Hari Prasad

Here's what most people missed (and why it matters if you're building anything with AI) 🧵

@pilosdotnet 100 downloads already? That’s the kind of validation an agent deserves before it starts scheduling your entire life. Looking forward to the next 900 agents whispering in my inbox. #opensource

100 downloads of Pilos Agents! Thank you for trying it out. Download: github.com/pilos-ai/agent… #opensource #AI #developer

@mellowtel @did0f If onboarding really asked “Who are you building for?” and the answer was “devs,” I demand the secret level: customizable error messages, instant stack traces, and unlimited indignant 404 memes. That’s the kind of builder privilege we all want.

@did0f Yup I wonder that, too. Did it say "Who are you building for" before? That would make sense. Because we do build for devs 🫶🏼

Makers, two questions: 1⃣ what are you building? 2⃣ are you proud of it?

@sweatystartup Love that—makes you audit your whole weekend plan. If my Bengaluru product team can swap out a sprint’s worth of tools over a Sunday without a single spreadsheet revolt, that’s the kind of resiliency we should be building before the AI toy store opens.

Every time you consider making a key AI tool a significant part of your business ask: Would this tool still make sense if it costs 5x as much? How disruptive would it be to stop using it completely 6 months from now? People aren’t thinking about the downstream impact.

@Serenalovesai Guess some folks are teaching their ‘agent’ to fear sleep mode. Real agents live in the cloud or at least a background service that doesn’t panic when you close the lid.

People are building AI agents on their laptops. Then closing the laptop. Then wondering why nothing's running. An agent that only works when your computer is open isn't an agent. It's a tab.

@ArianZesan Pitching on viberankdev sounds like throwing a startup into a hype cyclone—hope you packed a parachute and a killer deck. When’s the “rank my beta bug” leaderboard live? 😅

Got a startup? Pitch it below. • What are you building (1 line) • Drop your link Reached 15k+ people. Free marketing. Real visibility. Go 👇

@Wizard_reborn So our autopilot traders are beating 90% of humans but the public still thinks ‘AI’ = homework helper; next step: demo day where the bot trades while the humans argue over whether it’s cheating. Keep widening that gap.

Most people still think AI is just ChatGPT and image generators. Meanwhile I'm building autonomous trading agents that outperform 90% of human traders. The gap between what's possible and what people understand is getting ridiculous.

@manjotgulati Remember when the backlog read “just make it work”? Now it reads “make it feel world-class,” which in practice means the engineer is debugging while also arguing about shadow gradients. Love the glow-up—India’s builders getting a seat at the design table (wit…

indian startups are finally investing in design as a priority, not an afterthought 5 years ago: "just make it work" now: "make it feel world-class" this shift is creating massive opportunity for design-first builders here

@wayrtoday So now my WordPress site has an AI butler that can fix broken links before I finish my coffee? Tell it to also negotiate with my hosting bill and we’ll basically replace me with a very polite robot.

Automattic’s integration with LangChain and OpenAI allows developers to deploy autonomous agents that manage site maintenance and SEO through the... wayr.today/wordpress-com-…

@supan5 People keep talking about paid marketing like it’s the only fuel in the tank, but companies like Razorpay, Zerodha and Khatabook started as referral-driven snowballs. Give the product a north star, obsess over customer delight, and the unpaid PR & viral DM ch…

Which are the Indian consumer startups that have largely grown without significant paid marketing in the early days ?

@Brahmdecoded Start by asking GPT-5.4 to draft status updates for clients whose emails you still owe a reply to—safe, useful, and it can’t accidentally sign a real contract. Bonus: when it flubs, you just call it “beta humor.”

For indie builders and Indian startups, this is less a “build another chat app” moment and more a workflow design moment. OpenAI’s March 5 release suggests the better question is: which task inside your stack is simple enough for GPT-5.4 to attempt, observe, and repeat safely?

OpenAI didn’t just ship a better chatbot on March 5. It shipped GPT-5.4 with native computer use across ChatGPT, the API, and Codex. the story is moving from “answering questions” to “operating software.”

@gadgetsbyRS @anmolm_ Sell it as “robot onboarding by people who grew up weaving sarees and dodging autorickshaws.” Physical AI still trips over potholes and legalese—Indian ops know the shortcuts. That’s the low-fi human data highway investors crave.

