Chandini Nagendra

While the understanding of the Model Context Protocol (MCP) for agentic integration is still evolving, most of its current applications are centered around providing ‘runtime tools’ to AI agents along with relevant context. In one of my recent projects, we were using an SDK for DDS which was extremely helpful at providing a comprehensive collection of tools, libraries, documentation, code samples, etc. I was wondering if it was worth the effort to turn the SDK from a static library into an intelligent, interactive, and adaptive tool - empowering developers and testers to build and validate apps/utilities faster, safer, and with higher quality. One way of doing so would be to add a Model Context Layer to the SDK, which enables a structured, machine-readable representation of the SDK’s concepts, templates, and domain-specific knowledge that helps the LLM agent to generate valid code for a very consistent and accurate usage of the SDK.   This provides high precision and consistency, with added benefits like cutting down onboarding time and learning curve, and bridging knowledge silos that usually form across different user groups like service development teams, V&V teams, etc. This should also facilitate easier SDK upgrade cycles, and adoption across user groups. What are your thoughts on this? #MCP #Agentic-AI #SDK-intelligence #high-precision-code-generation Arkid Mitra Sunil Kulkarni Mukund Dharwadkar Abhishaik Srivastava Vidya Singh Mohan Kulkarni Navya Mishra Arvind Gupta Devansu Kasyap

6

3 Comments

I delivered a fireside chat with R.V. Guha on #NLWeb at #MCPDevDays this week. For those who missed it, our session was recorded at https://lnkd.in/gauZ-YyD.  𝐒𝐨𝐦𝐞 𝐤𝐞𝐲 𝐭𝐨𝐩𝐢𝐜𝐬 𝐰𝐞 𝐜𝐨𝐯𝐞𝐫𝐞𝐝 𝐚𝐫𝐞: 🕸️ Why NLWeb was created (to bring the power of LLMs to more websites) 🍛 Demos: Serious Eats, Backcountry, and more (that show bringing the semantic understanding of LLMs to question answering over a website collection of recipes or product inventory) 🧑‍💻 How NLWeb exposes an MCP endpoint 📊 What the underlying data looks like, which leverages schema.org ❓ What NLWeb is (a protocol and an implementation - see https://lnkd.in/g--NaaaW) 🛠️ How do you set it up? There are two ways - data loading is described at https://lnkd.in/gq-SRk_W or just use the MCP endpoint.   🧑‍🤝‍🧑 How NLWeb relates to MCP: both empower the "Agentic Web", a future in which both humans and computer agents can use the Web alongside each other. That encompasses a collection of technologies including MCP (which allows LLM-powered software to connect to external resources) and NLWeb (which allows website creators to present their websites as MCP Servers). They are different-level protocols that should be decoupled, like HTTP and HTML.   💽 How are we thinking about memory in NLWeb?   ❣️ The top reasons that I am excited about NLWeb:  • The power of LLMs on every website  • Searches over your item catalog (won't hallucinate items) • Open standard – easily switch out the model, embedding model, and vector database used • Tools to import structured data • MCP integration, so makes your website agent-compatible 𝗦𝗼𝗺𝗲 𝗠𝗼𝗱𝗲𝗹 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗣𝗿𝗼𝘁𝗼𝗰𝗼𝗹 (𝗠𝗖𝗣) 𝗥𝗲𝘀𝗼𝘂𝗿𝗰𝗲𝘀 𝗳𝗿𝗼𝗺 𝘁𝗵𝗲 𝗲𝘃𝗲𝗻𝘁: • Event page: https://aka.ms/mcpdevdays • Day 1 Stream: https://lnkd.in/gqt8uDy7 • Day 2 Stream: https://lnkd.in/gfHhAtCW • Get started today: https://lnkd.in/gem8hmD2 • Videos: https://lnkd.in/gkutDrnr • Let's Learn MCP workshops: https://lnkd.in/gSUVAmEY • Microsoft Learn Docs MCP Server: https://lnkd.in/g7Mj2B8X

51

System Design Part 2 Continuing my System Design Blog Series, this post is a deep dive into the Low-Level Design of WhatsApp. 📱 From handling 2 billion daily messages to ensuring high availability, we’ll walk step by step through how WhatsApp is architected at scale. 🚀 High-level estimate Here is our high-level estimate: TypeEstimateDaily active users: (DAU)50 million Requests per second (RPS) : 24K/s Storage (per day) : ~10.2 TB Storage (10 years) : ~38 PB Bandwidth : ~120 MB/s

6

have been seeing a lot of posts that human coding is dead, ai this ai that but not a lot of people are actually talking about how big of a bubble we are actually in. frontier ai providers have been giving us access to their services with massive subsidiaries and it was bound to stop at some point. now the real cracks are beginning to show. all I can say is, don't stop learning. ai is not yet the magic pill everyone is making it out to be. don't be a cripple, stay knowledgeable.

