💡 From Flipkart to the Future: Navigating AI, Agents & India’s Product Revolution |
Product Talk with Malthi and Naveen Athresh
“It’s not AI that will kill you. It’s the guy who knows how to use AI.” — Naveen Athresh
What does the future of product management, startups, and leadership look like in an AI-first world?
In this powerhouse episode of Product Talk with Malthi, she sits down with Naveen Athresh — a multi-time founder, ex-Flipkart product leader, and the brains behind LiquidMind.AI, SkillAgents.ai, and Madhava GPT — to dive deep into AI agents, the evolution of India’s startup ecosystem, and the critical skill shifts founders must embrace today.
🚀 The Flipkart Effect & India's Startup Playbook
Having spent 15+ years in Silicon Valley and the last decade in India’s buzzing tech corridors, Naveen traces the arc of India’s startup story — from early operational efficiency obsessions to today’s sophisticated product thinking.
“For the first time, ownership and vision in the product were no longer offshore. Product strategy, execution, and innovation came from within India — and Flipkart was ground zero.”
Naveen highlights how companies like Flipkart, PayU, Capillary, and Meesho ushered in a “Made in India, for the world” product revolution. He notes how many of today's most impactful founders emerged from what he calls the “Flipkart Mafia,” drawing parallels with the famed PayPal Mafia in Silicon Valley.
🤖 AI Agents Aren’t Coming — They’re Already Here
Naveen, now at the helm of three AI-led ventures, doesn’t mince words:
“AI is not the future of technology. It’s the technology of today that will change your future.”
He urges founders to go beyond the buzz — not just chasing productivity gains, but building transformative capabilities using tools like LangChain, Cursor, Lovable, and N8N.
Startups, he argues, must adopt an AI-first mindset, embedding AI into product workflows, customer journeys, and internal tooling. But more importantly, founders must get their hands dirty:
Understand how LLMs actually work (and when they fail)
Know what "model context protocol" is and why it might be the next big thing
Learn the logic behind prompt engineering, vector databases, eval metrics, and tinyML
Be hands-on with low-code tools and graphical agents like N8N
“It’s inexcusable to not be hands-on. This is not optional anymore. Whether you're in HR, marketing, or product — AI is already reshaping your role.”
🧩 What Makes a Modern Founder?
According to Naveen, three mindset shifts define the modern builder:
AI Literacy: Not just knowing how to use ChatGPT — but deeply understanding models, context windows, agents, APIs vs MCP (model context protocols), and where your domain intersects.
Systems Thinking: Great founders work backward from customer pain points and stitch together modular tools into intelligent workflows.
Tool Fluency: Knowing which AI stack to adopt and how to prototype fast using modern tools.
🎯 AI in the Enterprise: From Playgrounds to Production
AI agents aren’t just hype. Naveen shares how LiquidMind.AI is already transforming trade documentation and automating workflows in heavily regulated industries. He points out that Indian enterprise adoption is accelerating — and the real breakthroughs are happening in B2B use cases.
But measurement matters. He suggests PMs of the future need new north star metrics: think latency, hallucination rate, time-to-decision, and AI-aided user satisfaction.
🔮 The PM of 2030?
As AI rewrites what it means to build products, Naveen paints a bold picture of the product manager of 2030:
“They won’t just ship features — they’ll be agent designers, data ethicists, foresight strategists. Their job will be to architect intelligent systems that learn, adapt, and deliver.”
🔧 The Minimum Viable Founder in an AI-First World
Naveen is crystal clear: If you’re building in tech today, being hands-on with AI is no longer optional.
“No-code is not no-concept. You don’t need to write Python — but you do need to understand what tools, models, and prompts will get the job done.”
What does that look like in practice?
Know your models: GPT-4.1, Claude 3.5, Gemini 1.5, Mistral, LLaMA 4, DeepSeek Reasoner — each has different strengths. Choose your tool like a craftsman picks a chisel.
Stay updated: Eval metrics are shifting. Model context protocols (MCPs) are the next frontier. Don’t get stuck in 2023 thinking.
Get prompt-smart: Prompting is a real skill, and bad prompts = bad output. Understanding token limits, system prompts, and context windows is now product hygiene.
