Aravind Putrevu

Aravind Putrevu

@aravind.dev

Dev Tools | Builder | Angel Investor Ex - Fireworks, SurrealDB, Elasticsearch aravind.dev

1.8K followers19K following98 posts

I agree that PR Review is an educational experience. Only when it is not used for blocking or doing silly politics (like teaching an unnecessary lesson on the history of programming) It should be fun. Coming to this question about testing in the interview - you are being tested. Surprised?

Yes, all the team fit rounds, HR rounds, those technical problem explanation questions (where they ask a silly question and we fail, but we get moved to the next round)

Most Developers are naturally introverted, except a few who write personal blogs and post on hacker news :P PR Reviews is a situation where they have to speak and explain, but it is also a harder soft skill.

I hate it when good products suck and slowly become obsolete. I have seen this happen with Databases and Dev Tools. And, now, with AI, the SaaS apps, which were cool yesterday, are lagging behind.

View full thread

I have this feeling: Products lose traction when they stop innovating and stop working on the mission that they started with. - Organize the world's information: Google - Make work life simpler, more pleasant and more productive - Slack

I hate it when good products suck and slowly become obsolete. I have seen this happen with Databases and Dev Tools. And, now, with AI, the SaaS apps, which were cool yesterday, are lagging behind.

Let us take a look at Slack. Coming from IRC and the mailing list world, I instantly loved it. Infact, I built a massive community on Slack for Elasticsearch, despite knowing it is a paid-only tool. Today, with AI, Slack search is poor, and the AI summarization is terrible, let alone workflows.

I asked Claude 3.7 + GitHub MCP to do one thing. Very specifically, one thing. It did 3. I understand the line of thought, and it uses the chat context to make decisions autonomously. But that is not what I want it to do!

View full thread

I guess the emergent behaviors with LLMs (ofc configured by LLM providers) are making and executing decisions. --- This is where we and AGI-ready LLMs probably diverge.

They think they know, while they might not. It is like an infant trying to catch the orange-ish charcoal, thinking it is a color that doesn't harm.

I asked Claude 3.7 + GitHub MCP to do one thing. Very specifically, one thing. It did 3. I understand the line of thought, and it uses the chat context to make decisions autonomously. But that is not what I want it to do!

I used DeepResearch 100 times last month 🙈 Why a lot many don't understand how cool is DeepResearch. Discovered it first on Google Gemini.. never stopped using it after introduced on ChatGPT.

The beautiful thing about LLMs that differentiates from working with humans - When I ask it something like "generate x. Here are a few examples," it delivers alpha that average or below-average workers can't. If the task is too complex, you need more examples, and it will deliver results.

For an average human being, ramping up on a task might have more questions, and examples might not be enough to generate alpha. Basically, the "grasp" of the problem is slow. Here is the place where we lose, and LLMs would win.

That being said, LLMs, too, lose their charm if the extra examples don't give me results. For example, several times, I try to generate something that fails miserably. After three attempts, I turn to human help!

Suddenly while doing something, this window comes up on the Mac. Do you see this often? It's one of those unfortunate things apple built in the whole scheme!

Tried using Gemini 2.0 Flash over the last 4 days. - lack of clear documentation. Search on docs is more confusing. - Multiple SDKs for the same stuff - What works on AI studio, might not work for real on API. This is the company I adore. I loved for their tech throughout my teens 😕

The best engineers are obsessed with learning new things The great thing about software is there is always a new framework or trending LLM to learn If this sounds like you, don't give up on coding

The AI revolution is going to be even bigger than the industrial revolution. The pace of change is happening in months, not decades AI enables anyone to make a software business Be part of the change

No-code tools are in trouble ⚠️ Every no-code influencer is using Bolt or Lovable now Softr costs $59/month Bubble costs $32/month Do you think no-code tools can recover?

Hot take: Saying that you ace'd a AI benchmark leaderboard and then fail to get customers for your product is like Getting a Gold medal from Ivy League and not able to get any job.

The more new things you try in your career, the more you fail, the more you learn, your chances of getting lucky increases, and you start getting wins. Of course, you will not fit in any shallow pond of corporate soup. Your career is analysed and might look filled with a lot of experimentation.

I just now saw a killer demo from an AI Founder. Sadly, the only moat for them is the prompts + ux. Tons of AI Product launches all across, some very interesting ones. The only thing stopping big players at this moment to launch all solutions parallelly is TIME. Why?

Unlike money, infrastructure, and talent - time is inelastic. We all have 24hrs! People who execute better in that 24hrs are going to be the winners. A little loss in that productivity will be costly for them because big players have everything else on their side.

OSS Maintainers are in for a ride! I just got the first nicely worded PR spam. The person tricked me into reviewing the code. I thought it was a genuine contribution. Fun part: it came to a @coderabbitai repo.

View full thread

Personally, it’s not that writing docs is hard—it’s the psychological, organizational, and technical hurdles in fast-paced dev cycles that push documentation to the bottom of the backlog. “We’ll do it next sprint,” we say, knowing deep down we might never get around to it. The fix? start small.