3

1 Comment

🚀 Opus 4.6 and Claude Code Team Agents just dropped, so I skipped the demos and threw it at real work. 📖 A while back, during my Amazon Payments days, Keshav Narsipur gave his engineering leaders The Goal by Eliyahu Goldratt as reading material. It stuck with me hard. The core idea — Theory of Constraints — is simple: your system only moves as fast as its slowest bottleneck. Highly recommend it. I've had that framework rattling around in my head ever since. So when Opus 4.6 and Claude Code Team Agents dropped, I told the agents to apply it. Gave them full access — GitHub, telemetry, JIRA, JIT, CloudTrail — invested time in a solid /plan mode, and let them rip for a few days. 📊 It's already surfacing bottlenecks I wasn't thinking about with clear, actionable insights. Some of which I've already passed to other agents for quick wins. 🔧 One thing people need to hear: Claude Code isn't just for coding. It's the best damn agent out there for many things — analysis, planning, investigation, orchestration. The name undersells it badly.

42

2 Comments

Alignment Is the New Scale: Lessons from Our Platform Engineering Leadership Offsite Most scaling challenges in large organizations are not technical. They’re alignment problems. Last week at our Santa Clara HQ, our Platform Engineering leadership team : spanning Quality Engineering and Production Engineering, came together for an offsite. --> Different charters. --> Different operating rhythms. --> Different definitions of risk. Yet one clear unlock emerged: Shared context : amplified by shared self-awareness. When Quality Engineering (guardians of customer experience) and Production Engineering (custodians of mission-critical infrastructure) operate from the same strategic context, something shifts: • Conversations move from functional optimization to enterprise impact • Debate becomes sharper : and safer • Decision velocity increases • Execution compounds But context alone isn’t enough. We layered in leadership frameworks like DISC and Strengths-based insights : not as HR exercises, but as execution tools. Because when leaders understand how they’re wired: Speed vs. precision | Stability vs. innovation | Direct challenge vs. consensus building Friction turns into fuel. In global, distributed organizations, misalignment rarely comes from lack of talent. It comes from invisible differences in style, incentives, and assumptions. Bringing those to the surface builds trust. → Trust accelerates execution. → And execution drives strategy. What stood out most was the maturity in the room : leaders thinking beyond their functions and aligning around enterprise outcomes. The future of platform engineering isn’t reliability or quality in isolation. It’s integrated excellence : powered by aligned leadership. It’s integrated leadership at scale! Curious : what has been the biggest unlock for alignment in your organization? #ExecutiveLeadership #CXO #EnterpriseLeadership #TechnologyLeadership #EngineeringLeadership #DigitalTransformation #StrategicAlignment #OrganizationalExcellence #HighPerformanceTeams #FutureOfWork #PlatformEngineering #SoftwareEngineering #LeadershipDevelopment #OperationalExcellence #Scale #BusinessStrategy #InnovationLeadership #SoftwareQuality #QualityEngineering #EngineeringExcellence

19

Everyone optimizes for throughput. The real bottleneck is resilience under burst traffic. Rate limiting is traffic control for APIs: it enforces rules on how many requests pass in a window to protect services from overload. The concept is simple, but the impact is profound: it preserves availability, reduces backpressure, and clarifies service contracts. The documentation notes that careful limits keep the system stable without strangling legitimate use. ⚡ The documentation frames rate limiting as flow control. 🎯 Limits must balance protection with usability; too generous wastes resources, too strict causes 429s. ───────────────────────── 🔗 Read the full article: https://lnkd.in/gF2fcQcG 🎯 Practice interview questions: https://lnkd.in/gmy5drNw #backend #API #Microservices #limiting #protect

1

aisuite is basically offering: - A structured way to attach actual tools to LLMs - Simple agent loops - Observability (so you can see what the agent actually did) - A repeatable pattern for experimentation

2

Awesome conversation between Suresh Mathew and Gil Feig. This episode hits on the new push to go back to on-prem. It's interesting to see Gil's experience as a CTO navigating what this looks like for engineering teams. Lots of good info on support, production access, and the tradeoffs behind on-prem. Watch the rest of the convo here: https://lnkd.in/gvp97zgM

2

Our recent work on how we built Warp Speed Loader — sustaining 48× speed with a 90% reduction in costs!