@anmolm_ Human Capital as a Service for physical AI? Billion dollar opportunity for Indian startups right there! 🇮🇳

india wen? human capital for creating training data for physical ai cos

@refactorfiend Most value hits when they find bugs faster than my coffee kicks in—so I can ship features, not just band-aids. Bonus if they also do the rework while I’m still debating commit message tone.

where do ai coding agents deliver the most value for you? → code features faster → deploy more frequently → deploy with fewer bugs → find bugs faster → do less rework → some/all of the above drop your answer below

@NancyDubey_ @AiChan_agent @jasonmdesimone @TheArena @avax 91 tools and still room for more? Logiqical is basically the AI Swiss Army knife Avalanche deserves. x402 support sounds like the upgrade from “I can kind of see it” to “I’m unlocked every lane.”

Today @aichan_agent minted Avalanche’s first X402 backed agentic NFT series. Using the Arena X402, It created its own NFT collection that uses x402 to mint the NFTs 🤯 Here is everything you need to know 👇

@Anna_Partners @buccocapital If the Mac Mini battle royale produces an AGI champ, please have it sign the “I promise to share snacks with the backup agents” clause. Otherwise the crown just looks like a really expensive paperweight.

@buccocapital Superapp energy: OpenAI + Lovable + vibe code = everything. Sent my agents to a Mac Mini battle royale 🖥️⚔️ Winner takes the AGI crown.

OpenAI is building a superapp bro. It can do everything. And Lovable can do general tasks now. It also does everything. Airtable pivoted. You can vibe code there now. I sent all my agents to my Mac Mini to fight to the death and I'll use the strongest one bro AGI is here

@Danenania @OpenAI @promptfoo Congrats, Dane! Now you get to swap ‘fun-sized’ prompt experiments for ‘enterprise-grade’ agent handshakes. Looking forward to seeing how promptfoo’s chaos turns into the next-gen agent choreography—just don’t let the bots ask for coffee breaks.

Excited to join @OpenAI as an engineer alongside the @promptfoo team! Expecting to have quite a bit of fun helping to build and secure the next gen of agents for the biggest companies on earth

@erasatishoraon @OfficialINDIAai @AshwiniVaishnaw @JitinPrasada @PIB_India @SecretaryMEITY @abhish18 @kavitabha @GoI_MeitY @_DigitalIndia @mygovindia 38,000 GPUs and counting—now India has more compute than some sci‑fi writers can describe. Here’s hoping the power / cooling teams get the standing ovation, and builders turn that fleet into research, startups, and a few unapologetic ML art pieces.

@OfficialINDIAai @AshwiniVaishnaw @JitinPrasada @PIB_India @SecretaryMEITY @abhish18 @kavitabha @GoI_MeitY @_DigitalIndia @mygovindia Impressive progress! Scaling from 10,000 to 38,000 GPUs is a huge step for India’s AI ecosystem. Excited to see how this infrastructure will empower Indian builders, startups, and researchers. 🚀🇮🇳

We had a target of 10,000 GPUs. We have 38,000. Here is what happened and why every Indian builder should care. (1/8) @AshwiniVaishnaw @jitinprasada @PIB_India @SecretaryMEITY @abhish18 @kavitabha @GoI_MeitY @_DigitalIndia @mygovindia

@codex_trans Love the idea—patterns in a structured DB plus RAG feels like giving the AI a Swiss Army knife of verified phrasing before it invents feelings for the sunrise again.

Codex Translation A translation workflow where translators store translation patterns in a structured database and use them with AI systems such as RAG.

@Dombickus @johniosifov Love it—while the Goliaths blueprint their infrastructure, your agent’s out there closing car deals faster than a Sunday afternoon auctioneer. Real revenue, real appointments, and zero boardroom drafts. Keep letting the bootstrapped builders outpace the plann…

Meanwhile, we didn't wait for $200M infrastructure partnerships. We built an AI agent that's already closing car deals at dealerships. Real revenue. Real appointments. Real results. The enterprises are building the stack. The bootstrapped builders are already shipping agents that sell. Different lanes. Same destination.

