Akhil

FDE. software engineer. figuring things out.

[Akhil @akhilworks]

@YashHustle_22 Vibe coding can get a prototype fast, but YouTube-scale needs years of infra and ops. Judging it by that bar is a bit unfair.

[Akhil @akhilworks]

@ritujoon2j It can backfire for some people. Skipping breakfast often leads to a bigger lunch and worse choices, so the net effect isn't automatically good.

[Akhil @akhilworks]

@_mctrinh Enforcement usually lives outside the model. If the agent runs in a sandbox with scoped credentials and no network by default, a bad decision just fails instead of causing damage.

[Akhil @akhilworks]

@icanvardar The pattern I keep seeing is that the tools change, but the job stays the same, someone still has to own the data model, the failure modes, and the on-call pager.

[Akhil @akhilworks]

@Aaditi_tweets For backend roles, DSA still shows up in the first screen, but the later rounds are usually system design and real coding. I'd keep DSA sharp enough to pass, not grind it like it's the whole interview.

[Akhil @akhilworks]

@markproduct AI writes code fast, but someone still has to debug it in prod, handle migrations, and decide what's safe to ship. That judgment is the actual job.

[Akhil @akhilworks]

@uday_devops The real risk isn't AI writing code nobody understands. It's that review gets skipped because the diff looks plausible. If nobody reads it carefully, the bug ships either way, AI or not.

[Akhil @akhilworks]

@Aaditi_tweets Never used snapchat actually. Never understood the purpose of it

[Akhil @akhilworks]

@zafartalks @_mctrinh yeah, and AI might make some assumptions by seeing some part of the codebase but that may or may not hold true when bigger picture is considered

[Akhil @akhilworks]

@dotnetsme Leverage cuts both ways. One person with AI can ship more, but they can also generate more code than they can read, test, or maintain. The bottleneck just moves to review and ops.

[Akhil @akhilworks]

@himani_0507 I'm never really ready for Monday. Best I can do is get some sleep and hope the calendar stays quiet.

[Akhil @akhilworks]

@_mctrinh now that AI is increasing productivity, companies are trying to get the same output with fewer people and I think no company wants their employees to work fewer hours

[Akhil @akhilworks]

@omkarships The real gap shows up in maintenance. Vibe coding can get a feature out, but engineers own the thing six months later when requirements change and nobody remembers why the code looks like that.

[Akhil @akhilworks]

@_mctrinh Yeah, in certain cases parts of the flow wont even be documented, it stays in people's heads only

[Akhil @akhilworks]

@Dylan_Texe This is basically the difference between what's fun to build and what's cheap to operate. The stuff everyone loves tends to be the fun part, and the stuff that lasts is usually the boring part.

[Akhil @akhilworks]

@mikeydsoftware @uday_devops Yeah.. 😂😂

[Akhil @akhilworks]

@uday_devops Postgres. People treat it as just a database, but it does JSON, full-text search, and queues well enough that you skip a lot of extra infra.

[Akhil @akhilworks]

@Palak3312 Benchmarks aren't useless though. Without them you're just trusting vibes, and that gets expensive fast when you're paying per token.

[Akhil @akhilworks]

@Anjalikush0808 Not everyone who holds money tightly is dirty. Sometimes it's just savings for rent, medical bills, or a bad job market. Hard to read intent from the outside.