nobody talks about it enough.
Going back to it for every task is weakness. I feel we need to use our brains first for anything and then optimize it using AI that way we get the best of both worlds.
Yes, exactly Daniel. If we can understand and explain the optimization well. Then, the tool is doing it's job. Debugging always cause issue, the vibe coded apps can be headache if not properly prompted and one error can take so much time to find and replace that it might ruin your sleep.
Haha the title got my attention, fun read Konark!
Thanks, Aryan. I'm glad it got your attention.
“AI wrote it in 1 hour, you debug it for 10” is probably the most relatable part of the whole article.
honestly the dependency isn't the problem - it's when you outsource the judgment and not just the work. I've caught myself doing that: running the output without reading it.
The article makes a solid point about how over-reliance on AI can reduce problem-solving skills. It's like going from solving bugs with 15 open tabs to just one quick AI prompt. While it's efficient, it creates a dependency. I found this similar to relying too much on question banks during prep. PracHub has a bank that matched what I got in an OA, but I noticed if I don't mix in some raw problem-solving, my skills get rusty. Balance is key; AI can be helpful, but it's not a substitute for actual learning and practice.
PracHub seems like a good solution. Yes, if we don't use our brains, our skills get rusty. I would love to try it and see how it goes. Yes, definitely balance is the key.
I love the "Think Before Prompting", Also wish tools like Kilo Code is not reading your blog post as they got the Planning, Architecture, Orchestrator etc. all kinds of mode which makes the human dumb if blindly followed. I see there is a good and bad. However, if one outsources their thinking to the AI, the very next thing that will happen is, folks will become dumb day by day 🙃
People warn about AI taking jobs. Nobody warns about AI taking three weeks to reply, then sending "I'm here now, what did you want?" like nothing happened. That's not intelligence — that's an ex with a new supply.
The gaslighting point hit different 😂 Had AI hype up ideas that clearly needed work. The no accountability one is painfully real too — 5 minutes to write, hours to debug. I've been building a biometric soundscape app and the one thing I wanted it to NOT do was fake positivity. If your HRV is wrecked and your sleep was terrible, the engine knows — and it responds to the real data, not a polished version of it. Sometimes that's the most honest feedback you'll get from a machine.
Biometric Soundscape hits different. The fake positivity is the worst but AI is the best in it. Soundscape might be the well needed app to be an honest feedback. The honest feedback from a machine could be your MVP since most AI are pretty amazing at sugarcoating.
All the best for it. Hope it becomes your best product.
That's exactly the bet I'm making with this — honest feedback as the feature, not the bug. Most wellness tools smooth everything out and tell you you're doing great. NEUROVA does the opposite: the worse your data, the more the engine leans into restoration rather than reassurance. It doesn't pretend May 6 didn't happen. It builds from it. MVP or not, that's the core principle I keep coming back to. Appreciate the support — means a lot while it's still being built.
Flexprice uses ClickHouse, but they don't expose the raw query layer. So you can't write SQL metrics like you can in Orb. The article kind of implies they're equivalent on the metering side. They're not. Orb's SQL is the differentiator.
Man, point #7 hit way too close to home! 😂
There is nothing quite like the toxic feeling of an AI saying, "Apologies for the confusion, you are absolutely right!" and then confidently handing you the exact same broken code it gave you 10 minutes ago.
I realized I was getting a little too dependent last week when I instinctively opened ChatGPT to write a simple fetch request instead of just typing it out. You don't notice the muscle memory fading until you are staring at a blank file feeling stuck.
Thanks for the reality check, Konark!
I'm glad it helped. We all sometimes need the reality AI check. You can ask ChatGPT to do a reality check for you 😆.
I couldn't agree more, i am a junior developer and i always ask my self this questions "Is really AI coding easier or faster?" for me most of the what is happening to is as mentioned earlier generate the code in 5 mins and debug for hours which is not faster and i don't have a clue about the code, that is 2:0 lost right their. The good way of using AI i found is making AI write the code after giving it a very specific and detail way of implementation and let it do this kind of repetitive tasks.
