Chetan Dube gave himself a couple of summers. In 1998, working at New York University’s Courant Institute, he thought the computing research he was pursuing could be extended toward machines that appeared to think. His adviser was less persuaded. Dube remembers being warned that even John McCarthy, a founding figure of artificial intelligence, had found the problem harder than expected. The summers accumulated. So did the companies.
There is something wonderfully human about underestimating the time required to reproduce humanity. A young researcher looks at a formidable intellectual problem and sees a manageable project. The calendar disagrees. Dube’s response was to keep working, taking the question out of the university and into the machinery of business: IT operations, customer conversations, service desks, and eventually the possibility of a software colleague.
He founded IPsoft in New York in 1998. Its conversational AI platform, Amelia, arrived in 2014; the company adopted her name in 2020. Today he is founder and CEO of Quant, where the old question has acquired a new job description. His intended digital employees should carry out work across business systems. Across these changes, the ambition has remained recognizable. Dube wants to give routine work to software and leave people with more room to think.
A bow tie at the service desk
Before considering the software employee, consider its creator. Dube’s public appearances often feature a bow tie and an air of formality that would be quite at home in a faculty common room. When futurist Mike Walsh met him over coffee in his New York office in 2016, the conversation moved between childhood recollections of New Delhi, Paris and London and explanations of neural networks. The itinerary and the mathematics belonged to the same conversation.
That mixture helps explain his appeal as an interview subject. He can talk about the construction of an intelligent system and then widen the frame to the construction of a company. Walsh’s discussion with him concerned an enterprise in which software could learn and anticipate customers’ needs. The destination was an organization with a different operating rhythm, rather than a clever answer displayed on a screen.
His academic background is part of the practical story. Before building IPsoft, Dube served as an assistant professor at NYU; his research concerned deterministic finite-state computing engines. The subject sounds remote from an irritated customer waiting for help. Yet moving between a defined state and an appropriate next action is a useful starting point for thinking about service. A business has rules, permissions and processes. An answer needs somewhere to go.
The long route from a question to a company
IPsoft began with automating IT work. By June 2019, an Everest Group company profile described offices in 15 countries and more than 550 clients. It listed partners including Deloitte, Accenture, NTT, NCR and Unisys. Those details give the early ambition a physical scale: offices, contracts, implementation teams and relationships with companies already responsible for other companies’ technology.
This matters because enterprise software has an unusually thorough way of questioning a theory. A demonstration can be charming. A deployment has to survive the customer’s existing systems, procedures and expectations. Dube’s career took shape inside that less glamorous examination room. The goal was intelligence, but the work passed through the everyday obligations of keeping a business running.
In 2011 he received Ernst & Young’s Entrepreneur of the Year recognition. His LinkedIn profile also lists patents involving the communicative behavior of virtual agents, including expressions and responses shaped by a user’s affect. Even in the patent language, the concern extends beyond answering correctly. People bring tone, mood and personality to a conversation. A digital counterpart has to decide what to do with them.
The tension is familiar to anyone who has dealt with customer service. A perfectly accurate response can still leave a person dissatisfied. A warm response can leave the problem untouched. Dube’s work sits at the awkward intersection of those two failures, where a system is asked both to understand the person and to finish the task.
Why the employee was called Amelia
The name was a deliberate choice. Dube named the platform after Amelia Earhart, admiring her willingness to challenge accepted horizons. It gave a piece of enterprise technology an association with exploration, and it gave him a way to describe the research as an attempt to cross a boundary. The platform eventually lent its name to the whole company.
In a 2019 conversation, he described Amelia as learning different kinds of work: banking, insurance, IT operations and call-center support. His language made the training sound almost collegiate, with the same underlying platform acquiring different professional skills. This was the software worker imagined through human institutions: learn the rules, acquire competence, take up a role.
The choice of a human name also puts a question in the room. How much of a colleague can be represented by an interface? A name makes a system easier to address, but expectations follow it through the door. If customers speak to Amelia as though she understands them, they are likely to want the same continuity and competence they would expect from a person. Naming the employee is the easy part of the employment process.
He was talking about the office before the chatbot boom
In a 2012 essay, Dube began with textile manufacturing and the Spinning Jenny before turning toward office automation. It is a revealing choice of historical company. He was placing software within a much older argument about machines, productivity and the organization of work. The office, in his telling, was available for the kind of change that manufacturing had already experienced.
