Agents work by synthesizing data from different parts of your Amplitude account: , , , as well as . So the more you're already using Amplitude to analyze user behavior, the more context your Agents will have to work with.

Start by defining a goal for your Agent—like improving conversion or boosting feature adoption. The Agent will use your Amplitude data to identify patterns and make recommendations that you can approve for the agent to act on.

You don't need to be a data scientist to work with Agents. We built them with a chat interface, Amplitude Agents are designed with natural language interfaces that make them accessible to team members across skill levels. You can interact with agents via chat, give them feedback, and adjust their autonomy level based on your comfort.

We're using a combination of models from OpenAI and AWS Bedrock. We've also created a bunch of specialized tools and sub-agents that handle different parts of the work - some focus on analysis, others on running experiments or creating guides. Most of the orchestration, memory, and session management is supported by our own backend work. This approach gives us the flexibility to really customize how agents behave for product analytics specifically, rather than just using a general-purpose AI.

Agents will be part of the core Amplitude platform - we want every customer to benefit from having AI teammates. There will be additional capabilities you can unlock with other products, but we'll share more specifics about packaging when we launch generally later this year. Right now we're focused on making sure the experience is great for our beta customers.

It's likely that we'll support these standards down the road, but we're taking a deliberate approach right now. We're focused on getting the core agent experience right first, then we'll expand to broader integrations.

Amplitude does not train on customer data. Agents learn from a few different sources. They start with general best practices that come from the latest foundation models - things like optimization patterns that work across different industries.

Then we layer in Amplitude's own expertise through system prompts that we've developed from working with thousands of teams.

Finally, each agent builds up its own memory about your specific product and what works for your users - and this knowledge stays completely separate and secure for each customer. Over time, agents get smarter as the underlying models improve, as we refine our approach, and as they learn more about your particular business.