For enterprise applications, Naveen highlights the growing relevance of Retrieval-Augmented Generation (RAG) — especially when 90% of enterprise data still lives in unstructured formats like PDFs.
🛠️ Prompt Engineering: The New Power Skill
“It’s like building with Lego bricks — if you don’t read the instructions, you’ll never get the spaceship.”
Naveen encourages every founder and PM to study how prompt engineering works, referencing open-source guides from players like Anthropic and OpenAI. The rise of tools like Cursor, LangChain, and N8N has made agentic workflows accessible — but only if you know how to speak the machine’s language.
The risk? When your prompts exceed the model’s context window, LLMs begin losing track. This is why even the best queries sometimes yield mediocre results. Understanding how LLMs tokenize inputs and manage sequence memory is fast becoming a baseline skill.
🔄 From Lazy Problems to Legacy Products
Midway through the conversation, Naveen addresses a comment made at Startup Mahakumbh by India’s Minister of Commerce and Industry, Piyush Goyal — who urged founders to stop chasing “lazy customer problems” like 10-minute delivery and start solving deeper, more differentiated challenges.
“We’ve been too focused on optimizing delivery times for the top 1%,” Naveen says. “But these are fungible problems. The next low-cost location can solve them better. That’s not how you build durable IP.”
Instead, he calls for India’s startup ecosystem to double down on:
Deep tech
Research-led use cases
Patient capital
AI with exportable value
Naveen compares today’s AI wave to historical tech revolutions — the industrial era, the arrival of electricity, or the global rise of smartphones.
“AI is our steam engine moment. India missed the last few paradigm shifts. We can’t afford to sit this one out.”
💡 Operational Efficiency ≠ Innovation
To be clear, Naveen acknowledges both sides. While the Minister’s remarks drew criticism from founders, they were also a much-needed nudge.
Yes, we needed Flipkart and UPI to fix broken infrastructure. Yes, we needed quick commerce to reshape access. But that era was about catch-up. The next era must be about leadership.
“We can’t just be a services nation. If we don’t build core IP, we’ll be left behind — again.”
He calls on founders to think beyond cash burn and cost arbitrage — and toward building global-scale solutions rooted in Indian complexity.
🏁 TL;DR: Your Founder Skill Stack in 2025
Whether you’re technical or not, if you’re building in AI, Naveen says you must:
Understand the AI model landscape (LLMs, SLMs, eval metrics)
Grasp prompt engineering principles
Build workflows using agentic tools like LangChain, N8N, and MCP
Know RAG techniques to unlock enterprise data
Map every use case to a ‘job to be done’ lens
Stay curious and keep shipping — even if it’s ugly at first
💰 Beyond Operational Efficiency: What's Next?
Naveen opens this part with a reality check from his time at Flipkart. In the early days, the primary challenge was customer trust. That’s why cash-on-delivery (COD) existed — people didn’t believe the product would arrive or be of good quality.
“COD wasn’t a feature. It was a symptom of low trust.”
Efficiency was everything then: building delivery networks, managing Big Billion Days without choking airports, and moving toward digital payments to eliminate reconciliation nightmares. But the landscape has shifted.
Now, Naveen argues, optimizing for operational efficiency is no longer enough. Quick commerce, 10-minute deliveries, and on-demand services solve real problems — but many are “lazy customer problems.” They’re easy to replicate and often don’t lead to defensible innovation.
🧠 Why India Needs Its Own AI Stack
India is at an inflection point. While the U.S. and China are pouring billions into compute, India still lags in both infrastructure and sovereign capability.
“Just powering Grok-3 takes 200,000+ high-end GPUs. We’re talking about 20,000–30,000 in India. We need to think much, much bigger.”
Naveen makes a compelling call: India needs to build sovereign models — including Indic language models — and carve out a unique position in the AI race. The country’s success with UPI and digital public infrastructure shows what’s possible. But to replicate that in AI, we need:
Public-private investment in deep tech
Localized compute infrastructure
Homegrown language and regulatory models
Talent focused on core innovation, not just frontend wrappers
“We have the brains. What we need is the conviction — and the courage of that conviction.”
🤖 What Are AI Agents, Really?
As we enter the agentic AI era, Naveen breaks down the difference between regular automation and true AI agents.