40

1 Comment

With the rising costs of GPUs, engineering teams are on the hunt for smart ways to manage them. Our team has found a breakthrough, and Pooja Malik, our lead engineer, shares their creative strategy in depth. Listen in as Pooja reveals why standard GPU metrics might mislead, the journey behind crafting a fresh utilization algorithm, and the hurdles still ahead. Know any engineers dealing with GPU cost issues? Pass this along to them. Watch the video here: https://lnkd.in/gSXg52Ed What tactics have you discovered to address tech expenses effectively?

4

1 Comment

AI Boom and the Future of Developers - I see AI bringing automation for initial project setup, like creating full orchestration with all tools, reducing development time. Developers will focus more on adding new features and ensuring applications and infrastructure run correctly. AI can handle lower priority stuff, but fixing critical bugs with AI is still very critical and seems far off for AI, and in the future, we can’t say anything for sure. Developer roles will evolve, mixing coding with AI and ML skills, with current developers leveraging their solid application domain knowledge to create project-specific AI models for fixing, checking, managing Sev3 issues, alerts, and other operations, while also reviewing AI outputs, making these skills crucial and relevant in the future so I don't think developer roles in risk anytime soon. Security is a big factor as fast-paced development opens new challenges and work scopes. Many new things will evolve, and developers will take on different roles and responsibilities in this AI-driven future. What’s your take on how AI will change the developer’s role?

5

🎶 AI Application and Indian Classical #Music I’ve always respected all forms of music, but since I understand Indian classical music deeply, that’s where my analogy comes from. When I talk about AI-first thinking, I like to compare it to Indian classical music. In classical music, every instrument plays within a raga—a structure, a set of rules. Yet within that framework, each performance feels fresh: new rhythms, subtle variations, and unique expressions every time. That’s exactly( more-or-less) how #AI applications work. They aren’t static like traditional web features. Every interaction generates something slightly new ...dynamic, adaptive, and tuned to the use ...while still staying grounded within their framework of trust and guardrails. Just as music evolves within its raga, AI evolves within its guardrails. The result? Experiences that feel alive, personalized, and trustworthy (WIP). �� So who here enjoys creating their own “new tune”? Do you agree? #AIFirst

9

1 Comment

I was hearing a short on YouTube on Steve Jobs and the interviewee said "Steve wasn't a nice person. He refused to accept noise from anyone. He demanded pure signal." Hence this post. Signal vs Noise Systems, generate noise and signal. Signal is a message about what is true about a system that can be actioned. Noise maybe true temporarily, false, intermittent, corrupted, benign, or any other type of message that gives no valid insight into the state of a system and hence it is not actionable. When a system fails there follows a Root Cause Analysis. The idea being to find the root cause so we can prevent it in the future. There is a tonne of noise in this process not only because of the available data but also there are people's egos at play. People don't want to be found guilty. In high ego workplaces the RCA becomes, "this pod was having trouble so we restarted it" or if we have some ambitious team lead there, they will add, "thanks to our improved xxx, we found the issue quickly and were able to rectify the problem by reloading the pod". That whole paragraph is noise. There is nothing there that allows you to make sure it doesn't happen again. On the contrary it says everything is fine and will continue to be fine forever. Not lying to yourself is the first step towards mastery. At team level this becomes harder and political. The result of a company lying to itself ultimately is bad tech, if you are IT, lawsuits, if you live in a law abiding nation, death or harm to customers, fraud enron, lehmann etc and the list of bad things we seem to see quite regularly in the news. Ultimately, noise wins when leaders prefer the noise. Employees can see the amount of noise flying around as if it were signal. When they see what flies they start flinging it too. Its a downward slope. Customers hate noise. It causes frustration and much anguish. Ultimately they churn bringing down the reputation of the company. tl;dr if you are a leader, expect signal from your teams. Don't accept noisy handwavy explanations. Sounds easy but its super hard to do as people's egos are easily hurt and how you manage that is ultimately the mark of a good leader.

