Varun

Coming back to X after a while, been busy building. Now focused on AI, RAG, and Agentic systems. Plan is simple: learn in public, share what I build, no pretending. Looking forward to connecting with people who actually think & build 🤝

I have been thinking about on device inference lately. If models get good enough locally, would you still rely on cloud APIs, or prefer running things on your own device even if it means a trade off in performance or quality of outputs?

@bendee983 It feels like this could shift over time, especially if the models keep getting better. But in the end, adoption will depend on convenience and the ecosystem around it.

I feel like all those arguments about users not caring to run local LLMs will collapse sometime before 2027. On-device AI is becoming very good.

@cgtwts So the workflow is like, ask Claude to make Claude better?

Anthropic CEO: “At Anthropic, we essentially have Claude designing the next version of Claude itself, not completely but most of it”.

@CarawayDJ @Adidotdev Makes sense, you start valuing stability more over time.

True, but I still resist. At the first software company I worked for the CEO was a complete cynic about anything new. He basically believed most of it was just a rehash of something old with lipstick and new marketing. I rolled my eyes. He loathed switching to new toolkits or chasing new technologies. He valued stability and predictability. I rolled my eyes. I am now him lol. I get it now.

No cap, which IDE is actually carrying developers in 2026?

@CarawayDJ @Adidotdev Hard to blame them, a lot of these shiny objects eventually become the standard.

@vij_03 @Adidotdev Developers have been chasing trends for years and years. They are always chasing the latest shiny object, especially the younger devs. Find something that works well for you and stick with it.

@anandrishv @buildwithsid NASA is already using it🤡

@buildwithsid primary language would be sanskrit, proud

What if India built VS Code?

@seanonolennon Bound to happen when people start seeing it as an answer to everything rather than just a tool for convenience.

AI is starting to feel like a modern religion. And some are willing to sacrifice everything on its alter.

@bendee983 This is a great way to put it and exactly why understanding the output still matters.

Compilers are deterministic. Give them the same code with the same compiler settings, and you'll always receive the same binary. You can take responsibility for your software at the code level. LLMs, on the other hand, are stochastic. Even if you set the temperature to zero, you're likely to get different responses on the same prompt. Therefore, you need to understand the code it produces if you want to take ownership and responsibility for it.

@xoaanya New update drops and suddenly everyone switches sides.

who knew chatGPT would be the first AI to lose its job because of an AI.

@xoaanya @real_akireev They don’t really have another option 😂

@real_akireev People still use it tho

Curious, how many of you are actually paying for tools like Claude, Codex, or any other tools available or just squeezing out every single token from free tier plans?

@ta2025 @CodeEdison It is, until it keeps changing the same line again and again and you are just watching your tokens disappear.

@vij_03 @CodeEdison AI is also very good at debugging.

Is coding still worth learning in the AI era?

@omarsar0 Sounds great, but I have one doubt, how do you even debug a system like this? If everything is in latent space, we don’t really see or understand whats happening.

// Recursive Multi-Agent Systems // Great read for the weekend. (bookmark it) Multi-agent systems often pass full text messages between agents at every step. This leads to token bloat, latency, and context dilution which all grow with the number of agents. RecursiveMAS asks a different question: what if agents collaborated through recursive computation in a shared latent space, instead of through text? A multi-agent system can be treated as a recursive computation, where each agent acts like an RLM layer, iteratively passing latent representations to the next and forming a looped interaction process. They introduce a RecursiveLink module that generates latent thoughts and transfers state directly between heterogeneous agents, plus an inner-outer loop learning algorithm with shared gradient-based credit assignment across the team. Think of it as agents passing notes in their own internal language instead of rewriting everything in English each turn. Less talking, more thinking. The numbers are strong. Across 9 benchmarks spanning math, science, medicine, search, and code generation: 8.3% average accuracy gain over baselines, 1.2×–2.4× end-to-end inference speedup, and 34.6%–75.6% reduction in token usage. Why does it matter? If agent-to-agent communication is the next real bottleneck (and it is), latent-space recursion is one of the cleaner ways to scale collaboration without paying a token tax for every coordination step. Paper: arxiv.org/abs/2604.25917 Learn to build effective AI agents in our academy: academy.dair.ai

@Tekeee Gamble☠️

I just made $6,000 today I’m 21 years old I work out twice a day what do I even do now

Seeing a lot of mixed opinions on MCP. Some call out production concerns, while others seem to be moving ahead with it anyway. What are your thoughts?

@NitinthisSide_ Never got a call, but those forms are always long as hell.

Have you ever received a call for a job you applied to on Workday ?

@kashishonchain Feels like some founders forget that interviews go both ways.

INDIAN FOUNDERS NEED TO HEAR THIS 👀 I went on an intro call with the founder AI for a growth intern role. He asked me about my work, background, and experience. I explained everything clearly. Then I asked him basic questions about the company: what the product vision is, who the target audience is, what kind of users they are building for, and what direction they want to take. Instead of explaining the company himself, he asked me to explain it. I had read about the company before the call, but this was still supposed to be an intro call. I told him what I understood, that they are building an AI agent. Then he asked me the name of the agent. I had read it, but it genuinely slipped my mind in that moment. I told him honestly that I could not remember it right then. His response was rude. He said I should have known that, quickly said he would get back to me, and cut the call. Honestly, I found that very strange. If someone is applying as an intern and asks basic questions about your vision, product, target audience, and company direction, the founder should be able to explain it with humility. You cannot expect every applicant, especially for an intern role, to come in like they have done 100 percent deep research on your company. And if you are creating an AI research lab thinking you are some big shot, let me break the egoistic world you are living in: there are companies with 10x more funding already doing it better. If your ego gets hurt because someone forgot the name of your AI agent, maybe the problem is not the applicant. A founder should be able to communicate the vision clearly. That is literally part of leadership. And if you are building in AI with real humility, real product vision, and need someone on the growth side, my DMs are open.

@Nithya_Shrii This is more of a usage problem than a tool problem. Don’t just stop thinking and accept the ans gpt provides.

ChatGPT is rotting your brain and killing your critical thinking skills.