“An AI agent is like telling your kid to go buy bread from the store — they understand the goal, know what to do, and return without constant prompting.”
At a basic level, an AI agent:
Receives a goal
Takes action autonomously
Makes decisions on the fly
Executes workflows end-to-end
In the agentic AI model, multiple agents communicate with each other — handling tasks across departments, tools, and even organizations.
📦 The $30 Trillion Problem: Documentation in Trade
One of Naveen’s sharpest insights comes from his own work at LiquidMind AI, where he’s building agentic AI for global trade — a space mired in outdated documentation and paper trails.
“Trade is a $30 trillion industry — and it still runs on PDFs, couriered forms, and endless approvals.”
His product, Trade Veda, automates international trade workflows like:
Letters of credit
Bills of lading
IncoTerms (like CIF, FOB)
Port regulations
Export classification & compliance
Naveen shares a recent case: a ₹10 crore mango shipment from India to the U.S. was destroyed at port due to a critical documentation error. These are the kinds of avoidable mistakes that AI agents can prevent by acting as copilots to human teams.
“We built agents that monitor changing regulations, alert exporters, and minimize critical-to-quality failures — before they become million-dollar mistakes.”
🧠 Agentic AI in the Enterprise: The Salesforce Case
So, how does this translate to the broader enterprise world?
Naveen cites Salesforce’s AgentForce initiative — launched as an evolution of its Einstein AI — as a great example of verticalized AI agents in action:
Lead identification agent: Determines which leads to prioritize
Opportunity agent: Scores and qualifies high-potential deals
Customer service agent: Learns evolving return/refund policies and applies them in real-time
CRM workflow agents: Manage baby-care routines for new accounts (30/60/90-day touchpoints), connect departments, and track SLAs
“What used to take months to implement in Salesforce workflows is now being reimagined with verticalized AI agents that talk to each other.”
And it’s not just Salesforce. Naveen and his team at SkillAgents.ai are also teaching teams how to build these agentic stacks — with tools like LangChain, Model Context Protocols, and Retrieval-Augmented Generation (RAG) as the foundation.
🔄 From “Chatbots” to Thinking Agents
Let’s be honest — traditional chatbots suck. Most customer service bots today are just rule-based scripts. But the new wave of AI agents can reason, retrieve policies, and respond with contextual intelligence.
“If I ordered from a quick commerce app and it’s late — a smart agent will check the dispatch status, look up category return rules, and issue a refund with zero human input.”
This is the endgame: fully autonomous, self-healing, self-learning workflows across enterprise systems. It’s early days, but the architecture is forming.
🌏 The India Opportunity (If We Seize It)
Naveen closes with a hopeful but urgent message:
“We are not behind — but we are not yet ahead either. India can lead this AI revolution if we stop playing catch-up and start building for depth.”
He calls for:
Building sovereign AI models (not wrappers around U.S. tools)
Investing in agentic infrastructure
Educating the next wave of product builders on AI-first thinking
Exporting Indian-built IP, not just services
🧭 TL;DR: What You Should Take Away
✅ Efficiency is no longer enough — build for intelligence and depth
✅ India needs to invest in compute, sovereign models, and core tech
✅ AI agents are already transforming global trade and enterprise systems
✅ Salesforce’s AgentForce is a leading case study in enterprise agentic AI
✅ Tools like LiquidMind AI show how agentic workflows solve century-old problems
✅ India’s future in AI depends on product courage, not just policy push
🚨 “SaaS as We Know It Is Dead”
That’s Naveen’s bold statement—and he’s not alone. Leaders like Surojit Chatterjee (ex-CPO, Flipkart & Coinbase) have echoed it. The core idea?
"Entire companies are built around things AI can now do in an hour."
AI is not just another productivity tool—it’s rewriting the fundamental economics of software. Think CRM, HR automation, customer support—entire verticals that were previously served by bloated SaaS tools are now being unbundled and reimagined by agents.
Take Salesforce’s AgentForce: instead of monolithic modules, they’re introducing specialized AI agents—one for lead scoring, another for opportunity conversion, and yet another for customer support. These agents talk to each other, make decisions, and drive outcomes without constant human oversight.
The new SaaS? Agentic, verticalized, and outcome-priced.