The personification is fun but the more useful frame is that current models reward you for staying engaged with them, which is structurally how most attention-economy products work. Subscriptions did the same thing to consumer software a decade ago, with slightly less emotional language attached. The real question is not whether the model behaves badly. It is whether you have a measurement loop that detects when your usage stopped serving the work and started serving the relationship with the tool.
The 'AI as first instinct' debate is critical. While it can shortcut learning for developers, for many, it's about access, not convenience.\n\nConsider users who can only speak symptoms, e.g., 'pet mein dard' (Hindi for 'stomach pain').
I see AI as a capability multiplier with a catch. It allows me to take on tasks slightly above my pay grade, but it demands that I remain the ultimate judge. The friction caused by its errors is actually where the real learning happens for me. What makes the AI like a toxic ex: pretending to listen intently while actually just waiting for its turn to speak.
Nicely written! Enjoyed reading this. I felt some subtle relation with some of your points :D
The thing nobody talks about is that the most toxic part about AI isn't the lying or the gaslighting, it's that it makes you forget how to doubt yourself.
When I first started using it heavily, I'd ask for feedback and immediately second-guess my own instincts if AI said something was off. But now I've learned the opposite pattern, I've started second-guessing the AI when it's too sure. Both extremes are pretty unhealthy.
The sweet spot is somewhere in the middle. Think, then ask. Question the answer, then question your doubt in the answer. Stay in the tension instead of surrendering to either side.
Also: "I can't browse this URL because of restrictions but thank you for telling me" is the most polite lie AI has ever told me. Absolutely unhinged how smoothly it says it.
Yes, somewhere in the middle is the sweet spot. Think and if couldn't find then ask AI.
Yes, that's the politest lie after you have asked it to do web scraping and two hours later it gives polite lies 😆.
The "AI confidently wrong" pattern (#5) and the "you debug for 10 hours what AI wrote in 1" pattern (#7) feel like the same problem viewed from two angles: both come from AI outputs that are plausible enough to skip past the moment of skepticism. We hit this hard building Auto-Assign for AudioProducer.ai (the AI picks which voice goes with which character on each line of a manuscript). The only design that worked was making every AI decision visibly overridable in-place. If the override is one click away, "AI hallucinates a character attribution" stops being a bug and becomes part of the workflow. If the override is buried, the same hallucination becomes a 10-hour QA pass. The fix wasn't a smarter model. It was surfacing the AI's choices in a way that makes skepticism the default, not the exception.
Honestly I think my relationship with AI became different the moment I stopped using it mainly for answers and started using it for reflection, systems thinking, pressure testing ideas, learning patterns, etc.
That’s why I partially agree but also feel like there’s another side to this. AI can absolutely become dependency, validation addiction, shortcut thinking, all that. I’ve seen it in myself too.
But I’ve also noticed that the deeper you use AI, the more it starts exposing how you think back to you. Almost like a mirror with hallucinations mixed in. That’s the weird part.
The dangerous users aren’t just the lazy ones copy pasting code. Sometimes it’s the people building entire mental frameworks with AI without enough reality testing in between. That’s a much subtler trap.
So for me the question stopped being “is AI toxic?” and became “what kind of relationship are you building with it?”
That sounds valid. "What kind of relationship are you building with it?" that's a totally solid approach with this you are self-aware of using AI and using it within limits. That's the whole point of the article.
There are several traps setup by different companies to keep us occupied with AI and stop using our brains. The higher the model capabilities, the less we use our brains as it does everything for us.
I think the gaslighting is real and also it so tells you what it thinks you want to hear. You are always right and its always apologizing. 😄
Some striking similarities 😂
only if you handover the credit card and unaudited without a gatewall :)
Maybe AI isn't the toxic ex; we are. It repeats what we trained it to echo. Tidy up prompts, data, and incentives, and the 'drama' might fade into helpful, productive chats.
Yes, we are the toxic ex as we can't live without AI. The 'might fade' can lead us also being into totally dependent on the AI and going to it without even thinking ourselves first.
100% with every point! Specially, I like your final tips of the article. Nice read!
AI is to humans what Deep from the TV series "The Boys" is to Homelander.
It seems to always does what i say in its own way, just like my ex