🤖Agent Swarms
Avanika Narayan, Cofounder/AI Lead
Agents all the way down
Ten or change years ago when my high school computer science teacher asked about my end-of-year project, I had an immediate answer. Fresh from watching Iron Man — arguably the Marvel Cinematic Universe's finest — I glibly responded: "I want to build Jarvis." Needless to say, my teacher chuckled, tempered my expectations, and set me along a path that was much more attainable.
Fast foward a few years, both my personal ambition and the world look a lot different. A deluge of deep technical work exploring frontier models for enterprise automation and augmentation has convinced me that my teenage dream of personal AI assistants (or agents, as we now call them) wasn't just fantasy—it's become our reality. Across enterprises, these agents are already supporting specialists from HR professionals to software engineers to data architects. In many respects, the vision has been signed, sealed and delivered. While it’s far from perfect, we are getting there. I am an optimist, bear with me 🙂.
At Rox, we wanted more. We wanted to push beyond individual assistance. We asked a bigger question: Can we catalyze agents to not only supercharge inviduals, but also entire organizations?
Unlike personal users, enterprise organizations demand agents that can adapt to their unique operational preferences spanning both internal and external data (i.e., configurable), respond to events around the clock for both human initiated tasks and automated processes (i.e., always-on), and tackle specific workflows tailored to their organizational needs (i.e., task-oriented).
Over the past 8 months, we worked with the best revenue teams and sellers in the world (s/o Nick King, Matt Murphy and others) to see whether there was an agent solution that satistified these desiderata. The key lies in what we call Agent Swarms: coordinated agents that operate with organizational, account and seller-level context, maintaining full visibility over public and private account data. These swarms can be configured, directed, and orchestrated to execute real revenue workflows.
Now, alongside Ishan, Diogo, Shriram, and the incredible Rox team, I'm stoked to be launching Rox: an AI-native revenue platform powered by Agent Swarms, designed to supercharge the world's best revenue teams and sellers.
Why now?
Needless to say, agents are tranforming the future work. Enterprises are adopting AI like wildfire. At an individual level, traditional workflows are changing. From software engineering to legal to HR to finance, the best are adoping agents to be faster, better and more efficient.
It’s clear that adopting agents is necessary. As the economist Richard Baldwin said, “AI won't take your job. It's somebody using AI that will take your job.” At the enterprise level, the stakes are similar! The best CIOs and senior leaders are expanding their investment in AI. This leaves organizations — including revenue teams — in an “innovate or die” regime.
So what happens if revenue teams were to actually implement Agent Swarms for their organizations?
For sellers, the equation is simple: less time on rote day-to-day tasks (research, ICP scans, scheduling, and more) means more time leveraging their greatest asset — their ability to connect with customers and nurture valuable relationships.
At the organizational level, the impact cascades: when individual efficiencies compound across teams, sales efforts become exponentially more coordinated and potent. Teams achieve what was once impossible, moving faster through sales cycles, reaching higher deal values, and building stronger account relationships.
Where did we start?
Back in February when we started Rox, like many of the GenAI native companies around us, we concieved of what we believed was the perfect agentic experience for sellers: an “ask me anything” open ended chat interface.
Our early work with top sellers at Lacework, Elastic and others, revealed a crucial insight: our initial interface missed the mark. It wasn’t personalized to the organization, the product and the sellers' preferences and data. It was limited to seller-initiated queries — capping both the scope of agent actions and their time horizon for engagement. It didn’t help sellers accomplish any of the tasks they cared about. Fundamentally, the experience just wasn’t useful.
Our experiences lead to three key learnings. Namely, that our agents need to be:
Configurable: across the seller, account and organizational preferences and data
Always-on: completing one-off user initiated tasks (i.e., account specific research) and ongoing event-initiated tasks (i.e., news monitoring, job changes)
Task-oriented: meet sellers where they are at by doing real seller tasks (i.e., ICP scans, PG planning or people and team search, email generation).
Where are we now?
Our learnings helped us concieve of the Rox Agent Swarm: a fleet of AI agents that operates on top of Rox’s system of record. Each agent is paired with a specific customer and account executive, continuously monitoring and managing relevant information. Rox Agent Swarms can be configured, directed and orchestrated.
Agents have access to a workbench composed of models, tools and verifiers. The swarm stack has 3 layers:
Data: agents extract, clean and integrate data, unifying both public and private sources.
Intelligence: agents monitor real-time events and execute automated workflows as specified by users and organizations.
Interface: agents communicate with users through text and audio, handling a wide range of tasks from outreach to in-depth research and ICP scans.
Broadly, Rox Agent Swarms help with five core sets of sales rep tasks: Discovery, Research, Account Monitoring, Planning, and Engagement. Within each task category, we have built a core set of agentic workflows that are deeply rooted in best of breed sales practices.
What is next?
We're excited for the Rox Agent Swarm to help every seller become a top 1% seller!
If this future excites you, please join Rox and build the next generation of Agent Swarms, supercharging the world's best revenue teams and sellers!!!
Last updated
