Anaiya Raisinghani, Senior Tech Evangelist for AI Startups and Ventures at MongoDB, kicks off the episode with a warm welcome to Ishan Mukherjee, co-founder and CEO of ROX. The brief intro frames the conversation ahead: a deep dive into how ROX is using AI agents to autonomously manage the revenue cycle, and the technical architecture choices — including MongoDB — that make it possible. In just a few seconds, the listener is oriented to who is speaking and why this conversation matters for anyone building or buying enterprise software in the AI era.

Ishan Mukherjee grew up in India as a robotics major before moving to the United States in 2009 and graduating from MIT. He joined Kiva Systems as an early employee — a company that would eventually become Amazon Robotics, the Roomba-like warehouse robot fleet that powers Amazon's fulfillment operations. ^{[1] — Ishan Mukherjee "Pixie Labs acquired by New Relic in 2020: Ishan's Kubernetes infrastructure company Pixie Labs was acquired by software monitoring company …" 01:40 Play} After Amazon's acquisition, he relocated to the Bay Area in 2013 and joined the early team that built what became Apple's knowledge graph — the system that answers factual queries like 'How old is Steph Curry?' on Apple devices. His first independent venture was Pixie Labs, a Kubernetes infrastructure company that was acquired by software monitoring firm New Relic in 2020. He then served as Chief Growth Officer at New Relic, overseeing product-led growth, marketing, and high-velocity sales, until that company was sold to private equity in 2023. On Valentine's Day 2024, he and co-founder Sriram launched ROX — the company that now sits at the center of this conversation.

Pressed to describe ROX for first-time listeners, Ishan delivers a crisp, confident answer: ROX builds AI agents that own the entire revenue cycle without human involvement. That means agents conducting customer research, reaching out via email, engaging in conversations, managing contracts, and effectively replacing everything a human would previously have done inside a CRM. ^{[1] — Ishan Mukherjee "ROX isn't a CRM upgrade — it's a replacement. Its AI agents autonomously handle customer research, outreach, engagement, and contracts, doi…" 02:23 Play} The founding insight, he explains, came from observing in late 2023 — before agents were a household term — that AI could take over the mundane, data-intensive tasks that consume a salesperson's day, freeing humans to do what they're actually best at: building genuine relationships. The ambition is not incremental CRM improvement but wholesale replacement of a category.

When asked what distinguishes ROX from legacy CRM tools, Ishan offers a framework that's both simple and damning for incumbents: Siebel was great on-premise; Salesforce is great as SaaS; ROX is an agent product — and the differences are architectural, not incremental. ^{[1] — Ishan Mukherjee "Every major CRM shift has redefined who does the work. Siebel put it on-premise; Salesforce moved it to the cloud; ROX removes the human en…" 03:33 Play} The first key distinction is that agents do the work. In SaaS, a human feeds data into the system; with ROX, the agents conduct research, send outreach, run meetings, and manage contracts autonomously. The second distinction is even more structural: ROX's agents run on top of a customer's internal data warehouse, not on top of the CRM. Most of what you actually need to know about a customer — their business context, past interactions, internal signals — doesn't live in Salesforce. It lives in a data warehouse and a constellation of other systems. That's why ROX coined the term 'warehouse-native revenue agents,' and it's also what drove the architectural decisions that eventually led them to MongoDB.

Ishan walks through ROX's technical architecture in concrete terms, framing it as a three-layer cake. The bottom layer is the context system: a knowledge graph that aggregates all available data about customers, relationships, and business signals, governed by enterprise-grade access controls. On top of that sits what Ishan says ROX pioneered — the concept of agent swarms, an orchestration system where thousands to potentially millions of pre-built agents communicate and coordinate with each other to get work done. ^{[1] — Ishan Mukherjee "Think of ROX as a three-layer cake: a knowledge graph and governance system at the base, an agent swarms orchestration layer in the middle …" 05:05 Play} The top layer is the application surface: a web app, a mobile app, and integrations into platforms like ChatGPT, Microsoft Copilot, and Slack. He connects this architecture directly to MongoDB: the context system deals overwhelmingly with unstructured data — emails, call transcripts, meeting notes, news — the kind of data that traditional relational or CRM databases were never designed to handle at scale, and that document stores like MongoDB are purpose-built for.

The decision to use DynamoDB at launch was almost reflexive: both Ishan and Sriram, ROX's CTO and co-founder, came from Amazon and had deep familiarity with AWS infrastructure. For the 0-to-1 phase, that bias was fine — DynamoDB worked. But as ROX hit genuine enterprise scale, the team reached a point where MongoDB's feature set, scalability, and enterprise-grade security characteristics were meaningfully superior for their specific workload. ^{[1] — Ishan Mukherjee "ROX started on AWS DynamoDB — the founders came from Amazon, so the choice was instinctive. But as scale hit, MongoDB's feature set and sec…" 06:44 Play} The migration happened in late 2024, and MongoDB now serves as ROX's core document store, ingesting and serving the full range of unstructured data sources that feed into the knowledge graph. Ishan is careful to note this is an evolving relationship — MongoDB is a fast-growing part of a larger data infrastructure, not the entire stack.

