Sierra: From Chat to Voice AI in 2024 - A Decade-Long Journey

Hi everyone, my name is Zach Reno Adine, and I'll be sharing stories and insights on how we build and improve conversational AI agents at Sierra.

  • * Zach Reno Adine from Sierra, a conversational AI platform for businesses, is giving a talk.
  • * Sierra is known for chat experiences and customer service, but is expanding into more areas such as sales, subscription management, and product recommendations.
  • * By the end of 2023, most of Sierra's interactions will be over the phone.
  • * In 2016, Zach was working at Google on Google Lens, which was in its infancy and only able to identify plants with some accuracy.
  • * Building AI is like a slot machine - it works sometimes, but not always.
  • * Present day Google Lens can identify objects, translate text, do math homework, and more.
  • * Sierra's agent development life cycle involves iterative refinement with customers in production to make it as productive and bulletproof as possible.
  • * Quality assurance for Sierra agents includes looking at every conversation, providing feedback, filing issues, creating tests, and making new releases.
  • * Chubbies has a budget for each of its agents to delight customers, such as door-dashing items from retail locations if they're not available online.
  • * The agent development life cycle is more effective the larger the customer is.
  • * Reasoning models are a force multiplier towards each step in the agent development life cycle, allowing for more effectiveness in applying AI to development, testing, QA, and every step in between.
  • * Sierra's voice capabilities allow companies like Serius XM to pick up the phone right away to answer their customers.
  • * Building for voice is similar to building for web development - it's the same platform and agent code, but able to be responsive to whatever channel someone reaches out in and whatever modality they
  • * Large language models remind us of ourselves in that they're unpredictable, slow, and not great at math, but also allow us to be great designers by having empathy in a way that we probably couldn't
  • * At Sierra, they're building voice agents that are much more robust than just transcribed text with a delay.

Source: AI Engineer via YouTube

❓ What do you think? What are your thoughts on the ideas shared in this video? Feel free to share your thoughts in the comments!