From Vibe Coding to Vibe <…’ing>
My first hand experience of Vibe <...'ing>
“I’m not writing code; everyone is just typing prompts into Claude Code.” I heard of this from a Databricks intern yesterday during a dinner gathering.
I nodded.
Vibe Coding
I am using Cursor and Claude Code at the same time. For large refactoring, I let Claude Code handle it. For small changes with IDE view, I let Cursor handle it. And I heavily rely on Cursorbot in github to review pull requests.
Where I still hold the bar, is to make the key design choice, and to give feedback why a solution doesn’t work well enough.
Is this vibe coding? Close, but not the same.
Vibe coding is good for prototyping, but for production deployment, not there yet. (Maybe it will be ready in a year? Who knows. I do know the coding agent ability was not that great just 2 years ago, but things have improved drastically.)
However, what about other tech functions?
Vibe Designing
This is what we tried first. We already had an app design, and we needed to add a new feature. We didn’t need pixel perfect design, so we used Claude Cowork. We fed it existing designs, and our desired new feature, and asked Claude Cowork to mock up a design that fits the existing app.
It did okay.
So we added more. And it did okay each time.
As time goes, one day our senior design lead asked: “the app design is so inconsistent; want a holistic redesign?” We looked at each other, and agreed.
The new design is leaps and bounds better.
Key takeaway, overall design thought leadership is not replaced by AI.
Vibe PM’ing
I’m also serving the PM role for my team at the moment. I enjoy product thought process, I enjoy product brainstorming. I just don’t like writing PRDs. I thought, great, let AI do that!
For one specific feature, I asked Gemini to create a Google Doc from scratch. It did well in forming the background context, and discussing the tradeoffs.
But the formatting was so inconsistent. I ended up spending a lot of time fixing the formatting so that the doc is somewhat easy to digest. I’m not sure that actually saved my time.
In the end I believe no one actually read the PRD. Everyone just relied on me giving the verbal answers.
Did AI add value? Yes. It helped me do research, and made good suggestions on some aspects. It’s been very helpful in reaching the high level product idea.
However, it also stopped there. The PRD’s existence only proves that we have thought about it and this idea wasn’t completely random. But no one can actually rely on the doc for the decisions.
Why? I found the reason very interesting. It’s because no one actually wrote this PRD, no one felt responsible to update it when things changed. (Translation: I didn’t personally write the doc, which led to me not feeling responsible to update the doc as things evolved.) Very quickly, the doc became outdated.
Key takeaway, overall product thought leadership is not replaced by AI. And in this case, even the tactical task of keeping a doc updated to reflect the latest decision isn’t achieved.
Vibe <...’ing>
There are other functions that we are vibing too, but the story gets similar.
The consistent trend: overall thought leadership is not replaced by AI. AI provides tactical values, but runs short in strategic value at the moment.
That being said, this doesn’t deter the hype being there. The belief is, give it some time, it will be good to replace humans.
Will that happen? I’m torn. I’d love to see a day when these things get easier, and AI helps me focus on the core tasks of building my product. But at the same time, I can’t clearly see any time soon that I can relax without providing judgment.
Should this feeling be called Vibe Tearing?