I met Amy several months ago when a colleague invited me to a meeting. My first thought was jealousy over his ability to afford an assistant. I later found out that his assistant, Amy, was the manifestation of an AI scheduling service. And then my jealousy went from his income to his place in the beta queue, which I signed up for immediately.
Several months later, after a few polite notes saying they had not forgotten about me, I was invited into the beta. Amy is now my assistant.
Amy hails from New York’s X.ai, a company founded for one reason and one reason alone, to schedule meetings.
Despite the hype and fear mongering around AI these days, X.ai knows that all AI has hard edges over which an experience fails. IBM’s Watson, for instance, which famously won Jeopardy, was not able to take part in the interview portion of the show. Interpreting questions and rapidly searching for answers is very different than informal repartee. X.ai has limited its ambitions to scheduling meetings. It is hard enough to solve one tough problem, and for those of us who spend a lot of time in meetings, we know that converging on a common time for a group of people in today’s increasingly geographically distributed work environment — often consisting of loosely affiliated teams – can be very time consuming, distracting and occasionally labor intensive.
X.ai has created a scheduling service that eliminates the burden of schedule negotiation, as well as rescheduling. Amy, and her male sibling or alter ego Andrew, are a breed of vertical, focused AI, meaning that they are designed to get one “human” job done, not to be general purpose problem solvers.
Schedule automation solutions have failed in the past, according to X.ai CEO and Founder Dennis R. Mortensen, because “all of these solutions suggest that setting up a meeting can be a shared activity.” Mortensen is pretty sure that isn’t really the optimal answer. He and his team spent over 14-months creating a conceptual model of the universe of human interactions related to scheduling. Rather than discovering some grand unified model of scheduling, X.ai deploys unique models for each intent, for instance, parsing the request language, rescheduling or setting up an initial meeting.
X.ai’s Customer Acquisition Director, Brian Coulombe, says that the company is focused on quality over quantity at this point, with some date verification still taking place to ensure that Amy and Andrew perform admirably. That said, x.ai is already scheduling tens of thousands of meetings a month during the beta. They did not disclose the number of actual beta participants.
The X.ai service is also in a class of emergent software being called invisible software. Amy requires no download, no app and it has no real user interface after sign-up. Simply copy Amy or Andrew on an e-mail message aimed at scheduling a meeting and they take over the triangulation of date, time and channel/place. If a place isn’t designated, they will ask. Once Amy converges on a meeting time, the service either posts directly, or sends an appointment through e-mail. Early on they copy their employer — you — to make sure the host is comfortable with how Amy is handling the scheduling process.
Marketing is subtle. Respondents are not hounded after being invited to a meeting through a beta tester. (Only the tag line in the host’s signature provides a link to the service, and this was put in place to let guests know that Amy and Andrew are AI, not human.) Coulombe stated that “our job is to schedule meetings. It wouldn’t be right for a company dedicated to productivity to add e-mails when they are focused on saving time.” It is the service’s unique technology, Coulombe asserts, that first attracts users, and the deft handling of time that keeps them.
Though pricing hasn’t been set, Coulombe said that X.ai’s scheduler is a pure automation play, not one that includes the costs associated with human curation. Users should expect, he said, “to pay something akin to a double latte” monthly to secure the more advanced features of the product. Like many cloud-based products, a personal version of the assistant will remain free. The product will also come in professional and business versions. Although final features aren’t set, the ability to name the assistant and to include it within a company’s domain are likely, as are features like putting tentative appointments on a schedule at the “boss’s” request.
Mortensen imagines a future where vertical agents “do jobs, be it reconciling receipts or setting-up travel.” Amy isn’t yet ready to meet Siri or Cortana, but perhaps someday people will be able to say, “Hey Cortana, would you schedule a meeting tomorrow with Bob at 2pm” and Cortana will handoff the scheduling task to Amy.
“And now comes the more exciting question, I think,” Mortensen continues, “How do individual vertical agents communicate amongst each other? So far, for the last half-century any two computers that needed to communicate would either exist on some open protocol, or set of standards we agreed to as an industry, or we go build distinct APIs. I find it unlikely that the next paradigm will play out similarly. I think it is much more likely that the next software paradigm will consist of intelligent agents, not apps, and they are going to end up communicating with each other in plain English.”
He goes on to say, “simply because when one agent communicates with another you aren’t yet sure if it is a machine or a human, and the safest way to communicate, if you can’t know that, is to assume the agent has a backup that will be forwarded to a human.” English provides the transparency into the black box.
As the beta grows, X.ai is already realizing their vision of low friction, near immediate schedule convergence, especially when two users of Amy or Andrew engage the system on same event. Because they have complete visibility of each party’s schedule, and, equally important, full understanding of the preferences of all parties involved, Amy/Andrew quickly negotiates with itself and sets the meeting. As use increases, that immediacy will reinforce the invisibility and utility of the system.
In my own testing, I invited some people with full disclosure as to why I was copying Amy and what Amy was intended to do. In other cases, I just copied Amy to see what would happen. In all instances Amy was able to book a meeting, though she has also had to delete a few or reschedule them — because of human reneging on commitments, not because Amy made any errors. One recent response sent to Amy from a communications professional at a large software company apologized to her profusely for the necessity of moving a meeting. Because of responses like this, Mortensen and his team have imbued Amy with rudimentary empathy, so for instance, she knows how to accept a “thank you” gracefully.
X.ai’s scheduling siblings will not only assist in business scheduling among teams, teams and partners, and individuals, like customer service representatives, and consumers, but may well be the biggest boon to individual entrepreneurs in the “gig economy” who are trying to look professional and stay productive while still making ends meet.
Production versions of Amy and Andrew will likely be available sometime in early 2016. Until then, those wish to queue up for a personal scheduling assistant can sign-up at X.ai and take their place in the beta line.