Himanshu Vinchhi

Software Developer 👨‍💻

[Himanshu Vinchhi @Himvins]

@badlogicgames @badlogicgames thank you for all the recommended post. Learnt a lot.

[Himanshu Vinchhi @Himvins]

Hard truth: AI won't make a bad developer good. It makes a good developer a 10x developer.

[Himanshu Vinchhi @Himvins]

Personally using AmpCode free, and GitHub Copilot regularly at work.

The harness/runtime speed difference is very noticeable.

Amp is crazy fast. Massive kudos to the Amp team - seriously impressive execution. @sqs

[Himanshu Vinchhi @Himvins]

Not because I code faster. But because coding is no longer the bottleneck.

Thinking is.

Feels like the shift from writing software

to designing systems.

And once you experience this, you can't go back.

[Himanshu Vinchhi @Himvins]

Before Al, I used to jump into code early. The design would evolve along the way.

Now, I spend more time connecting dots, defining boundaries, and thinking in systems.

And when I finally build-

it often comes together in one shot.

[Himanshu Vinchhi @Himvins]

@burkov prompting might become difficult at some point, no?

[Himanshu Vinchhi @Himvins]

Here is the snap

[Himanshu Vinchhi @Himvins]

Just saw Claude by Anthropic being advertised inside Amp 😄

Competitors? Partners? Frenemies?

Feels like the perfect snapshot of modern dev tooling — models want distribution, editors want the best UX.

Fun times.

cc @ampcode @sqs

[hårkl @harkl_]

Hard disagree

Tailwind isn’t struggling because people stopped reading docs or buying plugins

It’s struggling because AI collapsed feature scarcity.

Any competent dev can now build a plugin in <15 min

But AI didn’t kill the things that actually get budget approved:

• Risk transfer
• Time certainty
• Compliance
• Trust
• Deep integration
• Ongoing ownership
• Network effects
• Distribution

Code is cheap
Guarantees aren’t

If you want to monetise OSS in 2026, stop asking “what feature can we sell?”
Start asking “what responsibility can we own?”

Then make the model do the rest and iterate 🚀

[Himanshu Vinchhi @Himvins]

@maddyb65 @championswimmer I had the same doubt, tried it with this prompt and it did work

"Fix this math problem and generate updated image in my handwriting"

[Himanshu Vinchhi @Himvins]

@designcoursecom Many times those minor technical cues in the prompt improve code genertion a lot.

Also, you promptly knows when it is going in wrong direction and get it back to the track with tech hints.

I agree with you 💯

[Himanshu Vinchhi @Himvins]

@housecor One Interesting thing is it has so straight forward prompt to do the job.

Huge prompts are not needed to get to something special. With LLMs, less is more.

[Himanshu Vinchhi @Himvins]

@housecor Refactoring a component (or any file) once it reaches 500 to 600 line of code. once the component grow big, LLMs output quality reduces.

Current coding agents are anyway good enough to findout refactored components when needed

[Himanshu Vinchhi @Himvins]

Many of us are feeling the same way.

[GosuCoder @GosuCoder]

AI coding works better when you know the inner workings of what you are building… frameworks + language. I’m a noob at Rust for example and I find it painful to really guide AI effectively when working with Rust. While Node,C++,Java etc I feel like it gives me super powers.

How many of you feel the same way?

[Himanshu Vinchhi @Himvins]

@GeoffreyHuntley I have experienced this, I feel this new approach to the way we started working can lead to some sort of burn out at some point?

I mean some days are like my mind dont want to stop thinking as things are getting executed so fast.

[Himanshu Vinchhi @Himvins]

If you write code,

you can write your slides too.

Just use reveal.js -

it's Markdown + HTML + a little CSS.

No PowerPoint.

No drag and drop.

Just versioned, beautiful, dev-friendly slides.

Give it a try → https://t.co/LohCPtML5a

[Himanshu Vinchhi @Himvins]

@Tobihbest Gradually that is going to happen. Thats where it becomes scary. No one actually knows which kind of world it would lead us to.

Insightful podcast I watched around this topic recently - https://t.co/6pXu4g0qOp

[Himanshu Vinchhi @Himvins]

NLP was always hard.

Regexes, custom rules, domain models - a nightmare.

LLMs didn't just "understand language." They made NLP look easy.

That's why they're famous.

They turned one of the hardest things in tech into a prompt.