Amazon has counted on others to spread Alexa everywhere — first developers, then device makers, and most recently everyone else. It’s been about a year since Amazon unveiled Alexa Skill Blueprints, a series of templates designed to help anyone build their own capabilities for the digital brain.
Since the program debuted, customers have created millions of custom skills and interacted with them millions of times, Amazon said, declining to give specific figures. Going forward, Amazon is looking to add more skills templates and make the service more appealing for businesses.
To make a skill, users choose a category on the Skill Blueprints website and template such as trivia, facts, flash briefing and more. A four-step form has users input the content they want Alexa to say in response to customized voice prompts. Amazon takes care of the rest on the back end.
Amazon is battling it out with Google in the competitive digital assistant market. Both companies are looking to add new capabilities and features to draw both developers and users. In an interview with Geekwire, Steve Rabuchin, vice president for the Alexa division, said Amazon is the only company that gives anyone templates to build no-code skills. The company is aiming to spread Alexa while also making the digital assistant smarter and more accessible.
“Oftentimes we release these really interesting technologies and it’s only the biggest companies participating and really skilled software developers that are able to use them,” Rabuchin said. “And we have hundreds of thousands of developers in our ecosystem building skills. But if you just think about if we can open this up and make it accessible to everyone, it’s millions of people if not billions of people that can participate in artificial intelligence and customize their own voice interface and also help us make Alexa much smarter.
Overall, Alexa has proliferated rapidly since Skill Blueprints were introduced. When the program was unveiled in April 2018, Alexa had 30,000 skills. Now the digital brain has more than 90,000 of these voice-powered apps.
The Skill Blueprints program began with 20 skill templates across five categories to choose from. Blueprints has grown to 55 templates across seven categories, including a new Flash Briefing template for news organizations.
At the beginning, the skills lived only on the user’s device. Today, some parts of Alexa’s standard answers can be overridden, and users have the option to share and publish the skills they build.
With the ability to publish skills comes the need to watch out for toxic material. All Skill Blueprints have to go through the standard approval process for Alexa skills. Rabuchin said foul language is not allowed in skills, and users aren’t allowed to do things like reprogram the weather and news.
“We have a whole bunch of these policies we run it against,” Rabuchin said about quality inspecting skills. “And then we do an amount of testing as a customer would just to make sure it’s working and a few other steps before it’s public. Those processes have been around for years. They’ve been hardened and evolved and everything.”
Amazon was already kicking around ideas to make Alexa skills easier to create, but it was an email to CEO Jeff Bezos that kicked things into high gear. An educator who wanted to teach students to build Alexa skills reached out after not being able to find a simple how-to guide, Rabuchin said.
Bezos tasked Rabuchin and a few others with getting the ball rolling, and they drew up a six-page or less document describing the project, standard operating procedure at Amazon. Bezos met with the teams and pushed for an important template that eventually became one of the most popular uses.
“The biggest piece of feedback he gave us that was spot on was we didn’t have the notion of just a simple question and answer Blueprint, which is just the ability to write a simple question and answer. He said, ‘no, you’ve got to do that at launch.’ And it turns out that’s the most used blueprint to date,” Rabuchin said.
Rabuchin has seen two of Amazon’s biggest businesses rise from their early days. He’s been with Amazon since 2005, hired to work on Amazon Web Services by its current leader, Andy Jassy. He worked on Kindle and ran the company’s app store teams before coming to the Alexa team three years ago.
When he first started on Alexa, the digital assistant was on a single device — the original Echo — and had a limited of skills. Alexa has been built into 150 different gadgets made by other companies, and the total number of Alexa-powered devices sold is more than 100 million.
Rabuchin is tasked with spreading Alexa by opening up all the technology Amazon creates for the digital assistant to companies and developers. On Blueprints specifically, Rabuchin and teams have set their sights on digging deeper into current templates to add greater functionality, while also focusing on new capabilities for businesses and publishers.
The world of digital assistants is a new frontier, and Amazon has grabbed the early lead. However, the concept of a device that listens into to your conversations and records things, sometimes by accident, can be a little unnerving. Reports that employees review some Alexa conversations raised eyebrows among consumers.
Rabuchin reiterated the company’s commitment to privacy and security, and Amazon provided a statement saying it only annotates a small number of conversations picked at random to “train our speech recognition and natural language understanding systems, so Alexa can better understand your requests.” Amazon says it has safeguards in place to restrict access to that information and make sure employees can’t see any identifying account information.
Skill Blueprints feature very little tracking of customer use, beyond going through the approval process and watching what types of templates are used most often, Rabuchin says.
“We’ve abstracted so much of how the skills are built that they’re basically those templatized,” Rabuchin said. “So there’s not a lot of machine learning that’s going on there for us. We’ve made it in such a way where we just know they’re going to input things in these fields and there’ll be a response.”