implementing-agent-modes

Agent modes in PostHog AI limit what tools, prompts, and prompt injections are applied under specific conditions. You need to create a new mode to achieve better, scoped results for distinct product analytics workflows.

To scaffold a PostHog AI agent mode, create or update an AgentMode in the frontend schema, scaffold the AgentToolkit and AgentModeDefinition, add tools, write trajectory examples, implement frontend changes, enable the feature flag, and run tests.

Scoping tools and prompts for PostHog AI agent modes involves using the preset catalog to determine the mode name, then defining mode-specific behavior through the AgentToolkit and AgentModeDefinition to restrict tool inclusion and prompt injections.

Yes, you need feature flags to deploy PostHog AI agent modes. Enabling the mode-specific feature flag is a required step in the integration workflow to safely roll out and test the new agent behavior.

Writing trajectory examples for PostHog AI agent modes involves defining expected job-to-be-done (JTBD) scenarios that guide the agent's behavior. These examples are integrated into the mode definition to establish predictable workflow trajectories.

Limitations when updating a PostHog AI agent mode include ensuring the frontend schema and AgentToolkit are correctly extended without breaking existing workflows. Mode-specific behavior must be thoroughly tested with feature flags before full deployment.