When Dennis R. Mortensen hired his founding team members at x.ai, he pitched them by illustrating his vision of a world where everyone has a personal assistant to schedule their meetings. Recognizing the complexity of the challenge, he concluded by saying: “We may die trying.”
“It’s a great setting to hire qualified people,” Dennis says, recounting the late days of 2013. “We’re working on something that is clearly very hard, to the extent that we might not make it, but it’s not impossible.” That’s the sweet spot, he says. When a calling finds you, that is incredibly challenging yet still attainable, even if only at your fingertips. “It’s right in the middle,” he continues. “It’s not a weekend hackathon, but it’s not space travel either. It’s that area right in between where you find the best people.”
We jumped into a pool and figured out that it was the ocean. We can’t see the shore from here.
x.ai has since grown to 64 team members, who Dennis refers to as propeller heads, who are working day in and day out to bring Amy and Andrew Ingram, whom many of you know, to life.
From his initial meetings with his seed investors (including IA Ventures, Lerer Hippeau Ventures, and SoftBank Capital) to our first conversation in early 2015, Dennis has been adamant in demonstrating his long-term perspective for x.ai. Countless solutions currently exist to schedule meetings: actual assistants, virtual assistants, apps, plug-ins, and websites. The list goes on, and each has a different price point.
x.ai has taken a different approach to developing Amy. They refer to her and her brother as “invisible software.” “She can never be an app, plugin, or website. Amy has to exist in dialogue. You would never go up to your assistant at the front of your office and say ‘Show me your features! What can you do for me?' It would be completely unacceptable,” Dennis affirms.
The goal is for Amy to feel and communicate like a human, and she does. If you’ve ever had an interaction with her, you’ll find yourself asking how her day is and thanking her. More than 10 individuals have even asked her on a date. x.ai has an AI Interaction Designer, who focuses specifically on developing these humanistic conversational skills.
“When you communicate with Amy you should forget that she is a machine because it doesn’t matter. You can be confident that when you tell her something, she understands and will do what you requested.” The mission has required two years of deep learning, testing, and iteration. The first was spent primarily gathering data and developing the tools that would enable the team to start building Amy. According to Dennis, they are building the tools to build the machine while building the machine.
“In the darkest hours of our first year there was no meeting data that we could acquire and model on top of. Nothing. Zero," he says. "We started out by setting up one meeting then another, then 10, 1,000, and so on until we went through millions of emails that we annotated, with great pain, so we can create models on top of them.”
Only the brave start in a place where there is no data.
Despite acquiring foundational knowledge the last two years, the challenge is significantly more complex than the message that appears in our inboxes. Take the response “Sure, Jessica. How about next Tuesday-Thursday around 8?” sent at 11 p.m. on Sunday night. Countless questions arise for Amy or Andrew on the other side of this exchange: Is Jessica referring to this upcoming Tuesday and Thursday or the following? Does Wednesday count? Is she referring to exactly 8, or perhaps 7:45 or 8:15? Will this be a morning meeting or an evening meeting? Will you be commuting from a prior meeting? If so, from where? The most significant question then is: How do they proceed?
“Little things turn into massive complexities on our end,” Dennis asserts. “The more we learn, the more ambiguity we see in everything that humans do. The less complexity you put on the user side, the more you have to put on the creator.”
We’ll keep chasing it until we solve it, or we’ll die trying.
The mentality appears intense but scheduling meetings is a vital part of our personal and professional lives. Say you’re meeting with an investor to close your funding round and Amy makes a mistake. There is simply no room for error. It is precisely because of this that x.ai has made your calendar their sole focus, despite requests from customers asking for more capabilities such as arranging deliveries or scheduling travel.
“We just want to perfect Amy. We want her to be so good that even if you can hire a human to perform this task you won’t need or want to,” Dennis says. The goal is for Amy to grow up to be an adult who can handle everything on her own without human intervention. x.ai is on track to achieve that as the team prepares to release their first paid product this autumn. “We are working on a clearly defined list as we speak," Dennis confirms. "We know exactly what we need to work on to make Amy into a superhuman when it comes to setting up meetings.”
Similar to self-driving cars, flawless artificial intelligence isn’t going to be ready overnight. The progress will be based on deliberate experimentation, slow growth, and most importantly, resilient pioneers. “It’s not reasonable to think that an AI agent is going to arrive at an oracle level,” Dennis says.
The 'Ta-da' moment Hollywood’s been telling us about will not happen with AI.
The more likely scenario, according to Dennis, will be a team of vertical AI agents meaning that we’ll have individual artificial intelligence agents who each work on a single task for us such as scheduling our meetings, booking travel, or ordering our groceries. For example, if you’re traveling to Los Angeles your AI travel agent will arrange your flights, hotel, and transportation. Amy or Andrew will then communicate with him or her to determine the best place to schedule your meetings based on your arrival time, distance to the hotel, and trip duration. Your input will not be required.
Our lack of participation raises what Dennis believes is the common misconception that AI will lead to doomsday scenarios with little human interaction and mass unemployment.
“We all read the same blogs asserting that creating AI is like summoning the devil. I just don’t believe that. No one has ever come up to you and congratulated you on scheduling a meeting. It’s a task we are given when we don’t have anyone else to do it. There is no creativity required. I can’t play out a scenario where we end up worse.”
Dennis believes that AI will free us to do the things that we excel at and are passionate about like brainstorming ideas, creating art, or leading our teams. “Artificial Intelligence is an agent for change enabling us to catalyze on our creative pursuits,” he says. It will never replace human ingenuity.
To follow along with Amy and learn about the future of AI, subscribe to x.ai’s blog, where you can learn about topics like how to think about AI agents in the future, and sign up for the waitlist here.
- How MeUndies Made Merchandising Their Competitive Advantage — When MeUndies thinks about their underwear subscription service they compare themselves to Netflix. What the streaming pioneer did for TV, the Los Angeles startup wants to do to your underwear drawer. The goal is to provide monthly subscribers with the staples they love while delivering elements of surprise - Think briefs with donuts and dinosaurs - right to your doorstep.
- Seedling Relies on Childhood Curiosity to Inspire a Creative Team — On a recent flight home, I spent over an hour browsing a children’s play website, completely enamored by sets like Design Your Own Superhero Cape and Invent Your Own Insects. After sharing Seedling with everyone I know, despite few of my friends having children, I wondered why the brand struck a such a meaningful chord with me.
- eero Designs Wi-Fi for the Smart Home Era — Nikhil Basu Trivedi, Principal at Shasta Ventures, recently distinguished exceptional founders by their ability to prioritize. “It all comes back to prioritization, speed, paranoia, and knowing that if you don’t iterate, even after finding product market fit, you can be disrupted by the next product,” he says.
back to the blog