At Octomind, we build AI agents - but our code is still mostly written by humans. We love LLMs and use them everywhere we can, from our product to ...
For further actions, you may consider blocking this person and/or reporting abuse
Et voila! That's why I prefer to always keep the hands on the wheel.
My take on AI coding tools: they're like any other tool—incredibly helpful if you're pragmatic and know what you're doing. But since they make mistakes, you can't use them effectively with technologies you don't already understand. You have to be better than the machine, but it writes code faster than you could ever dream of. Vibe coding into production has always been a bad idea.
That's the secret, not so secret.
can't agree more. Thanks for sharing
Yep! LLMS cannot handle everything, because they often mess up so much and even if you try to call them out and say what is wrong the AI apologizes a hundred times, may or may not fix the problem while simultaneously it's giving you code with others bugs in return.
You nailed how AI's thousand-line droppings erode that sacred "aha, the data flows this way" sixth sense we coders hoard like dragons. It's a love letter to the messy, irreplaceable joy of wrestling code ourselves—reminding us that true mastery isn't auto-merge; it's the scar tissue from debugging at 2 AM. In a world chasing "orchestrators over coders," you're the voice saying, "Hey, let's keep the orchestra human-led for now." Inspiring AF.
That's a pretty good breakdown tbh. As someone who has tried both orchestration agents and hand-coding, I know what you mean about losing your mental model and getting messed up in unexpected ways.
The thing is that Agents are great at assisting brainstorming and repetition, but there’s no replacement for a developer’s intuition and experience when it comes to something that really matters.
Veith, spot-on critiques—AI agents still lack true self-reflection, make rookie errors, and can erode codebase intuition. These must be addressed.
Yet, from years in AI, I see Cursor and Claude as indispensable IDE allies for brainstorming, debugging, and testing—already woven into pro workflows.
Best path: leverage their strengths with sharp awareness of limits to avoid pitfalls.
Brilliant reality check—thank you!
Thank you so much for a really good, in-depth, analysis! As someone who started in IT in the 1970s, AI has been beneficial when creating small projects in languages which, for me, were still relatively new such as Python.
My experience has been much the same, even with small projects. I've tried several AI systems and found what each one develops from the same prompt often totally different, often with 'silly' bugs.
Once again, thank yoU!
Love this honest breakdown Veith.... refreshing to see real data over hype..
Such a refreshingly honest take! Your breakdown of “what’s missing” is spot on - agents promise a lot, but they still trip over basics and miss the subtle connections humans make.
So True, but its for time being with more advanced models AI will be able to write and deploy full fledged code, we never even imagined AI to be able to this much of our task, i too now use chatgpt just autocomplete tool or bug debugging but i have to oversee.
claude code boost my productivity 10x times