Skip to content

The CCS Interview

Sensay CEO: AI replicas should draft your email, not send it unseen

Dan Thomson, founder and CEO of Sensay, explains how the company builds AI replicas of people for knowledge management and day-to-day admin. He also covers trust, jobs and enterprise privacy. Recorded in October 2024, he said at the time that Sensay's email drafter was due at the end of that month.

Interview by Ashton Addison Recorded October 2024 6 min read

Key takeaways

  • Sensay creates AI digital replicas trained on a person's voice, video, and digital footprint to draft emails and manage communications on their behalf.
  • Voice and video training requires minimal data—approximately 30 seconds of audio and 2-3 minutes of video—while the more complex task is identifying which aspects of identity to replicate.
  • Human oversight features allow drafted messages to be reviewed before sending rather than automatically dispatched, addressing user comfort and accuracy concerns.
  • AI replicas can operate 24/7 across multiple languages, extending a person's availability and knowledge globally without being limited by their personal time constraints.
  • The uncanny valley effect causes hesitation around AI replicas despite their accuracy, though historical precedent exists with personal assistants and secretaries performing similar tasks.

Building a replica of yourself

Ashton Addison: What is Sensay focused on right now?▶ 00:22

Dan Thomson: We're building digital immortality: digital replicas that can be used in contemporary use cases for businesses and individuals. They work as a tool for knowledge management within organizations, for automations, and as a perfect personal assistant. Imagine a version of you that can pre-draft your emails, with your entire knowledge base drawn from your digital footprint, your previous messages and the documents you work from. Combined with a sufficiently advanced AI, it can reply, or at least draft replies, depending on your preference, to the questions we get asked every day. We're online only certain hours a day, and a lot of the time we're asleep or enjoying personal time. Replicas mean our teams and the people around us aren't limited by the availability of our own time. They act as an extension of ourselves that can share our knowledge, experiences and skills with people globally, 24/7, in any language.

Ashton: How would I teach an AI replica to be me and do my work?▶ 02:19

Dan: It's backwards from how people imagine it. The easiest things are your voice and image. They're quick to train, and a few companies are developing the technology for that. About 30 seconds of voice is enough, plus maybe two to three minutes of high-quality video. Higher quality and longer is better, but it's relatively simple. The trickier part is searching for the information you want it to have and filtering through your own life to decide which parts of you to replicate. You have the version of yourself you put out on social media, another for work, and the versions reflected in the people around you. You may be one way to your closest family and a completely different person to a colleague. So we try to capture the relevant information for the relevant use case, so the replicas become useful and actually represent the person they're meant to be a copy of.

They act as an extension of ourselves that can share our knowledge, experiences and skills with people globally, 24/7, in any language.

Trust, oversight and jobs

Ashton: Won't people be creeped out by an AI using your voice and video?▶ 05:21

Dan: The uncanny valley is still there. People get a bit creeped out interacting with non-human robotics and AI. Video and voice have a wow factor, but people recognize it's an AI acting on someone's behalf, and a lot of the time they just prefer text messages. There's also an uncomfortable side when someone has their replica acting on autopilot, like I do a lot of the time. Not everyone is brave enough to let an AI do their replies automatically, so we're working on a human oversight feature. It's a bit like having your emails drafted rather than automatically sent, so you can check them. Having them drafted every morning will save a huge amount of time, and it's more comfortable for both sender and receiver to know a human approved the message. It's a bit like a personal assistant who sends mail for you with your approval first. That's weird, because people trust PAs all the time. Then suddenly, once it's done by a robot, which arguably is more accurate, there's something odd about it.

Ashton: Will this technology replace people altogether, or shift their jobs?▶ 13:51

Dan: I think there'll definitely be a shift more than a replacement. We see this as an empowerment tool, an extension of ourselves. Certain industries are more at risk than others, and certain jobs will inevitably be lost and shifted into different jobs. It will be difficult to start, with fast-moving shifts that disrupt a lot and create inequality at first. But if you believe in the good of AI and the good of people, AI fixes a lot of other problems around the world, and work could become almost irrelevant. There may be a global basic income and living standard we could achieve using AI. As with any industrial revolution, a lot of people are afraid and fight back, but those who embrace the new technology inevitably win out. I don't think we should be afraid of AI in general. It's not quite Skynet just yet, and in fact it's quite a long way off. I think we're probably more of a risk to ourselves than AI ever will be.

