It was so fun to have Wade Chambers, Chief Engineering Officer at Amplitude, on the latest episode of How I AI.
Wade shared how Amplitude built Moda, their internal AI tool, which has automated a bunch of their product development process.
Instead of relying solely on off-the-shelf solutions, Amplitude decided to build its own, and they were pretty happy they did.
In this blog, we'l look at the technical details behind the scenes and some practical takeaways to help you build your own internal AI tools.
Build internal AI quickly
My favorite part of this story is the beginning: Wade's team built Moda using spare time over just three weeks. How did it start? He saw another company's internal slackbot and wanted one for Amplitude. A few engineers jumped on the task, learning how to build AI agents and enterprise search as theygo.
Workflow 1: Social Engineering for Viral Adoption
Moda's success can be attributed to two things: it's functionality and its UX. While the capabilities of the agent itself are cool (data analysis, full enterprise search, open availability), the fact that the team chose to deploy a slackbot vs. a web app or other platform ensured it would get adoption across the team.
Building an internal AI agent step-by-step
- Building a Slack Bot: The team started by developing a simple Slack bot that interacts with all of Amplitude's enterprise data. This made it incredibly accessible for all employees.
- Publicly Visible Results: Instead of hiding the tool or its development, the team made it public and encouraged employees to experiment. This approach generated excitement and curiosity, as well as feedback requests.
- Leveraging Existing Behaviors: The design mirrored tools employees already used in slack, reducing the learning curve.
- Observability and Learning: The slack platform and open channels allowed employees to see how colleagues used Moda, borrowing questions and prompts. This fostered organic adoption and community building around the tool, which helped it become a favorite across the copmany.
- Results: Within a week, Moda had become a company-wide tool. This rapid adoption shows the power of transparency and user-centric design in driving AI adoption.
Workflow 2: Analyzing Customer Feedback with Moda
One of Moda’s most valuable applications lies in its ability to analyze customer feedback from various sources, providing actionable insights for product managers.
Step-by-Step Process:
- Broad Data Input: Moda gathers data from multiple sources including Slack, Zendesk, Product Board, and transcriptions of customer calls.
- Thematic Analysis: Product managers can ask Moda to identify top themes and inquiries from this diverse data set. This is a prompt example:
"Analyze recent customer feedback from Slack, Zendesk, and call transcripts to identify top themes and inquiries." - Narrowing Down the Scope: Once initial themes are identified, Moda can be further queried to drill down into specific areas. For example:
"Give me more details on customer requests for connecting session replay to funnel analysis." - Actionable Insights: Moda provides not just summarized themes but concrete quotes and context from customer feedback, ensuring that analysis aligns with real user needs.
Key Tools: Moda (custom built), Glean, Zendesk, Product Board, Slack
Results: This workflow significantly accelerates the process of gathering and analyzing customer feedback, streamlining decision-making and product development.
Workflow 3: Rapid PRD and Prototyping using Moda
Perhaps the most impressive workflow demonstrated was Moda's ability to rapidly generate Product Requirements Documents (PRDs) and even prototype instructions.
Step-by-Step Process:
- Single-Sentence Prompt: The process begins with a concise prompt, like:
"Customers want to see session replays directly linked to funnel steps so they can watch where users drop off or convert." - Automated PRD Generation: Moda takes this prompt and automatically generates a comprehensive PRD, including problem exploration, solution exploration, detailed requirements, and even prototype generation instructions.
- Multi-Tool Integration: The generated prototype instructions can be directly integrated into tools like Bolt, Figma, Lovable, and v0. This facilitates rapid prototyping and cross-functional collaboration.
- Iterative Refinement: The process is not static. If the initial output doesn't meet expectations, Moda allows for iterative refinements and adjustments based on user feedback.
Key Tools: Moda (custom built), Confluence, Bolt, Figma, Lovable, v0, GitHub
Results: This workflow compresses what could take weeks into a single meeting, dramatically increasing the pace of product development. It also fosters collaboration by seamlessly integrating product, design, and engineering efforts.
Everyone should build an internal AI bot
Amplitude’s Moda showcases the power of building internal AI tools tailored to your company's specific needs.
The three workflows Wade showed off demonstrate how AI can change how your team works together, what data they use to make decisions, and how products get built.
When you choose to build, not buy, your AI platform, you can customize your workflows to your specific data sources and tools, resulting in significantly higher efficiency and enhanced product development, which mean more people will actually use it.
Thanks to our sponsors!
This episode is brought to you by:
- CodeRabbit: Cut code review time and bugs in half. Instantly.
- Vanta: Automate compliance and simplify security.