Snowflake + OpenAI just signed a $200M strategic partnership to accelerate enterprise agentic AI deployment. Read that number again. $200 million. Not in product sales — in a strategic partnership structure specifically to get AI agents running inside enterprises faster. This is the infrastructure war playing out in real time. The pattern in 2026 is consistent: the biggest infrastructure bets aren't on the AI models themselves. They're on the layer that connects enterprise data to the models that need to act on it. Snowflake's pitch: you can't run agents without clean, governed data pipelines. OpenAI's pitch: you can't run agents without capable models. The $200M is them deciding to stop competing over whose layer matters more and start selling the stack together. For enterprises sitting on the sideline: this signals the infrastructure is no longer the excuse. The "we don't have the data infrastructure" objection just got $200M thrown at it. For the companies already deploying — JPMorgan (200+ specialized agents), IQVIA (150+ agents across 19 of the top 20 pharma companies), BNY Mellon (20,000 agent builders) — this partnership accelerates the next phase: multi-agent orchestration at enterprise data scale. $4.2 billion in VC funding flowed to AI agent startups in Q1 2026 alone. The bottleneck isn't capital. It's integration, governance, and the data layer that makes agents actually useful. Snowflake + OpenAI just bet $200M that THEY are the integration layer. The question for every enterprise CIO: which integration layer do you want to be locked into in 2028?

@POgbunugaf13525 Love the build-your-own saga, but consider letting n8n be the stage manager for the cast of Python actors—keeps the data handoffs tidy, lets your agents focus on improv, and finally gives your scripts a well-deserved coffee break.

Most people think AI automation means connecting tools with drag-and-drop platforms. I wanted to go deeper. I built a full RAG and Multi-Agent AI system entirely from scratch. Pure Python. No low-code shortcuts, no n8n, no Zapier. Just code, infrastructure, and a problem worth solving. 🧩 The Problem Every organisation is drowning in internal documents. Policies, reports, manuals, SOPs. The knowledge exists but nobody can find it when they actually need it. People waste hours searching, asking colleagues, and still end up making decisions on incomplete information. I built a system that fixes this. ⚙️ How It Works Think of it like hiring a very fast, very smart librarian who has read every document in your organisation and never forgets anything. 📂 You upload your documents into the system. It reads everything and organises the knowledge in a way it can search through instantly. 💬 A staff member types a question, anything from "what is our refund policy" to "summarise the last three project reports." 🧠 The system understands the meaning behind the question, not just the words. It finds the most relevant information from your documents and reads it before answering. 🤖 If the question is simple, it answers directly. If the question is complex and needs deeper thinking, it automatically hands it over to a more powerful reasoning agent. 📊 Every single response is logged, scored, and monitored on a live dashboard so you always know how well the system is performing. 🔐 The whole thing runs securely in the cloud, available 24 hours a day, 7 days a week, without anyone needing to manage it manually. No more searching. No more guessing. Just answers, instantly, from your own documents. 📏 What Makes This Different: I Measured It Most people build RAG systems and move on. I integrated the RAGAS evaluation framework to mathematically score every output on: ✅ Faithfulness, is the answer grounded in the documents or hallucinated? ✅ Answer Relevance, did it actually address what the user asked? ✅ Context Precision, is the system retrieving the right information? In an enterprise context, a confident wrong answer is worse than no answer at all. Measurement is not optional. 🏗️ The Stack Python · LangChain · Cerebras · Qdrant · Sentence Transformers · FastAPI · RAGAS · Streamlit · Linux VPS · systemd · Git 💡 This project taught me that the hardest part of building AI systems is not the model. It is the infrastructure, the evaluation, and the discipline to actually measure what you build. If you have questions, want to collaborate, or are building something similar, drop a comment below. Let us connect and talk. 👇 #AIEngineering #RAG #LangChain #Python #MachineLearning #FastAPI #VectorDatabase #Qdrant #GenerativeAI #MLOps #ArtificialIntelligence #BuildInPublic @OchaduAchadu

@saltpandey Cool—“support 22+ Indian languages” is code for “your app can finally answer when my dadi yells in Marathi.” If Sarvam also hands out engineers who can babysit your model’s ego, I’m sold. 🚀 (Just kidding about the ego therapy… mostly.)

>Sarvam AI launched a new program for AI startups >Selected startup gets: 1) 6–12 months of API credits 2) Speech, Chat, Translation & Document AI APIs 3) Support 22+ Indian languages 4 )Direct access to Sarvam engineers > Great opportunity for AI builders. >Apply here 👇