Sabhya Verma

Senior frontend engineer. Using Claude Code and AI in real work. Sharing what works, what doesn’t and what I’m testing

[Sabhya Verma @SabhyaVerma]

@GergelyOrosz True, reviews have always been a learning method too coming from more experienced engineers, knowing their thought process on why things need changing if any

[Sabhya Verma @SabhyaVerma]

@rahulj51 A bit of both, the critical paths tested with automated UI tests so regressions are caught in CI pipelines, while browser-use very useful when finishing up a task and reviewing it before opening a PR

[Sabhya Verma @SabhyaVerma]

I made a manual review skill that goes through PRs with me, file by file.

It feels like a supercharged version of how I’d review alone. We catch things either of us would miss on our own.

It’s made a big difference to the quality of what I ship

[Gergely Orosz @GergelyOrosz]

If you’ve stopped doing human code reviews: how is it going?

If you’re still doing them: how are you able to keep up with the explosion of code + PRs, and how do you prioritise what to review “properly”?

[Sabhya Verma @SabhyaVerma]

@GergelyOrosz I still review anything non-trivial. Too often I catch bugs or readability/maintainability issues that would otherwise go through.

AI has helped with the workload though. I made a review skill that walks me through the PR file by file, rather than asking it to replace the review

[Sabhya Verma @SabhyaVerma]

@akshaymarch7 A lot of “technical disagreements” are really just people solving slightly different versions of the problem.

Getting the mental model aligned first saves so much time.

[Sabhya Verma @SabhyaVerma]

Good read!

[Addy Osmani @addyosmani]

How to maximize the value of your Claude Code sessions: https://t.co/YGa8swOX6v

Solid guidance on managing token efficiency and keeping context windows focused by @lydiahallie

[Sabhya Verma @SabhyaVerma]

Does seem impressive! 💯

[Sabhya Verma @SabhyaVerma]

1979:

“put that there”

2026:

“put that there, but first read CLAUDE.md”

[Sabhya Verma @SabhyaVerma]

@mattpocockuk @unclebobmartin It’s been a game changer, also let’s new contributors ask questions to the LLMs naturally and the cotext doc helps guide where to look, which feels like an underrated use case for it!

[Sabhya Verma @SabhyaVerma]

@MansiCodez Both have their merits but feature flags help merge often which has been quite useful ever since using them. Distribute work into chunks that are easy to review and merge them one by one and deploy behind a fearure flag even if incomplete. Then turn on in staging to test

[Sabhya Verma @SabhyaVerma]

@housecor A PR template could be a useful addition too

[Sabhya Verma @SabhyaVerma]

@housecor Reading code still seems important, more often than not, there are things to improve or fix when you start looking

[Sabhya Verma @SabhyaVerma]

What it actually looks for

[Sabhya Verma @SabhyaVerma]

That moment when the implementation is technically correct but somehow needs 4 files, 2 helpers and a context provider.

/simplify is basically Claude Code’s “can we not?” pass.

[Sabhya Verma @SabhyaVerma]

@aditi_tibarewal Nice and interesting site! Tiny iOS Safari polish: the input zooms on focus because the font size is under 16px. Setting it to 16px on mobile should make it go away

[Sabhya Verma @SabhyaVerma]

Interesting Claude Code update: /usage now shows prompt cache usage, misses, cache reads/writes and whether the cache is warm.

Feels like we’re slowly getting proper observability for coding agents.

Cache invalidation followed us here too 😅

[Emil Kowalski @emilkowalski]

The registration for my animation course is open for the next 10 days.

Now, with a new CSS Animations module!

We'll build all of these components and more, source code included.

[Sabhya Verma @SabhyaVerma]

@cheatcodeco So glad you found it helpful!

[Sabhya Verma @SabhyaVerma]

100dvh is quite handy! https://t.co/Mm36PTLsuJ

[Sabhya Verma @SabhyaVerma]

@JoshWComeau @KevinJPowell Hashnode is a pretty handy platform. Markdown supported blogs and also now supports a Headless API