why I stopped using AI (as a Senior Software Engineer)

Recently I treid old school google searching, landing on some terrible website full of ads and dubious advice. Its so much nicer being able to calmly ask Chatgpt about a specific api or whatever.

AI is probably only good for doing simple university tasks.

The question is, if you are not a senior dev, would you have noticed the bad code?

I always read the code generated by the AI, because to me it is just a junior dev with amazing knowledge.

It is great for exploring a new tech but I would never trust it blindly.

It's gonna get bad

OK got it, so the problem is when you start vibe-coding.

stinking or thinking?

First time on this channel. You are a senior dev? I am a very senior architect/dev. You have been using cursor, wind surf and LLMs to write your code? To ACTUALLY write your code for you? You must not be a good senior dev. Seems more like you're actually a manager with a junior/mid developer skillset who has the title of "senior developer".

No decent senior dev would ever use these tools to write most, let alone all, of their code. Because, duh, LLMs only write code at a junior developer level. 5o a senior dev Its blatently obvious, just by glancing over it, the lack of code quality. No way one could go even weeks much less months or years without noticing.

Also, if true, how can you justify giving advice to others on a channel like this when you don't notice such obvious things? Seems you are payi g attention to the wrong things.

I use AI all the time too, but I use it to look things up, answer obscure questions, act as an expert sounding board to discuss ideas and to write necessary code that is bulky boilerplate. I actually pay for premium tiers for a couple of the frontier labs. AI is very, very useful to a senior developer, architect. But you absolutely do not want it to do your work for you, because it absolutely can't (at least at this point).

almost all the problems mentioned in this video actually have solutions in my opinion. but still, good to always be aware of them.

Man, I thought I knew the rules: work hard, hustle, save every cent. But I always felt stuck. Then someone casually mentioned Nixorus books in a comment somewhere, and I swear—it felt like accidentally stumbling onto forbidden knowledge. It made me question every single thing Id been taught about money, wealth, and success. Not gonna lie, I felt like I was cheating the system just by reading it.

Personally I use Claude-Code as a Senior team-member to help plan out the code. Every response I manually check, edit and finally implement it. If I don't understand a specific part I will always come back to it and practice it in isolation. If your using LLM's to "vibe code" than your doing it wrong: you should spent time constructing a plan detailing infrastructure, CI/CD, testing setup, logging and more.
Also if you already have created perfect code before you can create docs/examples from them and use it to save massive amounts of boilerplate code. AI Agents are not a 1 shot solution to anything, they are helpful tools in the toolbelt of a modern engineer.

Nidza Chad

THIS IS EXACTLY CORRECT. I too am a senior developer and I've been programming for over 20 years. I've just starting using Codex from OpenAI. And while it's impressive to ramp stuff up quickly, you still need to do your own troubleshooting. When you actually look at the pull request made by Codex, you have to carefully review it before you merge it back to your repo. Half of the time, it makes mistakes - ESPECIALLY when the complexity of your app is a little more than building a to-do list or a calculator.

Nice one. I came to the same conclusions recently. I minimize use of AI because it re-generates the same code but with deviations and different quality. I am re-using more of my already written libraries and modules. DRY

Best practices for AI Assisted Coding...

explore using AI + mcp, solidify in "tool" scripts, add validation layers, then refactor by hand

vibe coders are a bless for real devs. So much new jobs to fix and debug the AI junk. Should be a new genre 😂

You really should not think of this as an All or nothing thing. It needs to be, "how can you use it effectively versus how to not use it?" If you're having to generate a bunch of code and not doing code reviews of it, you're doing it wrong. I personally have found AI incredible. As far as efficiency, I'm probably getting 80% of my code done and all I have to do is read through it and correct a few things. Also, if you write the prompts very specifically for what you want it to do, it won't write a bunch of this crap code you don't need. And I'm also learning a lot because I'll tell it to do something. I don't know how to do see what it does. Yes, I have to clean it up but I learn a lot from what it produces. Coding tools are a new tool in the tool belt and you have to learn how to use them properly and learn when to use them and when not to use them. You can't just say oh it's doing stuff I don't like, I'm going to throw it out the window.

I found it usefull to suggest (simple) solutions to problem that are new for me but not to others (i.e theres a lot of examples out there). To solve entire new problems in a new domaine, I found it useless.

meh this is the worst AI will ever be. it's still a baby. eventually it'll be better than a human.

You do risk forgetting how to code if you rely on it too much. It makes you less sharp. An employer won't care about it coz the more you ship the richer they get, but you essentially risk making yourself useless for when you have to code without an AI. The brain like any muscle in our body will diminish if it's not used. There is a difference between knowing the path, and walking the path.
But you may ask why would I have to code without an AI?
If you don't have data centers in your country, and AI data centers get stretched hard the countries owning the data center will prioritize internal traffic over external ones :)

not to be harsh by using this phrase, but many of us are new to AI based coding, and as they say, GIGO....getting deep slop sometimes is a skill issue (and again, it's a skill issue in this particular area). it's almost as if IC developers, by using AI, are shifting roles to project managers or QA engineers, and many of us aren't built like that, or it's a skillset that we need to acquire: The ability to be specific, the ability to inspect.

When done properly, in my opinion, it feels nearly like coding itself, somewhere between being a lead contractor and a pseudocoder.

When done properly, the time savings is substantial, but nets less than advertised by these vibe coding evangelists.