The essay appeared more than a decade before conversational AI became a routine subject at dinner. Its title asked who was running the office, with a nod to Siri. Dube was already considering what happened when software participated in operations, rather than merely storing the paperwork. His vocabulary has changed since then, but the office keeps turning up.
By April 2022, in a discussion with Paolo Baldesi, he was arguing for businesses to adopt a workforce combining people and AI. He presented automation as a way to address labor and talent gaps, and to remove repetitive work from employees’ days. The appeal had two audiences: an executive worried about capacity and a worker tired of spending attention on the same small chores.
When he discussed ChatGPT with Nidhi Arora in February 2023, the emphasis included its strengths and limits and the different purposes served by generative and deterministic technology. For someone who had spent years selling conversational systems, the new fascination with generated language raised a practical question. Which parts of that fluency could be useful inside a process that needed dependable outcomes?
A new company, the same unfinished assignment
SoundHound announced its acquisition of Amelia in August 2024. The combination brought Amelia’s enterprise customer-service relationships into SoundHound’s voice and conversational AI business. Finance, insurance and retail were among the sectors involved. For Amelia, it was a change of corporate home after years of developing and deploying conversational technology.
Dube’s next venture, Quant, launched in September 2024. The company identifies him as CEO and chairman. Its focus is agentic AI: systems intended to carry out work with a degree of autonomy. The progression from IT automation to conversations to digital employees reads like successive attempts to make the original research question useful enough for a company to buy.
THE WORK HE WANTS SOFTWARE TO DO
A conceptual illustration of Dube’s stated ambition for digital employees, rather than a performance benchmark.
The word “employee” is doing serious work here. It implies a role, responsibilities and an outcome someone can inspect. A tool can sit unused in a drawer; an employee is expected to contribute. Calling software a worker invites the questions that come with managing work, including who authorizes an action and who takes responsibility when it goes wrong.
Quant’s leadership also includes Martijn Gribnau as chief customer success officer. Dube’s professional circle continues to connect technical development with people who have run institutions and dealt with the consequences of operational decisions. The customer-service setting makes that connection unavoidable. A conversation becomes part of a real transaction.
The suitcase test
In July 2026, Dube offered a travel example grounded in his own experience. He recalled a San Francisco-to-New York flight delayed by fog the previous year. Weather, he observed, could not be controlled; the response to it could. Anyone who has watched an airport queue grow can recognize the appeal of that distinction without studying a single neural network.
His proposed improvement involved agents able to handle many passengers’ requests at once and arrange alternatives. Elsewhere in the same interview, he contrasted instructions for reporting a missing bag with actually locating and delivering it. These were illustrations of what he wanted agentic systems to accomplish. Their attraction lies in the modesty of the customer’s desire: the suitcase should turn up, and the next flight should exist.
It is a useful way to understand his persistence. The philosophical question concerns a thinking machine. The customer’s question concerns a bag. Between them lies an immense amount of unromantic work: connecting systems, obtaining permission and carrying a request through to completion. Dube’s preferred future has to pass through that baggage hall.
Who manages the new colleague?
His February 2026 writing argued that leaders must redesign jobs as AI takes on more execution. He called for clearer responsibility, different ways of assessing people’s contribution and time to review the systems’ work. His proposed human role places greater weight on judgment, exceptions and care for the customer, with someone remaining accountable for outcomes.
That adds an organizational obligation to the technical ambition. If routine tasks disappear, a job description does not automatically rewrite itself. Managers have to decide what the person should do next and how that work will be valued. Otherwise, the promised extra room for creativity is just an empty space on somebody’s calendar.
“Dad, are you going to be a robodad?”
Dube recalling his son’s question, 2025
Dube has also advocated taxing revenue generated through digital workers to help fund universal basic income. It is a proposal, with the political questions a proposal deserves. His son’s question about a “robodad” gives the same subject a smaller, sharper form. Building a substitute for human work means considering what people want to keep human.
The early timetable proved optimistic. Dube kept the assignment. From a university conversation to an airport queue, his career has returned to the possibility that software could relieve people of work they would gladly hand over. The remaining question is what they gain in the exchange. A returned suitcase is a good beginning. A more interesting working day would be better still.