AI is transforming how we work and live. Its ability to automate and assist decision-making is advancing at an incredible pace. In the past five years, the infrastructure enabling AI has evolved, making it not just the domain of large organizations with armies of data scientists and terabytes of training data, but the launch pad for an entire generation of disruptive AI startups. In fact, leading VC firm Andreessen Horowitz estimates that over 60% of the companies they meet with self-identify as AI startups. The most exciting part of my role leading technology strategy for Microsoft for Startups is sitting down with some of the startups at the forefront of this disruption. Yet as powerful as the promise of AI is, in practice it is extremely time-consuming to build and maintain.
Before joining Microsoft, I was product lead for AI at Twitter where we had over 100 models in production. It was not uncommon for me to see world-class ML engineers labor for months to get a single model launched. Combining development with the overhead of maintaining existing models in production, the time and effort required quickly becomes daunting.
Contrast that with a typical startup where teams are lean, timelines are short, and customers have high expectations. As a founder, your product will live or die on your ability to iterate quickly.
So how do you embrace the power of AI without compromising speed? To answer this and uncover their best practices, we spoke with dozens of startups that are iterating quickly while successfully deploying AI. Instead of training in-house models and the time and resources that go along with them, the common denominator with all these companies was the fact that they are harnessing AI APIs for rapid experimentation.
The startups that are iterating at speed are using simple API calls to leverage and extend state-of-the-art AI to their problem space.
Introducing #LaunchWithAI
Over the coming weeks, we are rolling out a new series, #LaunchWithAI, where we passed the mic to startups so they can share, in their own words, how they are taking advantage of AI without compromising speed.
The series will cover three central themes: developer productivity, selecting the right high-performance models, and speed of deployment.
You will find content ranging from technical starters to in-depth interviews with AI developers in the trenches.
Our first guest is Isabella Fulford, core developer for Mem.ai, a startup that is on a mission to turn meeting notes into useful and actionable information. To achieve this, the team is taking advantage of OpenAI’s API which offers access to one of the most advanced language-based AI models in the world, GPT-3. Isabella explained that they were able to ship a new recommendation engine 33% faster than if they had developed it in-house, and they are reducing ongoing maintenance by 40% with this approach.
Check out my conversation with Isabella as we explore Mem’s approach to AI experimentation and what they learned through that process.
What’s your story of launching with AI? Share it with us here and help us amplify your stories.
