HN Explorer

In the era of "Vibe Coding", when Agents are writing code – what are you doing?

9 points by madagang ↗ HN
In the vibe coding era, AI agents are busy generating functions, wiring tests, and filling in boilerplate.

I catch myself just sitting there, scrolling on my phone — and it feels like wasted time.

So I wonder: what’s the right human role in this loop? Higher-level design? System thinking? Critical auditing? Or are we destined to be “idle supervisors” while machines do the typing?

How are you using those gaps when the Agent is writing code?

9 comments

[ 0.22 ms ] story [ 26.6 ms ] thread
smoovb
When taking a break from tending my agents, I like to eat lunch, some exercise or check on other work and emails.
adyashakti
maybe it can write code, but it still sucks at technical editing
Leftium
You can fill the gaps with more "Vibe Coding."

While one agent is busy, you can start another agent on a different task. (Or maybe even the same task with a variation in the prompt. To hedge against the unlikeliness of the first agent always one-shotting acceptable code.)

Even if you aren't verifying the resulting code, prompting six different agents and confirming their results should keep you busy. (I guess you could even do more if you wanted...)

I have seen various descriptions of this type of workflow. Most recently: https://hw.leftium.com/#/item/45180353

> Now I see him spending most of his time doing what product managers traditionally do: talking to users, understanding their problems deeply, figuring out what's actually worth building. Coding has become maybe 20% of his job, and even that 20% is mostly about understanding requirements and translating them into clear specifications. The actual implementation work that used to consume 80% of his time is now handled by machines.

Also here is a short video of this type of workflow in action: https://www.youtube.com/shorts/tP1fuFpJt7g

giantg2
There isn't any real gap time. It tends to generate things quickly, then you have to code review it all.
the__alchemist
I'm solving problems that LLMs assist with, but are unreliable. Using vibe coding or Agents sounds like a disaster.
cdaringe
Imagine a job where all you had to do was take requirements and punch out code.

While i expect that to be the case for junior dev positions, it very rapidly is not the case for even entry senior developer roles.

To answer your question, I’m busy solving non coding problems.

mikeinseattle
Thinking deeply about architecture and trying more high-level prototypes out -> then if one really sticks digging in a bit more and injecting my own code is what I've started to do.

The downside is I don't really like starting from "scratch" without these tools anymore. They save so much time for drudge work / migrating changes across large portions of code - I miss the manual way but if I'm being honest not really.

I code as much as I can without ai tools in order to stay fresh for interviews though. It's really eye opening to see how lost non-technical people without a software background get even with the latest cursor tools.

In the future, architectural understanding and knowing how to scale a project will pay top dollar.