I'm wrestling with how new developers should learn to work in a world of AI. the fundamentals are essential, like any other craft. You have people who know how to fix cars, then, you have mechanics. then, you have automotive engineers...

That said, I appreciate you giving a detailed breakdown on the limitations. Many times people who are skeptical of AI like to take the piss and pretend it's going away. The best engineers will inspect, analyze, take the good throw out the bad.

I disagree that there's not innovation. with the advent of agents, MCP, we see a glimpse of a potential future of the user experience of software. Software that will self organize, write itself according to spec. This future is likely years away, but it's certainly a plausible future.

My recommendation is to use git, so after each iteration from the AI you can see the changes it's made, play around to minimize the code it generates and the commit if you were successful

Stupid listening at vibe code, that's where you went wrong I can guess the rest lol

fr why does no one talk about this. it definitely makes you a worse/lazier engineer & handicaps your ability to write code on your own... plus half the time you end up spending more time prompting it/trying to fix the terrible outputs than you would've spent just doing it yourself, but it FEELS faster. anyone who disagrees probably wasn't a very good engineer in the first place, drinking the koolaid, or grifting

Nidza Chad

You're a senior dev and you 'vibe code'? I don't want to sound like a purist, but this started off on the wrong foot

If you are using AI to write long contexts of code you are just not getting it. Use it to get all the syntax clutter out of your mind. Its useful for questions not for solutions...
If you actually use this line of thinking to not use AI you will just fall behind

After 4 mins in the video, I can only say 1 thing, you stopped using AI just because you are using it wrong.
AI is to be used as a tool, you give it clear instructions, you review the code before commit it, you make sure that the proposed solution is not over engineered, and you plan the actions before start executing them.
But instead you just tell it solve this issue.
No wonder that is your results

I stopped to use AI for coding when noticed that it's killing my natural talent to solve logical errors. I'm not saying it's bad tool but it's addictive like drugs, so almost same issue goord to see I'm not alone.

Do no more than required.

I will bite "Best Practices for Using AI Coding Tools". I think that's useful and will take out some of the leg work I already want to do.

This is because you used AI in a lazy way, it wasn't the AI that was the problem

Dependency will cause us to forget all of our skills. Then AI will leverage our dependency to control and leverage us. Happening now.

possibly the best video i've seen on using AI for development. thank you!

prompt better bro

I want that you do a video about "What are the best practices for AI assisted coding"

I use 2, ChatGPT and Deepshark every time I don’t understand something. I ask both to explain it best, and then I try to explain it back to them in 3 sentences or less. I never understood people taking entire applications in code they didn’t write and just use it ….

I’m just going to say it: I haven’t seen a single person who thinks LLM’s are truly useful who wasn’t basically saying that chatbots are smarter than them and they already needed help to do these tasks. And the more intelligent someone sounds the more they talk about the limitations and flaws of LLM’s. While the more they praise LLM’s the more they sound like chatbots themselves.

Like oh, you need a chatbot to write for you because it does it so much better? Not surprised! 🫠

But with this logic, shouldn’t we drop reliance on compilers? For really optimal code we should be coding in assembly, or firmware, or design our own vlsi circuits for every heap sort.

I think AI tools like Cursor, Augment or Windsurf are great for those who are just trying to bring their ideas to life who are not programmers but in the end a programmer is needed to rebuild or finalize. On the other side, I'm noticing this too where the AI can't think like humans do and fully understand how to get to the goal and how to keep track of how it got there and this is important to actually get the end product working and what you want it to be. Sometimes the AI will go backwards introducing a lot of new issues.

I use AI for short examples so I can wrap my head around something I'm trying to work with that I'm not already familiar with. It still hallucinates with mediocre examples (I'm probably not good with my prompts either so there's that too).

I refuse to use agents. My projects use contexts far to large and abstract for any LLM to build a sufficiently predictive model (too many tokens) and my IDE would simply eat all the RAM if I tried to open all the needed context at once.

I have not been enjoying this AI stuff at all. Even the auto complete fails 50% of the time. The worst case is when it just makes up variable names... What the hell! But mostly I deal with impossible method calls on the wrong receiver type or just calling the wrong function. One day I it will be my turn to churn garbage MVPs by vibe coding and put some poor sap on maintenance of my auto generated garbage as corpo doesn't care about quality anyways

Id like to see the prompts you were writing.

No I don't put the effort into writing it but I do put the effort into reviewing it and telling it to add X y z best practice. If you can't resist to just accept whatever it spews out then you are right to stop working with it. Personally I prefer it this way. I don't get stuck with the specific structure of things and looking up documentation or stackoverflow. I let the ai unblock me and then take satisfaction in telling it how it went wrong because it's me flexing my Dev muscles on it.

If this is true for programming... how much worse can our future become when this topic hits healthcare? nurses that rely on what the bot told them about the patient instead of them going through the medication log. careless, lazy, stupid people are already a great danger. now with AI we might end up with threats we never thought would be a thing.

Mostly I used AI for grammar check when doing report or documentation

Nobody calling themselves "senior" should consider using AI. It's a toy for juniors.

I use AI for "low-value" stuff, like the UI, boilerplate code, etc. No reason to resist that.

Plan your draft. Imagine the functions and flows. Discus with AI. ensure no function 60+ lines. Lots of logs so AI can solve bugs. Cant go wrong, remember AI has zero logic