This is the section where Ishan's thesis sharpens into a single, shareable idea: the AI model is a commodity. OpenAI, Anthropic, Google — they're all building world-class models and selling access to everyone. The companies that win the AI era won't win on model selection; they'll win on context. ^{[1] — Ishan Mukherjee "GPT, Claude, Gemini — they're available to everyone. The actual competitive moat is who can aggregate, activate, and make their data agent-…" 08:25 Play} The ability to pull together all the signals a customer leaves across emails, Slack, meetings, CRM notes, product usage data, news coverage, and LinkedIn activity — and then make that entire corpus immediately queryable by an agent — that is where enterprise value accumulates. It's also, Ishan argues, why systems designed to ingest and serve unstructured data at scale (like MongoDB) are going to become a fundamental layer of every serious AI company's stack, regardless of what they're building.

There's a philosophical debate in every engineering team about which technologies to adopt, and Ishan's framework cuts cleanly through it: separate the 0-to-1 question from the 1-to-scale question. ^{[1] — Ishan Mukherjee "Technology stack decisions aren't one-size-fits-all — they depend entirely on what phase you're in. Early stage needs developer-friendly AP…" 09:50 Play} In the 0-to-1 phase, what matters is developer experience — great APIs, no sales call required, easy onboarding, a strong community and brand. That's why ROX chose DynamoDB initially; it was familiar, fast to start with, and backed by AWS. The 1-to-scale phase asks different questions: can the architecture handle massive enterprise traffic spikes? Is query performance fast and reliable enough for real-time agent operations? Does the security model meet enterprise procurement requirements? MongoDB, Ishan says, is where ROX found its answers for that second phase — and he adds a note that MongoDB appears to be 'regaining its mojo' after a period when vector stores and other specialized systems were capturing developer attention.

When Anaiya asks how Ishan leads AI adoption internally, his answer is almost dismissive in the best way: at ROX, AI tool usage is second nature. There's no approval process, no controlled rollout — engineers use whatever tools help them move faster, including Cursor, Cognition, and Claude Code. ^{[1] — Ishan Mukherjee "At ROX, there's no AI tool approval process — engineers use Cursor, Cognition, Claude Code, whatever helps. The whole company runs on ROX i…" 11:20 Play} The distinction he draws is between engineers working on the platform and foundation layer (who think deeply before reaching for AI-generated code) versus those on the product and design side (who are exploring and experimenting constantly). But the most striking internal AI story is simpler: ROX runs its entire business on its own product. No Salesforce. No HubSpot. No external workflow tool. Just ROX, end to end — both as a CRM and as the team's general operating system.

Ishan's answer to 'what comes after agents?' is refreshingly honest: nobody knows what the world will look like in six months. But the fuzzy long-term view is becoming clearer. In the B2B world ROX operates in, every business and team will become dramatically more productive. The best performers — the best engineers, the best marketers, the best salespeople — will be 2 to 5 times more productive than they are today. ^{[1] — Ishan Mukherjee "Predicting AI six months out is nearly impossible — but the long-term direction is clear. The best performers will be 2 to 5x more producti…" 12:36 Play} That upside, however, comes with a hard corollary: those who don't reskill and aspire to become the best will fall behind. The transformation is already happening, and it will only accelerate. For ROX's own planning process, the implication is practical and immediate: monthly planning cycles are already obsolete. ROX runs on weekly targets and executes daily, because the window in which a monthly plan remains relevant has collapsed.

Wrapping up with advice for founders building AI products today, Ishan distils his company-building philosophy into one principle: be opinionated about the deep infrastructure (the architecture, the context system) and extremely iterative on everything closer to the user (the product, the experience). ^{[1] — Ishan Mukherjee "Build with strong convictions at the infrastructure layer and relentless flexibility at the product layer. That's Ishan's formula for navig…" 14:09 Play} That balance, he says, is the only viable way to navigate what he calls an 'insane transformation or supercycle.' The episode closes with a rapid-fire lightning round that reveals Ishan's personality as much as his views: one word for the future of sales — 'supercharged.' The most underrated part of building AI — 'the adrenaline rush.' Best city for founders — 'No question, San Francisco.' The conversation ends with Anaiya thanking Ishan for welcoming MongoDB into the ROX office, and Ishan signing off with a direct plug: 'Please use Rocks.'