Image generated by Deeptech Times using Google Gemini
As AI agents become increasingly capable of performing knowledge work, a critical question remains: how do they acquire the context that humans take for granted?
Nathan Xu, CEO and co-founder of Plaud, argues that the future of AI will depend less on raw intelligence and more on understanding people, conversations and intent. Plaud, best known for a small recording device that turns conversations into structured notes, now has more than 2 million users globally and is expanding its Singapore office into a larger APAC base.
In this interview with Deeptech Times, Xu discusses the rise of AI-powered memory, wearable agents and why context may be the missing layer in autonomous work.
What is your background?
I have spent most of my career as a serial entrepreneur. Plaud is actually my fourth startup, and the previous three did not succeed. I launched my first company before graduating from university, and each time I started a business, I invested virtually all of my personal savings into it. If the company failed, I faced the consequences personally. That experience taught me resilience and conviction.
My professional background began in finance, specifically venture capital and private equity. However, during an exchange programme in Amsterdam in 2012, I took a course on the Internet of Things and became fascinated by technology’s ability to transform lives and create meaningful societal impact.
I’ve always been drawn to solving problems through optimisation. Whether in business or everyday life, I instinctively think about how systems, processes and experiences can be improved.
My first startup focused on education. We built a platform that aggregated internship opportunities from company websites and news sources, helping students discover openings at organisations they aspired to join. It was a relatively simple product, but it addressed a real challenge faced by millions of graduates seeking meaningful career opportunities.
You have said AI capability is already superb but adoption still lags because human intent and context are missing. What do you mean by that?
One of the clearest demonstrations of AI’s progress over the past year has been the emergence of coding agents. Software engineering was traditionally considered highly specialised work, yet today’s frontier models can already solve many engineering challenges autonomously. A user can describe an application in natural language and AI can assist in building it.
Now consider managerial decision making. In my role, I review information from product, engineering, legal and finance teams. I read documents, absorb facts, identify gaps in context, consult stakeholders, gather perspectives, reason through competing viewpoints and ultimately make decisions.
Increasingly, much of that process can be supported and eventually performed by AI systems. That is why I say AI capability is already extraordinary. The real question is not whether AI can do these things, but how many people today can actually harness that level of capability in their daily work.
What is the gap?
The challenge is that while AI appears capable of doing almost anything, it lacks a fundamental understanding of what an individual is trying to achieve and the context surrounding that objective.
In everyday work, we accomplish tasks through conversations with colleagues, customers and stakeholders. Those interactions contain critical information: the motivations, priorities, constraints and nuances that shape decisions.
Humans naturally possess that context but AI does not. The problem is that we rely heavily on memory to retain these details, and human memory is inherently imperfect. Without that context, even highly capable AI systems cannot operate at their full potential.
How does Plaud try to give AI that missing context?
We believe AI should understand not only what is being discussed but also who is participating in the conversation. That has been a major area of focus for Plaud.
Speaker identification alone is a highly complex challenge. Perfect accuracy remains difficult, but there are multiple ways to bridge that gap.
Imagine a system that understands an organisation’s directory: each person’s role, responsibilities, current projects and voice profile. In a business conversation, AI could automatically recognise participants and understand their organisational context.
Once AI knows who is speaking and how they relate to the broader business environment, its ability to provide meaningful assistance becomes significantly more powerful. In effect, it creates an organisational memory that is far more complete and reliable than human recall alone.
Is that possible today?
Significant progress is being made in this area. While the technology is not fully realised today, I believe it is achievable within the near future.
The primary challenge is privacy. Many people are understandably uncomfortable with having voice-print data stored in the cloud.
Our approach is to explore architectures where identity matching can occur securely, with sensitive voice data processed locally whenever possible. Once a user’s identity has been linked to a voice profile, the underlying biometric data does not necessarily need to be retained centrally.
From a technical perspective, these challenges are solvable and we are actively working on them.
Does that mean Plaud has to be always-on?
We prefer the term “always ready” rather than “always on”.
Being always ready means the device can respond intelligently when needed, through multiple activation methods suited to different situations. It does not imply continuous recording or data capture.
