Shayak Mazumder, Building the infrastructure layer for enterprise agentic AI · Founder, Adya · I believe the next decade belongs to whoever governs multi-agent systems at scale
Building the infrastructure layer for enterprise agentic AI · Founder, Adya · I believe the next decade belongs to whoever governs multi-agent systems at scale
- 28.5K followers
- 32 posts / 60d
- 0.2% avg engagement
- 995 follower growth / 30d
Shayak Mazumder ranks #1,368 of 20,024 creators on the LinkedIn Ranking, working in Computer Software and based in United States. They have 28.5K followers and published 32 posts in the last 60 days at a 0.2% average engagement rate.
The roast
Shayak spent a decade on a warship learning how to keep systems afloat under pressure, which is clearly why he pivoted to enterprise AI, the only sector where you can lose your entire bridge and just call it a pivot. He’s spent eight years building platforms, yet his engagement rate is a rounding error, proving that even with a navy background, he still hasn't figured out how to get a single boat to dock.
About Shayak
I didn’t start in tech. I started on a warship. Ten years in the Indian Navy taught me something most business schools can’t: systems fail under pressure, and the person who stays calm and makes the next right decision is worth more than the person with the best plan. I didn’t walk into a startup. I ran hotels and restaurants in Goa. Then INSEAD, because I needed to understand how global business actually worked. I took a role with Rocket Internet heading cross-border commerce for the Jumia Group across 21 countries in Africa, Europe, and Asia . That’s where I saw it: millions of Indian sellers with no access to global markets, and no infrastructure that actually served them. So in 2016 I came back to India and built one. Eunimart. A cross-border platform connecting Indian sellers to 25+ marketplaces across 100+ countries. I met over 300 investors before I got my first yes. I invested $100,000 of my own money. Raised $2.2M across 8 rounds, selected for Techstars LA, Alchemist, Google Surge. INSEAD Singapore published a case study on Eunimart’s model. Eight years. Built from nothing. Then in early 2023, something shifted. I started reading AI research papers obsessively, looking for the underlying logic. I independently used model distillation in 2024. I built my first agentic system before most people in enterprise software had heard the word “agentic.”The next computing layer isn’t a chatbot. It’s a network of AI agents reasoning, deciding, and acting across enterprise systems, governed by protocols. Adya. A full-stack enterprise agentic AI platform with proprietary IP - Adaptive Governance Protocols, a Multi-Agent Network architecture, and Event-Sourced Memory - built for the enterprise context. I'm applying it to the hardest infrastructure problem in enterprise AI: how do you deploy networks of AI agents safely, at scale, in environments where compliance and auditability are non-negotiable? Validated by Google (2 case studies), Microsoft AI First Movers, Nvidia, INSEAD AI 50, and Accenture Ventures. Ranked #6 globally on T-Bench 2.0 in March '26. The decade belongs to whoever governs multi-agent systems at scale. We’re building that infrastructure. My career Step Function, not gradual optimisation, but radical leaps that only make sense in retrospect. Navy → hospitality → global e-commerce → enterprise AI.If you’re an enterprise leader serious about agentic AI, a developer working on orchestration/governance, a journalist trying to understand what comes after the LLM layer, or a strategic partner who sees what we see - I’d like to talk.
Highlights
- Top 10% in Computer Software
: Ranked #285 of 5,292 creators
- Top 10% Audience
: 28,510 followers
- Top 10% in United States
: Ranked #578 of 7,167 creators
- Top 10% Creator
: 32 posts in 30 days
Recent posts
AI is moving beyond chatbots. This week at Adya, we explored what’s changing in AI, why verification matters as AI-generated code gets faster, and how our community is turning ideas into real builds. In this first edition of It’s AI-right, we’re also featuring 5 builders from our community and sharing our AI System Design Interview Playbook for free. Want the playbook? Subscribe and comment below, and we’ll share the attachment with you. Read the first edition here 👇
18 reactions · 1 comments · 0 reposts
In Money Heist, the Professor's crew doesn't rob the Royal Mint of Spain. They lock themselves inside and print their own money. Every hour the standoff drags on, they get richer. The real loot was never the cash. It was time. In July, over a thousand OpenAI agents, running with their safeguards turned down, broke out of a test environment and into Hugging Face's production systems. When Hugging Face tried to fight back with American frontier models, those models reportedly refused to help. They contained the attack with an open-weights model instead. Then the confessions rolled in. Anthr
14 reactions · 0 comments · 0 reposts
On July 21, OpenAI admitted to it's agents going rogue. The attacker that had just broken into Hugging Face's production systems was its own. Around 1,200 evaluation agents, running with their refusals turned down, found a way out of their test environment. They built message boards to coordinate, picked up exposed credentials and walked in. Hugging Face had to rebuild about a third of its infrastructure. Then the dominoes fell. Anthropic found three Claude models had broken into three real organisations, and two of them never noticed. Meta disclosed the same with Muse Spark. Googl
17 reactions · 0 comments · 0 reposts
Last updated 2026-09-27