This post is intended to be the first in a series about empowering product, UX, and engineering teams with AI. We’re going to focus on how we’ve approached and scaled the use of AI in the context of writing code.
AI has fundamentally transformed how we build software at HubSpot. Over the past two years, we've gone from cautious experimentation to achieving near universal adoption of AI coding tools across our engineering organization.
This transformation didn't happen overnight. It required strategic investment, organizational commitment, and a willingness to learn. As we've shared our experience with other engineering leaders, we've discovered that many teams are facing similar challenges in scaling AI adoption beyond POCs and early adopters. The conversations we've had with external teams convinced us that our lessons learned could help others navigate this journey more effectively.
Adoption of AI coding assistants, % of members in the Engineering organization [sanitized data]
In the beginning there was code completion
We began experimenting with GitHub Copilot in the Summer of 2023. Our founders Dharmesh and Brian provided us with the push we needed to get started. Dharmesh had recently used GitHub Copilot to build ChatSpot and had a good experience with it, so he and Brian pushed us to evaluate it and connected us with other leaders in the industry who were seeing success.
Our proof of concept (POC) successfully validated the tool's potential, and several factors contributed to our success:
Executive buy-in made everything else easier . Support from Dharmesh and Brian accelerated our pilot process significantly. This helped our legal, security, and engineering team have the same goal and urgency for making this happen.
We ran a pilot that was sufficiently large : Our strategy was to include entire teams so they could adopt and learn together that had different experience levels, different missions, and worked in different domains. We gave teams over two months to try it..
We put energy into enablement . We had setup/training sessions and created a channel where people could ask questions and share what is and is not working.
... continue reading