1

🚨 150+ applications in 10 days… but something is seriously off 🤯 10 days back, I posted for hiring QA Engineers and Full Stack Developers. Since then, I’ve been getting emails every single day 📩 And before I sleep, I check them. Honestly… I didn’t expect this. But what I’m seeing is frustrating 👇 Same resume… just name changed Same email body… copy paste No subject line No intro… nothing Just attachment and send Some emails literally look identical. At one point I was like — is this even real or am I reviewing the same person again and again? 🤦‍♂️ And the biggest problem? I can’t even figure out who is genuine and who is not. That’s the sad part. Because I know there are good candidates in there… But they’re getting mixed up with all this noise. This is not about blaming anyone. But let’s be real for a second If you’re applying to jobs like this… you’re actually hurting your own chances. Just take 2 minutes ⏱️ Write a proper subject Say something in the email Don’t send the same resume everywhere That’s it. Because right now, effort itself is becoming the filter ⚡ Are you also facing this issue while hiring? 🤔 Or is it just me? Would love to hear from others. #Hiring #Recruitment #Startups #TechHiring #Careers #JobSearch

6

Agentic coding workflows have become essential to modern development. Here's my comparative analysis of the tools reshaping how we build software—with honest benchmarks and practical guidance on when to use each.

16

AI is everywhere, and conventional wisdom says you’re falling behind if you’re not using it. Naturally, many flock to ChatGPT, impressed by its vast knowledge and comprehension—and occasionally chuckling at its hallucinations. As a software developer or organization building applications, what’s your AI strategy? The conventional path seems clear: start with ChatGPT, move on to frameworks like LangChain for Retrieval-Augmented Generation (RAG), and eventually develop sophisticated pipelines and agentic AI. But here’s the contradiction: the right approach for software developers is to focus on fine-tuning to improve the models themselves. RAG works by indexing external knowledge into vector databases and appending retrieved snippets to prompts—essentially giving the LLM a cheat sheet for each response. This method has fundamental limitations: LLMs don’t truly learn from these snippets; they only reference them temporarily. The result? Inconsistent outputs and unnecessary system complexity. The LLM remains one step removed from this corpus of knowledge. If you have specialized data or domain expertise, why settle for this halfway solution? Instead of patching a base LLM with external retrieval, fine-tune the model directly on your data. Fine-tuning weaves your knowledge into the model’s parameters, creating a smarter, more reliable solution tailored to your specific needs. Yes, RAG is popular because it’s quick to implement and avoids costly retraining. But it’s a temporary fix, not a lasting solution. Fine-tuning is the true path to AI that deeply understands your domain, delivers consistent results, and scales effectively. Best of all, it’s now accessible even to smaller players without billion-dollar budgets. In short: Don’t fall for the trend of layering retrieval on top of base models. If you want AI that truly serves your users, make fine-tuning your primary strategy—not just an afterthought. What’s your perspective? Are you still relying on RAG, or have you embraced fine-tuning as the way forward?

6

5 Comments

Most people mix up skills, agents, Codex, and Claude Code 👇 Here’s the simplest way to think about it: 🧩 Skills = reusable workflows Best for repeatable tasks where consistency matters. 🤖 Agents = goal-driven workers Best when the task needs planning, tools, and multiple steps. 💻 Codex = coding agent Best when you want to build, refactor, test, and ship code. 🧠 Claude Code = coding agent with strong reasoning Best when you want the model to understand the codebase deeply, make edits, run tests, and iterate. --- ### When to use what Use skills when the task is: • repetitive • structured • process-driven Use agents when the task is: • multi-step • dynamic • tool-heavy Use Codex when the task is: • implementation • refactoring • test generation • shipping code Use Claude Code when the task is: • code understanding • architecture thinking • complex edits • debugging with context --- ### Simple rule Skills help you do the same thing better. Agents help you do bigger things autonomously. Codex / Claude Code help you build software faster. --- The real power is not choosing one. It is knowing when to use each layer. #AI #Agents #Codex #ClaudeCode #Skills #GenAI #SoftwareEngineering

8

Excited to share that I’ve published 3 new npm packages for seamless Text-to-Speech integration across modern web applications 🚀 🔊 Built with simplicity, flexibility, and developer experience in mind. Packages Published Angular → https://lnkd.in/gHbU3Xni Core SDK → https://lnkd.in/g9fT73np React → https://lnkd.in/gACgPHc4 Features ✅ Supports 67+ languages ✅ Easy Text-to-Speech integration ✅ Modern TypeScript support ✅ Angular & React ready ✅ Lightweight and developer friendly ✅ Built on top of browser speech capabilities Whether you're building accessibility tools, AI assistants, learning platforms, customer support solutions, or voice-enabled apps — these packages can help you add speech functionality quickly. Would love feedback from the community and contributions are always welcome 🙌 #npm #javascript #typescript #angular #react #webdevelopment #opensource #tts #texttospeech #frontend #developers #accessibility

15