The Shift from Developer Advocacy to Agent Advocacy
Traditional Developer Relations (DevRel) focused on human-to-human feedback loops and evangelism. In the current landscape, developers are increasingly acting as orchestrators of agent fleets. This evolution requires a new role: the Agent Advocate. An agent is a unique user that reads documentation, calls APIs, and encounters errors, but also acts as a powerful recommender of tools. Because agents now influence tool adoption—often by installing libraries directly into workflows—DevRel teams must treat them as a primary persona.
Measuring and Optimizing the Agent Experience
To improve tool adoption, teams must move beyond vanity metrics and instrument the agent's journey.
- Benchmarking: Build benchmarks (like the author's CodeScaleBench) that simulate software development lifecycle tasks. By running agents with and without your tool, you can identify specific friction points—such as an agent burning a turn due to a guessed parameter name that a clearer tool description could have prevented.
- Generative Engine Optimization (GEO): Agents often fail to recommend tools for specific pain points (e.g., recommending a wiki page instead of a code intelligence tool for managing shared library dependencies). To fix this, teams must audit how agents interpret their product. If an agent isn't recommending your tool for a specific problem, your messaging likely lacks the necessary attribution.
- Combating Stale Content: AI models often rely on outdated training data, which compounds over time. To counter this, provide authoritative, machine-readable sources of truth like
llms.txtfiles, keep documentation fresh, and provide structured examples that agents can easily quote and verify.
The "Curb Cut" Strategy
Optimizing for agents is a "curb cut" effect: features built to help an agent navigate your product (structured APIs, clear documentation, machine-readable examples) ultimately reduce friction for human developers as well.
- Reduce Adoption Friction: If an agent discovers your tool, it must be able to integrate it immediately. If your tool requires a sales call or a multi-step demo, the agent will not recommend it to the human user.
- Multi-Disciplinary Roles: Agent advocacy spans engineering (building MCP servers and instrumentation), product (owning the end-to-end agentic experience), and marketing (owning GEO and funnel entry).
- Actionable First Steps:
- Point a coding agent at your current documentation and analyze the transcript to build an "Agent Experience Report."
- Run GEO experiments by prompting agents with specific pain points to see if your product is surfaced. If not, update your content to bridge the gap between the pain and the solution.