Gupshup Built a Super Agent: What Beerud Sheth Taught Me About What Is Under the Hood

Gupshup Built a Super Agent: What Beerud Sheth Taught Me About What Is Under the Hood

Which AI model should we use? It is a question I hear repeatedly. But my conversation with Beerud Sheth on Dilse Omni Talks brought me back to a more important question: does the AI have the context and connections it needs to get work done?

Beerud is the Co-Founder and CEO of Gupshup, a global conversational AI platform for marketing, commerce, and customer support. An IIT Bombay and MIT alumnus, he previously co-founded Elance, which later became Upwork. At Gupshup, he has led the journey from messaging to conversational AI.

I have known Gupshup as a customer and through a consulting engagement. That history made this conversation especially interesting. I wanted to understand what its Super Agent actually does, beyond the excitement around AI agents.

What followed felt like a masterclass in connecting intelligence, enterprise knowledge, and execution.


Table of Contents

  • 1. Stop Treating Model Selection as the Whole AI Strategy

  • 2. Under the Hood: The Brain Needs a Nervous System

  • 3. The Real Test Is a Workflow Across Departments

  • 4. AI Marketing Is More Than Faster Campaign Production

  • 5. Easier Execution Makes Better Thinking More Valuable

  • 6. Gupshup’s Technology Changed. Its Customer Mission Did Not.

  • 7. A Useful Bot Needs Context, and a Failed Bot Needs an Exit

  • My Takeaway: Build Around the Work, Not Just the Model

1. Stop Treating Model Selection as the Whole AI Strategy

Early in our conversation, I drew a parallel with Google’s mission of organizing information and making it accessible. Public AI models have made internet knowledge easier to access. But somebody inside the enterprise still has to organize the company’s own information and make it accessible to those models.

Your customer records, finance data, HR systems, policies, and campaign history do not automatically become useful just because you have access to a capable model.

Beerud shared an experience from inside Gupshup. When he encouraged employees to experiment with public models, some found them interesting but struggled to see how they helped with everyday work. Once those models were connected to internal tools and company context, the value became much more obvious.

“The incremental value actually comes from that context.”

Beerud Sheth

My takeaway was simple: model capability matters, but enterprise usefulness depends on what the model knows about your business and what it can access. That distinction also runs through our broader discussion on why the future of marketing is conversational.

2. Under the Hood: The Brain Needs a Nervous System

Beerud described Super Agent as a harness. That might sound technical, but his analogy made it very accessible.

The AI model is the brain. The harness is the central nervous system. The brain can reason about what should happen. The nervous system connects that reasoning to the tools and actions required to accomplish it.

Take a request to research a company. The system first needs a plan: identify competitors, understand the market, gather relevant information, and combine the findings. Then it must execute those steps in sequence. If something fails, it may need to revisit an earlier step before continuing.

The harness coordinates that process. It is not simply passing one question to a model and returning one answer.

A Simple Framework for Understanding Super Agent

I would summarize Beerud’s explanation through this flow. This is my synthesis of the architecture he described, rather than a named Gupshup framework.

Business request → Plan → Retrieve context → Execute steps → Revisit failures → Deliver the result

  • Intelligence: The model reasons about the request.

  • Context: Connected systems provide relevant business information.

  • Orchestration: The harness coordinates tools, models, and the sequence of actions.

  • Outcome: The system completes a workflow, not just a response.

That last distinction is critical. An answer might explain how to follow up on outstanding invoices. A connected agent can coordinate the information and actions involved in doing it.

3. The Real Test Is a Workflow Across Departments

Inside Gupshup, Beerud said Super Agent connects with HR, CRM, ERP, finance systems, and other tools. Its context also includes material such as policies, campaigns, customer information, and press releases.

The interesting part is not the number of integrations. It is what becomes possible when information from different departments can work together.

Beerud offered an example that made this tangible:

  1. Use the HR system to identify the salespeople.

  2. Use the CRM to retrieve customer information.

  3. Use finance data to identify outstanding invoices.

  4. Draft and send customer emails.

  5. Compile a report and send it to the CFO.

For a human team, that involves several systems, handoffs, and checks. The agent’s role is to coordinate the sequence using context from each department.