Where replicas fit in business

Ashton: Which business areas are the obvious first use cases for Sensay?▶ 08:32

Dan: We see two major ones. The first is any company that deals in knowledge management: consulting, legal firms, education, healthcare, and many other industries that rely heavily on individual expertise. There's a huge brain drain when people move between organizations, retire, go on holiday, take maternity leave, or just sleep. Replicas of high-performing people let them share their knowledge even when they're unavailable, far wider than their physical and mental capabilities allow. Take one of the world's leading professors in a subject at MIT, teaching it on a very intimate basis to anyone globally, in places where that knowledge is scarce. Or take companies that pay tens of thousands of dollars for a speaker. With a replica, they get the speech plus an interactive version people can ask questions directly. The second use case is a very scalable message automator, like autofill on crazy steroids. It uses a bit of your personality, your internal communications and access to the right documents. It can draft messages from previous conversations, write summaries, book things into your calendar and handle the usual day-to-day admin, with human oversight if required.

Ashton: Is Sensay built for small startups as well as large corporations?▶ 16:58

Dan: I think the age of those big companies is kind of over, and AI really shifts that. We use a SaaS model, with enterprise plans for larger organizations, but it can be purchased by individuals too. The features are the same. The difference is that enterprises get an internally hosted version of Sensay, blocked off from the general platform and the internet. AI has been trained mostly on public data, but there's a huge amount of corporate and organizational data that's still relatively untapped and still closed off. Large organizations are being very cautious about AI, with good reason. The big guys move slower and integration will take longer, so there's a lot of opportunity for smaller, more agile companies. And it isn't just companies. Governments, nonprofits and charities can all benefit from a replica workforce.

Ashton: So an enterprise can run its own private model, segregated from everyone else?▶ 21:02

Dan: Exactly. We're very aware that this is a big issue for a lot of companies. So we can run a Sensay node within their existing systems, with its own model completely separate from the rest of our platform. It's available to larger enterprise customers who want that. For us it's just another element of security that we've had to build in from the early days.

I think we're probably more of a risk to ourselves than AI ever will be.

Staying ahead and getting involved

Ashton: AI is improving dramatically each year. How does Sensay keep up?▶ 22:10

Dan: Someone recently published that we're no longer following Moore's law, it's now Moore's law squared, which is crazy. To keep up, we focus on building the best-quality replicas of people: the personality engine. That means capturing what makes us us, which parts of us make us sound more human and less like chatbots, and how our nuances, characteristics and emotions can be presented digitally. It's not the easiest thing to build, so by focusing on it we become the experts in that field. The replicas will gain more and more utility as agents and other utilities are added bit by bit. It's the same way mobile phones went from a quick way to phone people to smartphones that keep adding features.

Ashton: How can entrepreneurs and startups get started and follow along?▶ 24:47

Dan: Check out our website, Sensay, and find us on social media. We're taking on a few design partners right now, especially larger enterprises, so we can build the solutions they're looking for. We have an ambitious roadmap over the next two years, both in sales and onboarding a substantial number of customers, and in features, inputs and external channels. Even now you can create replicas and put them on your own website as a widget, on Telegram and on Discord. WhatsApp and Slack are coming very soon, and the email drafter arrives at the end of this month.

This interview has been edited for length and clarity. Recorded October 2024.

Dan Thomson
Dan ThomsonDan Thomson is the founder and CEO of Sensay, an AI company that began with digital replicas of individuals and now focuses on knowledge capture and transfer for organisations. He studied philosophy at King's College London and completed an MBA at the University of Cambridge.… Full profile →
Share this interviewXLinkedIn
▶ Watching the interview