770 ⭐
SwiftDL has crossed 350 stars on Github⭐️
swiftdl.harshhh.me
github.com/harshsinghsv/swif…
3
106
v0.4.0 launched
github.com/harshsinghsv/swif…
swiftdl.harshhh.me/changelog…
SwiftDL has crossed 350 stars on Github⭐️
swiftdl.harshhh.me
github.com/harshsinghsv/swif…
1
5
298
Project got viral on chinese Twitter , dammm....
IDM 感觉被围剿了,Rust + Tauri 写的开源下载管理器,下载单个文件时最多开 32 条连接,某条连接提前完成后会接管剩余最大分段的一半,最终文件预先按大小占位、各连接直接写入,省去合并步骤。
github.com/harshsinghsv/swif…
6
374
SwiftDL has crossed 350 stars on Github⭐️
swiftdl.harshhh.me
github.com/harshsinghsv/swif…
Chrome ✅ Edge ✅ Brave ✅ Vivaldi ✅
Now in Firefox ✅
SwiftDL is a free, open-source download manager with 32 connections per file, crash-proof resume and YouTube support. No trial, no account, no telemetry.
Site - swiftdl.harshhh.me
Github - github.com/harshsinghsv/swif…
No trial. No account. No telemetry. No catch.
SwiftDL is a 5 MB open-source download manager with 32 connections per file, crash-proof resume, a browser extension and YouTube support. MIT licensed. PRs welcome.
Site - swiftdl.harshhh.me
Github - github.com/harshsinghsv/swif…
7
274
I also build open-source IDM in Rust
SwiftDL is a 5 MB open-source download manager with 32 connections per file, crash-proof resume, a browser extension and YouTube support. MIT licensed. PRs welcome.
Site - swiftdl.harshhh.me
Github - github.com/harshsinghsv/swif…
So somebody ACTUALLY built an open sourced Figma clone in Rust.
You can now open .fig files without any limits or anything.
6
385
Downloading Torrent is also supported now
Try it out 🙌
No trial. No account. No telemetry. No catch.
SwiftDL is a 5 MB open-source download manager with 32 connections per file, crash-proof resume, a browser extension and YouTube support. MIT licensed. PRs welcome.
Site - swiftdl.harshhh.me
Github - github.com/harshsinghsv/swif…
4
218
No trial. No account. No telemetry. No catch.
SwiftDL is a 5 MB open-source download manager with 32 connections per file, crash-proof resume, a browser extension and YouTube support. MIT licensed. PRs welcome.
Site - swiftdl.harshhh.me
Github - github.com/harshsinghsv/swif…
Harsh retweeted
Someone with your exact background, your exact resources, your exact 24 hours just did what you've been telling yourself you'll do next month. Same internet. Same tutorials. Same AI tools. The difference isn't privilege. It isn't talent. It isn't luck. At some point, you have to admit that the difference is what they did while you were thinking about doing it.
today while working on a project I realized that in the era where the most of the work is being done with the AI , have the absolute clarity with what are u doing , also have your fundamentals super clear
learn fast but implement slower
ss is of my audit 🥲
2
112
Harsh retweeted
You cannot optimize what you haven’t mastered manually. You haven’t earned the right to automate until you’ve felt the friction of the manual labor.
The AI Engineer Roadmap is updated now
Believe me you don't need anything more than this
ai-engineer.harshhh.me/
AI Engineer Roadmap v4 is live 🔥
What's new:
→ Harness Engineering
ai-engineer.harshhh.me
5
338
while reading this I realised that , I m hitting the sweet spot lately and it's fun and you actually know what you are doing
AI for coding has given me enough opportunities to actually think about the product rather than blindly writing code
How do you use AI for coding without becoming dependent on it?
The first answer is simple.
Let the agent do everything.
If you choose this, you will finish features much faster.
Large refactors become easy.
Boilerplate disappears.
You can build things that would have taken weeks in just a few days.
But this creates another problem.
You stop making hundreds of small decisions yourself.
The agent decides where to put files, how to structure functions, which abstractions to use, and how different parts connect together.
You review the diff, merge it, and move on.
After doing this for weeks, your understanding of the project slowly starts to disappear.
Not because the AI wrote bad code.
Because you never built the mental model yourself.
The second answer is to stop using AI completely.
If you choose this, you will understand every line of code because you wrote it yourself.
But now you lose all the boring productivity gains.
Writing tests.
Renaming variables.
Generating CRUD endpoints.
Simple migrations.
Documentation.
Repetitive refactoring.
These are things AI already does well.
Throwing all of that away does not make much sense either.
The third answer is to use AI like a junior developer.
If you choose this, you stay responsible for the architecture.
You decide the boundaries.
You decide the data flow.
You decide the interfaces.
AI only implements your decisions.
This is the sweet spot. let AI generate code, but never stop reviewing every change or asking why something exists.
If they cannot explain a piece of code, do not merge it.
Then another trade-off appears.
Reviewing code is not the same as writing code.
When you manually solve a problem, you naturally remember why every decision exists.
6 months later you may forget the details, but once you open the file, the reasoning comes back because you remember the process that got you there.
With AI-generated code, that memory often does not exist.
You understand what the code does today, but you do not remember why it ended up this way because you never fought through the implementation yourself.
This is not really an AI problem.
If you choose to blindly approve PRs, you would have the same problem with human developers.
The real issue is laziness.
AI just makes it much easier to become lazy because reviewing is always easier than creating.
Understanding every line is no longer realistic anyway.
Senior engineers already delegate work to teammates.
Why should delegating to AI be different?
Their point was that engineering is moving away from writing every line of code toward making architectural decisions, reviewing changes, defining quality standards, and building better testing systems.
Then another interesting point came up.
Instead of trying to remember everything, preserve the reasoning.
Keep architecture decision records.
Write better tests.
Document why something exists.
Make understanding part of the repository instead of depending on someone's memory.
That way AI can generate code while humans still keep the important context.
Another point that I found interesting was about company incentives.
Management simply does not care if developers understand every implementation detail anymore.
If AI lets the company ship faster with fewer engineers, they are willing to accept larger codebases, more technical debt, and weaker code comprehension.
They are betting that future AI will fix the mess later.
Whether that actually works is another question.
For me, the best answer is somewhere in the middle.
I want AI to remove repetitive work.
I want AI to help me read unfamiliar code.
I want AI to generate tests.
I even want AI to write code.
But I do not want AI making the important engineering decisions for me.
Because the biggest thing you lose is not typing speed.
It is your mental model of the system.
And once that is gone, every future decision becomes harder.
4
299
