He also described customer queries that require looking up information across three or four tools. In those cases, he said responses that previously took hours could be prepared in minutes. Sales teams use the system for custom pitch decks and customer communications too.

For me, this widened the conversation considerably. We often discuss conversational AI as something businesses deploy for customers. There is an equally important opportunity in the productivity of the people serving those customers.

4. AI Marketing Is More Than Faster Campaign Production

That internal productivity discussion led naturally to marketing. Historically, marketers built bots and conversation platforms for customer interactions. Now, agents can also help the marketer operate campaigns.

Beerud described customer-facing capabilities such as creating multiple templates, launching a campaign on another channel, setting up an agent, or adapting a WhatsApp agent for voice.

But his bigger point was not about completing the same checklist faster. It was about expanding what a marketing team could reasonably attempt.

A human-led campaign might involve a handful of segments, two or three journeys, and daily optimization. Beerud contrasted that with the possibility of AI coordinating thousands or even millions of segments, individualized copy, thousands of journeys, and optimization every hour or minute where needed.

These were illustrations of potential scale, not a promise that every campaign needs a million segments.

The Campaign Capability Ladder

Here is how I organize that shift:

  • Segmentation: From a few broad groups to much finer customer distinctions.

  • Messaging: From a limited set of variants to more contextualized communication.

  • Journeys: From a few predefined paths to many tailored journeys.

  • Experimentation: From a small number of tests to massively parallel A/B testing.

  • Optimization: From periodic intervention to continuous adjustment.

Beerud also discussed Gupshup’s Campaign Hub in this context. The ambition is to bring precise targeting, parallel experimentation, and real-time optimization together.

The lesson is not simply “generate more content.” It is to reconsider which useful ideas you previously abandoned because the manual execution was too difficult.

5. Easier Execution Makes Better Thinking More Valuable

This was perhaps my favorite part of the conversation. If AI simplifies campaign execution, does the marketer become less important?

Beerud argued the opposite. Marketing has always required thoughtful decisions about whom to target, what to say, and how to communicate. Yet so much time gets consumed by execution that the thinking gets squeezed.

“The thinking part of it is going to be far more critical.”

Beerud Sheth

His argument was that these tools could magnify the difference between good and great marketers. When execution becomes easier, a great marketer can translate more of their judgment and creativity into action.

I see this as a useful challenge for marketing leaders: do not spend all the time saved by AI producing more variations. Spend some of it improving the customer understanding and strategic choices behind those variations.

6. Gupshup’s Technology Changed. Its Customer Mission Did Not.

Having seen Gupshup evolve, I asked Beerud how he understood the progression from messaging company to platform company to AI company.

His response was grounded: the enterprise customer still wants to engage consumers. The technology enabling that engagement has changed.

He illustrated this through a delayed flight:

  • SMS: Inform the customer that the flight is delayed.

  • WhatsApp: Deliver the update with richer visual content.

  • Conversational AI: Help the customer explore rebooking, rescheduling, cancellation, upgrades, refunds, or a voucher.

The shift is from delivering information to helping someone resolve the situation. That is what makes conversational commerce and customer engagement more than another messaging format.

Beerud’s concern was that enterprise adoption still trails the capabilities already available.

7. A Useful Bot Needs Context, and a Failed Bot Needs an Exit

When Beerud asked me for examples of customer-facing AI, I shared one frustrating experience and one promising one.

The frustrating experience involved recovering access to Azure. I found myself moving through automated support instructions without resolving the problem. After a month of going in circles, I gave up.

My lesson was straightforward: automation must not become an endless loop. Customers need an exit and access to a human when the automated journey fails.

My more positive example was Amazon Rufus. What interested me was the context available around the conversation: product listings, reviews, ratings, and pricing information.

I described using price history to understand whether a purchase represented reasonable value. I also discussed a contextual shopping request, such as finding a gift for a son heading to college within a budget of 10,000. Follow-up questions about interests can make recommendations more useful.

Beerud pushed the idea further: what if the agent could monitor a product and buy it when the price fell below a specified threshold? He presented that as a next step, not an established capability in our discussion.

