Thirty years before AI agents became a business conversation, Beerud Sheth was researching how software could learn our interests and build a personalised newspaper. When I came across that work while preparing for our conversation on Dilse Omni Talks, I had to ask: has the future he imagined finally arrived?
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. His journey connects three ideas that once seemed ahead of their time: autonomous agents, remote work, and messaging as an interface for digital services.
What followed felt like a masterclass in technology, timing, and building businesses. My biggest learning was not about predicting the next breakthrough. It was about recognising a lasting human need, then working through the gap between a compelling idea and widespread adoption.
Table of Contents
1. AI Agents: An Old Vision Reaches a New Threshold
2. Elance: Technology Adoption Is Also a Trust Problem
3. Being Early Does Not Replace Product-Market Fit
4. Gupshup Began with Access, Not AI
5. From SMS to Conversational AI: The Interface Evolves
The Lesson I Took Away
1. AI Agents: An Old Vision Reaches a New Threshold
At MIT’s Media Lab, Beerud was among the earliest graduate students in a newly formed Autonomous Agents Group. His research focused on personalised news filtering: creating a newspaper around an individual’s interests that could evolve over time.
The idea is easy to understand today. Instead of giving everyone the same collection of information, a system would adapt what it selected to the person using it. But the technology available then was far more limited. Beerud recalled building strong demonstrations for that period, even though the broader vision was ahead of what the tools could deliver.
“It feels a full circle moment for me.”
Beerud Sheth
That was how he described the connection between his early research and today’s AI progress. Personalised recommendations, language capabilities, and voice are now making parts of that vision practical.
Yet his answer was not that the work is finished. He believes we have crossed an important threshold in solving human problems, while greater agent autonomy still lies ahead.
Meaningful progress and a long runway can exist together. That distinction matters. Businesses do not need to dismiss AI because it is incomplete, or assume every ambitious promise is already achievable. The useful question is which customer problem the technology can solve well now.
A simple way to understand the early agent idea
I found it helpful to reduce Beerud’s news-filtering example to a three-part flow. This is my summary of the concept, rather than a named framework he presented:
Personal interests: Begin with what matters to an individual.
Relevant selection: Filter information into a personalised experience.
Continued adaptation: Let that experience evolve as interests change.
The central ambition was not simply to produce more content. It was to make information more useful to a particular person. That is a valuable starting point for anyone thinking about AI-powered customer engagement.
2. Elance: Technology Adoption Is Also a Trust Problem
After graduate school and a period working on Wall Street, Beerud co-founded Elance in 1998. The name played on “freelance,” bringing the idea of finding and working with independent professionals online.
The proposition was to connect people with freelancers around the world. Today, that sounds familiar. At the time, it challenged a basic assumption about work: that you needed to meet someone physically before you could trust them professionally.
Beerud remembered the questions people asked. Could they trust someone they had never met? Would language become a barrier? What if the person disappeared with their money?
Those were not minor objections. They were fundamental doubts about whether the model could work.
For me, this was an important turn in the conversation. We often describe innovation as a technical achievement. But a product can be technically possible long before people are comfortable changing their behaviour.
Beerud reflected that remote work took almost two decades, and perhaps a pandemic, to become mainstream conventional wisdom. The journey from Elance to Upwork illustrates the length of that adoption curve. Elance was founded in 1998; Upwork went public in 2018.
He also clarified his own involvement: roughly eight years in an operating role, followed by another six or seven on the board. At the time of our conversation, he was a shareholder.
My takeaway: do not confuse resistance to a new behaviour with proof that the underlying need is weak. But equally, a strong vision does not remove the need to address the practical concerns stopping people from participating.
3. Being Early Does Not Replace Product-Market Fit
I asked Beerud about the difference between the cutting edge and the bleeding edge. What does it take to build when the market has not yet caught up?
His answer brought us back to something more grounded than foresight: product-market fit.
“Even if the idea is right, there’s lots of little details that make a big difference.”
Beerud Sheth
That line stayed with me. A founder may identify the direction correctly, but the service still needs to work in the details that matter to customers.
Beerud distinguished between two phases. In the startup phase, the business works to find product-market fit and resolve difficult foundational problems. In the scale-up phase, it grows and expands what is working.
He also gently corrected my suggestion that he had left before the major growth at Elance. His point was that the team had already solved many hard problems and established a model that continued to scale. The later journey still involved substantial work, but it was not a story of waiting passively for the market.
The startup-to-scale-up framework
Here is how I would translate his distinction into a practical sequence:
Identify the need. What useful outcome are you making possible?
Resolve the details. What stops customers from using the service successfully?
Establish product-market fit. Does the offering work for the market it serves?
Scale what works. Expand the model while continuing to improve it.
This is a useful counterweight to AI excitement. A new capability may open the door, but it does not automatically create a business. Customer value still has to be established.
4. Gupshup Began with Access, Not AI
When we moved to Gupshup, I asked about the name. In Hindi, “gupshup” means conversation or chitchat. The name captures a thesis that has remained remarkably consistent even as the technology has changed.
The early opportunity came from the mobile revolution and the next billion people becoming digitally connected. Many were using feature phones. They were not entering digital life through the web and email in the way users in Western markets had.
That changed the product question. How could digital services reach people through the devices and channels actually available to them?
For Beerud and his team, mobile messaging and voice were the practical routes. SMS became the foundation for an early social network, broadly similar in concept to Twitter, but operating entirely through text messages.
People could follow others and receive updates through a publish-subscribe model. They did not need a conventional web experience to participate.
Beerud said the network grew to about 70 million users in India during the 2007 to 2010 period, when Facebook and Twitter each had fewer than a million users in the country.
The scale surprised me. More importantly, it demonstrated the value of designing around actual access rather than assuming everyone would follow the same digital journey.
The channel was not an afterthought. It made participation possible.
5. From SMS to Conversational AI: The Interface Evolves
Gupshup’s early SMS network and its conversational AI business can look like very different products. Beerud’s explanation revealed the continuity.
The underlying idea was to connect people with digital services through messaging and voice. Over time, that extended to richer interactions between businesses and consumers. AI is now helping bring more of that original ambition to life.
That is why the name still fits. The technology has evolved, but conversation remains central.
For omnichannel leaders, I see a useful lesson here: start with how customers can comfortably interact, not only with the interface the business prefers to build. This connects naturally with our exploration of how conversational commerce is changing customer engagement.
My synthesis: need, access, trust, fit
Across Beerud’s three journeys, I came away with four questions. These are my synthesis of the discussion, not a formal model attributed to him:
Need: What lasting human problem are we solving?
Access: Can people use the technology through channels available to them?
Trust: What makes them hesitant to adopt the new behaviour?
Fit: Which details must work before the business can scale?
News-filtering agents addressed relevance. Elance confronted trust in remote work. Gupshup built around mobile access. Across all three, turning the vision into a useful service required more than being early.
The Lesson I Took Away
Beerud joked that anything he predicted would become a fact in 20 years. Behind the humour was a serious lesson about patience, execution, and technological progress.
Being early is neither a guarantee of success nor a reason to abandon a good idea. The work is to keep solving the customer problem as capabilities and behaviour evolve.
I began by asking whether AI agents had arrived. I left with a better question: which enduring customer need can we finally serve well with the technology available today?
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.


















