In a comment some weeks ago (link) i wrote about how it doesn't matter if you are using AI or not, or if you call yourself a vibe coder or a SWE. O...
For further actions, you may consider blocking this person and/or reporting abuse
Thank you so much for this insightful and inspiring article! Your reminder to stay curious and take ownership of our own learning really resonated with me. I truly appreciate the effort you put into writing this. Keep up the great work!
🧊🗻😁
The "if you can't explain it simply, you don't know it" bit is exactly why I write things down now. I really got obsessed with documenting everything I do or develop rn. AI can make my output look finished way before my understanding is. The dangerous part isn't using it, it's losing the discomfort that makes you ask one more why. My rule lately: if I can't re-explain an answer in my own words, it's not done. AI should makes us smarter, not lazier
Same here! I was writing things in an actual paper way before, nowadays is a must when my memory and attention gets more and more divided
I actually found that I ask more questions with AI, than without it. I research more, I learn more and I still get more done. Not cuz I let AI do it all, but because when it finishes a round, if there's anything I dont understand, I ask it to elaborate, if I still dont understand, I send the AI on for it's next round, then research it manually. Something I find quite useful though, is key-word queries. Eg. Write a function in rust, then ask it 'how would we optimize this for zero-alloc', it'll then check allocations and give you a theory on how to reduce it and whether true zero-alloc is possible. Yes, it's not perfect, but it helps you find the small things that can be the difference between a server hitting 1gb of ram usage under 1000 concurrent users, or stay stable at 50mb. It makes a difference, because it compounds and I would have never asked that question if I was stuck figuring out getting it working, let alone optimizing.
In that sense, AI really has given us the opportunity to look broader and dig deeper, because often we dont have the right question to research it ourselves, vs AI gives us the keywords we need to learn specifics.
Hmm, yes, techinical questions are easy to answer by an agent. But there are some that needs collaboration, like "why we are building this in the first place" or finding a limitation before writing a single line of code, that i see people just give up to ask. I see more and more that design and user requirements are moving after the development, which make the pipeline WAY slower.
In many ways, yes. But in others they make the end task more deliberate, depending on how you use it. Eg. When investigating a concept, I ask alot of questions, regarding common use-cases, edge-cases, industry equivalents, what's users' biggest complaints, what makes it maintainable, where do these systems generally fail, how can we optimize it without sacrificing compatibility, etc. All questions that get answered immediately, whereas they'd take a decent while to research manually. It's that instant nature that makes me more willing to ask them, rather than weigh the tradeoffs and ignore them, because it would take too long.
It's like that old saying about chopping a tree, you spend more time sharpening your axe than chopping. Do the things that are low-effort first, so they can make the high-effort task more deliberate and easy to manage. Like sketching out a roadmap, architectural diagram, Red-teaming, etc. All things you can just prompt your way through, before writing a single line of code, because with those files in place, you have a solid design to ground the implementation with and it simplifies it enough that you'd understand the concept, before you ever code. It makes the process slower, but lets remember, 10 minutes of planning beats 10 hours of debugging.
Yes, i agree 100%, try to convince a manager though 😂 Some things will take time, no matter what
That's why I tend to lean towards higher autonomy and ownership roles. Give me a task, let me do it my way... Dont micro-manage. You hire for experience, so that experience can be applied right? But seems to me that the more people get into AI in their companies, the more paranoid they get, despite proven results.
The last couple years i worked in more corporate environments, and ownership even though advertised was never an option. Autonomy not a chance. Especially after the introduction of AI, we had this impossible requirement to do things in a ridiculous short amount of time, and most of these things were not even thought a bit. At that point i also lean to roles with more ownership and autonomy, I am a bit "scared" to go back to this insanity that bigger companies are
The tell for me is whether I can explain why the thing I shipped works, not just that it does. When I cannot, I usually end up building it again six months later because I could not adapt it.
The other one is how long I go without reading someone else's code. When everything I touch is either mine or generated, the only ideas in the codebase are the ones I already had.
True, but what happens when many many people ship to the codebase? Along with agents. That makes it increasingly harder to keep track!
On a team you cannot hold it in your head, so the test has to move off the person and onto the pull request. If nobody wrote down why a thing exists, somebody deletes it by accident six months later and finds out in production. Agents just make the pile grow quicker.
The biggest thing I feel is, it is not about curious - as you mature in the industry, most of us just choose what is worth (few are exception).
'What is worth' in what manner?
At least in my context: Even if I propose an approach or solution for a problem would that be supported by the team and company OR can I do it on my own - is it worth to pursue so the time/effort vs gain etc.., so many what if run in brain and will end up skipping most.
I understand that this holds about the actual work, but learning is not necessarily about efficiency or gains.
True, but as you get older, that is what it will end up with (for the most people, not everyone - agree)
If that was true... then Prasad and I would not be able to work, because Prasad V and I would not have all those simplified tools which were made by people who invested decades of their life into pursuing things that Prasad V and I cannot even comprehend still.
Humbleness is not a property of Prasad V and little old me, it seems.
We salarymen (like Prasad V and I) must be careful with our words, when we are talking about people like Naur, Knuth, Liskov and of course centuries of Eulers laying the foundations for us.
So in summary, I agree with you Prasad V on one point you made: We got older. We now have to buy more candles for our birthday cakes. Good on us.
Years did not make us more humble though.
How was it about respecting our elders?
We are not the good apples, Prasad V... we are not.
Being evaluated on token usage is the detail that explains most of the rest. If the metric is how much you used the tool, asking why something works is pure cost: it takes time and burns no tokens. Curiosity didn't fade on its own there, it was priced out. Writing things up is a good counter precisely because it's one of the few places left where not understanding something shows up immediately.