That brought us back to the central lesson: relevant context makes the conversation useful; connected execution can make it actionable.

My Takeaway: Build Around the Work, Not Just the Model

I came away with three questions for any business exploring AI agents: What workflow matters? What context does it require? What systems and actions must be coordinated to complete it?

Those questions move the discussion beyond impressive answers toward practical business value. For omnichannel leaders, that means improving both sides of engagement: the customer’s experience and the team’s ability to deliver it.

I am Saurabh Agrawal and we come with a new episode on Dilse omni talks every fortnight and cover different aspect of omnichannel with amazing speakers.


Latest Episodes

dilse omni talks nilesh patel

The Lead Management Playbook

Nilesh Patel, CEO of LeadSquared talks about the funnels, customer journeys and the lead management.

dilse omni with swapnika nag

Decoding Personal AI Agents

On this episode of Dilse Omni Talks, Saurabh sat down with Swapnika to talk about what happens when you stop treating WhatsApp as a messaging app and start treating it as the operating system your entire business already runs on.

dilse omni talks with aditya singh

Funding The Next Unicorns

On this episode of Dilse Omni Talks, we covered how to read India's consumer market, what actually happens in a pre-seed pitch room, and why networking is the one compounding skill most founders radically underinvest in.

dilse omni talks with gupshup founder

Future of Marketing is Conversational

Gupshup founder Beerud Sheth on why the click is dying, how conversational AI is rewriting marketing, and why context, not intelligence, is the real moat.

Inside Arjun Vaidya's D2C Playbook

Arjun Vaidya on building, exiting, teaching, and investing, when to go offline, founder branding, and life as an operator turned VC.

View More

Latest Episodes

dilse omni talks nilesh patel

The Lead Management Playbook

Nilesh Patel, CEO of LeadSquared talks about the funnels, customer journeys and the lead management.

dilse omni with swapnika nag

Decoding Personal AI Agents

On this episode of Dilse Omni Talks, Saurabh sat down with Swapnika to talk about what happens when you stop treating WhatsApp as a messaging app and start treating it as the operating system your entire business already runs on.

dilse omni talks with aditya singh

Funding The Next Unicorns

On this episode of Dilse Omni Talks, we covered how to read India's consumer market, what actually happens in a pre-seed pitch room, and why networking is the one compounding skill most founders radically underinvest in.

dilse omni talks with gupshup founder

Future of Marketing is Conversational

Gupshup founder Beerud Sheth on why the click is dying, how conversational AI is rewriting marketing, and why context, not intelligence, is the real moat.

Inside Arjun Vaidya's D2C Playbook

Arjun Vaidya on building, exiting, teaching, and investing, when to go offline, founder branding, and life as an operator turned VC.

View More

Latest Episodes

dilse omni talks nilesh patel

The Lead Management Playbook

Nilesh Patel, CEO of LeadSquared talks about the funnels, customer journeys and the lead management.

dilse omni with swapnika nag

Decoding Personal AI Agents

On this episode of Dilse Omni Talks, Saurabh sat down with Swapnika to talk about what happens when you stop treating WhatsApp as a messaging app and start treating it as the operating system your entire business already runs on.

dilse omni talks with aditya singh

Funding The Next Unicorns

On this episode of Dilse Omni Talks, we covered how to read India's consumer market, what actually happens in a pre-seed pitch room, and why networking is the one compounding skill most founders radically underinvest in.

dilse omni talks with gupshup founder

Future of Marketing is Conversational

Gupshup founder Beerud Sheth on why the click is dying, how conversational AI is rewriting marketing, and why context, not intelligence, is the real moat.

Inside Arjun Vaidya's D2C Playbook

Arjun Vaidya on building, exiting, teaching, and investing, when to go offline, founder branding, and life as an operator turned VC.

View More

Latest Blogs

Dilse Omni Talks — Omnichannel Podcast by Saurabh Agrawal

Talks with Saurabh Agrawal

Dilse Omni Talks — Omnichannel Podcast by Saurabh Agrawal

Talks with Saurabh Agrawal