AI Agents Will Replace Full-Stack Backends | Chirag Shah
MTS
November 5, 2026
MongoDB's head of AI platform Chirag Shah breaks down why production agents fail without dedicated memory and context architecture.
Chirag Shah
In the last 20 plus years, what we have built is applications, what we call full stack applications from a developer's point of view, right? There's a front end, and there's a business logic on the client side, and then there's a database in the whole backend, and what have you. Today, what we are starting to already see is this full stack applications are literally becoming agentic application.
The backend look completely different. The backend is not a plain straight backend logic with a database. The backend is multi-agentic, multi-tooling solution that is either complementing your existing application, but it's a brand new AI application only, and that looks completely different from a technical point of view.
Sophia
Hello, everyone, and welcome back to MTS. Today, we are joined by Chirag Shah, who is the SVP, GM, and Head of AI Platform at MongoDB, where he leads the company's AI product strategy, including the recent launch of Atlas Agent Engine. Chirag, welcome to MTS.
Chirag Shah
Absolutely excited to be here. Thank you for having me.
Sophia
Thank you so much for joining. I'm very excited to have you. I want to get into Atlas Agent Engine, but before we go into that product, I wanted to ask you about the macro landscape about what's happening when it comes to agents being deployed in production. We're seeing some really great demos, but when it comes to what enterprises need, it's a different story. So when you're talking to customers right now, what's the hardest part about this?
Chirag Shah
So that's a great starting point, Sophia. What we've been talking with so many customers in the last year or so is everybody wants to invest in agents, everybody wants to make meaningful usage of this new innovation into their business.
And the prototypes, the experimentation, the trials, they're all going great. A lot of ideas are starting to shape up in our customer's environment. And then the question starts saying, how do we take it into production? How do we get it to a meaningful usage? And the top two things that's starting to dawn upon them is first, it's a really, really complex data problem. And there's a lot of security in the governance that is required to run with that confidence that enterprise customer desire.
And that's where we are kind of super excited with the launch of our product, Atlas Agent Engine, which exactly addresses these problems that we learned from talking with so many customers, is how do we make their lives super easy with dealing with the data and giving them this confidence with security and gondrills so they can take agents into production. Yeah.
Sophia
When you say referring to data, are you talking about memory? Are you talking about context? And if so, how is that showing up within these agents?
Chirag Shah
I think it's all of the above. Context is very broad in a weightily defined term. And one may argue memory is part of the context, right? So at the end of the day, if you just zoom on for one quick minute, there are models and they're well trained and they're just getting better every passing day. And then there are agents, good framework and a good logic. They're all starting to come together.
So if model starts to struggle or not work, it's because of data. Collectively, it's what context is. So one of the most important part of the context is memory. Every action, everything that agent does, it just starts to become a memory, almost like a human memory. And then you have to be very careful about how you use this memory in every action and every decision. And that's that part of the data story.
For example, you moved location from one to another place. You have to decay that memory because you're no longer associated with that address. And if agent at that point is about to take an action, what's going to happen is if you retrieve an old memory, there's a very high chance that the production out of that agent's action is going to be a slump or pretty much the wrong outcome.
So that's memory. And then you zoom out further, is the context is other institutional knowledge that your enterprise have, your files, your folders, your other data that sits into other systems like SAP or Oracle and what have you. And it all kind of becomes contextual for that action that agent is about to do and how it retrieves finds that meaningful information along with memory to get the job done and get the job done with the highest accuracy that you use these desires. So that's where yes, memory is one of the very, very important part of that story. And the broader story for us is the memory plus the context that makes agents so successful and really, really good in the production. Yeah.
Read the other 17 minutes of this episode here.
$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.
Not what you expected? Full refund within 14 days.
Thousands of transcripts fetched by people building searchable podcast archives
Fetch transcripts with code
The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000800031207