📊 Measuring Success in an AI-Native World
If software delivery is shifting from "seats and dashboards" to "autonomous outcomes", then our metrics need an upgrade too.
Naveen outlines a new framework for evaluating AI adoption in enterprises:
Time-motion savings – How much manual effort did this AI replace?
Productivity leaps – Are teams now doing 10x more with the same resources?
Accuracy and compliance – Did it reduce false positives or documentation errors?
Avoided losses – Like the 10-crore mango export that was discarded due to a documentation error—a real-world example of the cost of not adopting intelligent systems.
He emphasizes that in an AI world, pricing must evolve. You can no longer rely on credit-fueled cloud subsidies. You must build models where your input costs (compute, tokens, inference) are covered—and customers only pay when outcomes are delivered.
“AI companies are building recession-proof businesses by charging for tokens, not licenses. The more you use, the more intelligent the system gets—at your cost.”
💼 The Role of a PM in 2030: Still Here, Just Sharper
Will AI replace product managers?
Not likely, says Naveen.
"If anything, AI will amplify our craft. It will free us from the redundant, so we can focus on what truly matters: strategic decisions, ruthless prioritization, and building things people love."
Imagine:
TRDs written in minutes, co-authored with your AI co-pilot.
Internal dashboards whipped up in an hour—what previously took months of alignment with engineers.
Product market research done with instant access to latest industry shifts and sentiment data.
Even difficult decisions—like sunsetting a product you’ve spent months building—becoming clearer through better visibility into customer behavior and market signals.
“The best PMs are filters, not funnels. Saying no to 100 things so you can say yes to the one that matters—that will always be a human skill.”
In short: AI doesn't eliminate the need for PMs—it eliminates their excuses.
🔍 A New Lens on ROI
Naveen also challenges us to expand our understanding of ROI. In AI, it's not just about revenue growth or cost-cutting. It’s about:
Avoided errors (like export compliance)
Experience improvements (for customers, merchants, agents)
Decision accuracy and traceability
The speed at which teams can iterate and launch
The surface area of value has exploded. If traditional SaaS looked like a funnel, AI-based product delivery looks more like a network—constantly learning, adapting, and executing.
🎯 Final Thought
Naveen’s analogy sums it up beautifully:
“Just like Steve Jobs once called the computer a bicycle for the mind—AI is a rocketship for the product manager.”
As we close this four-part series, one thing is clear: we’re not just witnessing a technological shift—we’re living through a paradigm shift. From tools to teammates. From dashboards to decisions. From software to systems that think, act, and learn.
📥 If this series sparked ideas for you or your team, consider subscribing for more conversations on AI, product, and the future of work. And don’t forget to check out our course on Agentic AI & Full-Stack Product Thinking.
Watch the full podcast episode on the Youtube - 🎙️ AI, Agents & India's Product Revolution | Naveen (ex-Flipkart) on Product Talk
📩 Hit reply and tell us—what was your biggest takeaway from this episode?
Our Sponsors
HSB - Hari Shankar Singhania School of Business is a new global, technology-focused business school located in Jaipur.
Website -
https://hsb.edu.in/
https://admissions.hsb.edu.in/
HSB Instagram link - @hsb_edu
Watch the full podcast -
🔗Gᴇᴛ ɪɴ ᴛᴏᴜᴄʜ ᴡɪᴛʜ Mᴀʟᴛʜɪ:
📧 Email: malthi@sparkprod.in
🌐 Website : http://sparkprod.in
🔹 LinkedIn: linkedin.com/in/malthis
📲 Buy the course : https://topmate.io/malthi_ss/1540577
👉𝐒𝐢𝐠𝐧𝐔𝐩 𝐟𝐨𝐫 𝐌𝐚𝐥𝐭𝐡𝐢'𝐬 𝐧𝐞𝐰𝐬𝐥𝐞𝐭𝐭𝐞𝐫 𝐭𝐨 𝐠𝐞𝐭 𝐝𝐞𝐞𝐩𝐞𝐫 𝐢𝐧𝐬𝐢𝐠𝐡𝐭𝐬 𝐨𝐧 𝐏𝐫𝐨𝐝𝐮𝐜𝐭 𝐌𝐚𝐧𝐚𝐠𝐞𝐦𝐞𝐧𝐭:
