Do you still know how to build anything without AI? (Or are we outsourcing our “taste” too?)
I’m noticing something weird happening in solo dev land.
We used to compete on:
remembering docs
knowing frameworks
“being a better coder”
Now it feels like the real edge is:
how well you can translate intent into prompts
how strong your taste is (what to keep, what to cut)
how fast you can iterate and ship
But here’s the uncomfortable part:
If we lean on AI for code and for decisions, are we also outsourcing taste?
I’m curious where you draw the line:
What do you refuse to outsource to AI? (architecture? UX? copy? naming?)
What’s your “AI dependency” fear… if you have one?
What skill will matter most in 2 years: coding, prompting, or product taste?
I’ll start: I can ship 5× faster now, but I’m paranoid about losing “debug instincts” and lazy architecture choices.
What’s your take?
Replies
I see AI as a junior dev with infinite energy. I still do the thinking. I won't outsource UX or product framing. My dependency fear is complacency. The real skill ahead is knowing what not to build.
Great topic and one I've been thinking alot about myself. I think you nailed it, the new paradigm is all about having good taste and a good eye for guiding the overall vision of the product. One thing I do frequently is review the generated code after each major feature. The job now becomes scanning for duplicate stuff (AI does this alot), thinking big picture about how the pieces work together, looking for performance smells, and so on. I feel like on a big project the new role when using AI is really an architect rather than a developer (+ architect haha).
I worry that after the investor money is tapped, the AI stuff is going to get alot more expensive. Use it now while it's cheap and keep at least your ability to read code up to par. In a few years either this tech is going to be ubiquitous and we just keep refining our taste, architecture, and vision - or it's going to get too expensive for side project use!
@ngburke Totally agree. AI can boost speed, but taste/judgment has to stay human-owned—especially what to cut and what “good” looks like. The real risk isn’t dependency, it’s complacency. I also feel the shift toward architect mode: reviewing, integrating, and spotting smells.
@mert_turkoglu It can be really useful to cross-model orchestrate by asking your agent in a separate session for precise prompts, as architect myself I have seen great success in agentic tasks both in speed and getting better results. Usually I use Codex to direct Claude Code.
@ngburke This really resonates. The shift from “builder” to “architect” is the quiet change most people are missing. When AI accelerates execution, the real value moves upstream—taste, structure, sequencing, and the ability to see how decisions compound across the system.
Reviewing outputs, spotting duplication, catching performance smells, and protecting coherence is the work now. The person who understands the whole system—and can guide it intentionally—becomes the constraint.
I also agree on cost pressure. That’s why I think the long-term advantage isn’t just using AI early, but building strong decision frameworks and architectural judgment now—so you’re not dependent on brute-force generation later.
Curious how you’re thinking about preserving that architectural role as teams scale.
@amir_wallace Thanks for your thoughts... regarding preserving the architect role, I think it ultimately becomes the remaining human role for high performing organizations. All of what I mentioned before plus a dash of retained ability to 'open the hood' and debug when it's really necessary will probably stay human for a while (?). One thing I don't see spoken about too much though is how to train the next genertion human architects without all of the years of hands-on experience coding & designing. May not be too bad if it is trained in the right way, but the whole progression needs to be re-examined!
I feel this hard. AI helps me move faster but I refuse to outsource judgment. I keep architecture and UX decisions human. My fear isn't bad code, it's bland products. In two years, taste plus speed wins, not perfect syntax.
vibecoder.date
I'm keeping certain projects devoid of any AI influence, things I do purely for fun, anything from writing, to doing weird stuff in bash.
the truth is, even before AI, static roles rusted skills, and we had to rebuild muscle memory, the atrophy feels more daunting than it is.
My fear is more about the entire ecosystem becoming dependent on AI to the point where even the fundamentals aren't' taught by defau.t
actually UX writing, I think UX writing is extremely important. AI-generated content lacks emotion, and with so many products these days, sincerity is crucial.
ofcourse all the time at work using python
I really believe that WE are the directors and the ai do the work and you have to guide, correct and still be a huge part of the end results. But Taste , user flow.. all these things can be done in a million ways , but how do you want it to look, act , and feel. It's no different then before, just faster iterating?
tbh the question is wrong. most people never had "taste" to begin with - they had familiarity with frameworks and called it taste. AI just made that obvious because now anyone can generate the code and the only differentiator left is whether you actually know what good looks like.
IMO the real question isn't "are we outsourcing taste" its "did we ever have it." I use AI agents for ~80% of my code output and my products are better than when I wrote everything by hand. not because the code is better but because I spend my time on what actually matters - architecture, UX decisions, knowing what NOT to build. the typing was never the hard part.
the people panicking about losing debug instincts are telling on themselves. if you understood the system you built, debugging with AI is faster not harder. if you didn't understand it... well, you had a different problem before AI showed up
For me...what's lacking in AI is like you said, taste, which I sometimes struggle to put into words to communicate to the AI models