The goal is to make AI accessible at the moment it is needed, without requiring users to consciously initiate every interaction.
What kind of device does that point to?
Many people immediately assume smart glasses will become the dominant AI interface. I am not convinced that conclusion is as straightforward as it appears.
Smart glasses are typically associated with cameras, which introduce two important considerations. First, how do we ensure that people around the wearer feel comfortable and respected? Second, how do we transform the vast amount of visual data captured into meaningful value rather than overwhelming noise?
I believe smart glasses have tremendous long-term potential. However, product adoption depends heavily on timing and user readiness.
Plaud has grown from zero to more than two million users in just three years, while many companies pursuing AI wearables have struggled to gain traction despite years of development. That tells us the choice of interface matters.
Over time, visual capabilities may become part of Plaud’s product roadmap. But any introduction of cameras must be accompanied by strong safeguards, clear user benefits and a high degree of trust from both users and those around them.
More broadly, I believe the future interface between humans and AI agents will be wearable and always ready. It should be comfortable enough to wear throughout the day, capable of capturing context and intent, and allow people to interact with AI in the most natural way possible.
Today, interacting with AI typically requires opening an app or typing into a chat interface. Ultimately, conversation itself may become the most natural interface.
Where do you see Plaud being used most meaningfully?
Healthcare is undoubtedly one of the most impactful sectors for Plaud today. Sales and consulting are also among our strongest use cases.
Some of our most meaningful stories come from healthcare professionals and patients. I recently spoke with a professor at Stanford Medical School who also practises in a hospital setting. He shared the story of a 95-year-old former Nobel laureate who used Plaud because hearing challenges made it difficult to remember everything discussed during medical consultations.
The ability to accurately capture and revisit those conversations significantly improved her experience.
What does Plaud’s current user base look like?
A significant portion of our user base falls into what we internally describe as the “three-H” category: individuals with high knowledge density, who rely heavily on conversations to perform their work, and whose decisions carry significant impact. This includes executives, business owners and senior decision-makers.
Early adopters of Plaud were people whose work revolves around conversations. Senior leaders spend much of their day moving between meetings, absorbing information, evaluating ideas and making decisions. For them, Plaud provides an immediate and highly accessible AI assistant.
Beyond executives, our users include bankers, consultants, lawyers, product managers and a wide range of knowledge professionals.
The NotePin series has also gained traction among users who need hands-free solutions while working in dynamic environments, including wedding planners, field technicians, automotive sales professionals, real estate agents, insurance advisors, doctors, dentists and veterinarians.
Across these professions, conversations are often the primary vehicle for getting work done. Plaud helps capture those interactions, organise information, generate action items and integrate them into broader AI-powered workflows.
You call it “out-of-the-box AI”. What do you mean?
The idea is simple: users should not have to think about the technology. They should be able to take the device out of the box and begin benefiting from AI immediately.
Ultimately, the most effective technology becomes invisible. Many leading AI companies achieved their initial success by serving developers, which naturally led them to focus on technical sophistication and underlying architectures.
We take a different view. Users care less about the technology itself and more about accomplishing their objectives. The product should focus on the task, not the technology.
What role will Singapore play for Plaud?
We established our Singapore office in July 2025, and in less than a year the team has grown to nearly 100 employees.
Initially, we were attracted by Singapore’s strong talent pool in computer science, machine learning and data science, as well as its strategic position as a gateway to the broader APAC market.
As we expanded, however, we recognised the opportunity to build a truly regional AI company from Singapore.
Today, Singapore serves as our APAC headquarters and houses a broad range of functions, including AI agent research, subscription business operations, growth initiatives, user operations and regional marketing, sales and go-to-market activities.
How much bigger will the Singapore team become?
We expect the Singapore team to reach approximately 150 employees by the end of 2026.
While I do not have the exact figures in front of me, I believe roughly three-quarters of our current workforce in Singapore are local hires.
Building strong local teams has always been an intentional part of our strategy. People who understand local cultures, customer expectations and business environments are essential to building trust and driving sustainable growth.
From the outset, our philosophy has been to build locally for local markets. That approach has been a key factor behind our rapid expansion across the region.
