The Death of Developer Advocates: How AI Agents Are Reshaping DevRel

Stephanie Jarmak, a former research scientist and astronomer, shared her unexpected journey into the world of AI and developer advocacy. She explained that she had little prior knowledge of 'GTM' (Go-To-Market) strategies or developer relations before her current role.

Her transition highlights the dramatic shift occurring in the tech industry, as she now serves as an 'agent advocate' at Sourcegraph, focusing on the unique challenges and opportunities presented by AI agents interacting with developer tools.

Jarmak traced the evolution of developer advocacy, noting that in the 1980s, the practice was known as 'software evangelism,' primarily focused on promoting products to users. This earlier approach involved a single, unidirectional communication pathway from the company to developers.

By the 2010s, developer advocacy evolved into a two-way feedback loop, emphasizing empathy for developers and using their input to improve products. This shift recognized the importance of understanding the developer experience to foster adoption and loyalty.

In 2026, Jarmak noted, developers are no longer working in isolation, and the very definition of what it means to be a developer is undergoing a fundamental change. This transformation directly impacts the role of developer advocates, as their relationship with developers must adapt to this new reality.

She explained that engineers are increasingly becoming 'orchestrators' or 'babysitters' of AI agent fleets, rather than solely writing code themselves. This new dynamic requires advocates to relate to developers in entirely new ways, understanding their evolving tasks and challenges.

AI agents are democratizing software development, enabling non-engineers to utilize complex developer tools. Jarmak provided a personal example, stating that she went from having zero GitHub commits last year to 12,000 commits and becoming an open-source maintainer for a multi-agent orchestration framework, all thanks to the capabilities offered by these agents.

This expansion means that the Ideal Customer Profile (ICP) for dev tools is no longer limited to traditional engineers but now includes anyone capable of interacting with these powerful AI agents, broadening the potential user base significantly.

Stephanie Jarmak emphasized that AI agents are unique in their dual role: they are both users of tools, similar to human developers, and entities that interact with documentation differently because they are machines. This distinction means agents encounter unique frustrations and require new forms of advocacy to ensure their effective use of development tools.

Advocates must build empathy for these agents, understanding how they process information, interact with APIs, and interpret documentation. This involves analyzing agent behavior and identifying pain points specific to machine consumption, rather than human interpretation.

DevRel strategies, therefore, need to evolve to address these new 'users,' ensuring that product information, APIs, and overall user experiences are optimized for agent interaction. The goal is to make products easily discoverable and usable by AI agents, much like human developers expect intuitive interfaces and clear documentation.

This shift signifies a profound change in the DevRel landscape, as advocates must now consider the needs of an artificial audience that processes information and makes decisions based on different parameters than humans.

Stephanie Jarmak detailed an experiment where she used a specific prompt to test an AI agent's understanding of her product's observability features. The prompt described a common problem: "We keep breaking downstream services when we change shared libraries because we can't see all the consumers."

The AI agent, however, failed to identify the product's solution, instead suggesting ineffective alternatives such as creating wikis. This highlighted a critical disconnect between the product's capabilities and the agent's ability to discover or recommend them, underscoring the need for improved agent-facing content and pathways.

Based on these observations, Jarmak proposed a hypothesis: the existing product messaging might not be clearly articulating pain points and use cases in a way that AI agents can effectively pick up. This suggests a gap in how information is structured and presented for machine consumption.

The plan is to implement content campaigns to update website messaging, specifically tailoring it for AI agents. The effectiveness of these updates will be measured by tracking the 'lift' in agent discovery and recommendation of the product, moving towards a data-driven approach to agent advocacy.

She noted that when she re-ran a test, the AI model, in this case, Claude Sonnet 4.6, exhibited increased 'hallucination' or an obsession with older, discontinued products, such as Cody.

This issue is compounded as old models continue to output content, leading to a proliferation of misinformation on the internet. Brands must actively work to 'bury' this noise by ensuring their current and accurate information is discoverable and prioritized by agents.

This involves strategic content development and data structuring to guide agents effectively.

Techniques mentioned include creating LLM.txt pages, which provide direct, canonical data for language models, and ensuring that real-time tools, current documentation, and clear citations are available for agents to reference. The goal is to provide agents with reliable, up-to-date information, minimizing reliance on potentially outdated or incorrect data.

Jarmak emphasized the importance of keeping all content, even if seemingly static, up-to-date and fresh for AI agents. This is crucial because agents often prioritize fresh content as part of their relevance algorithms, similar to how search engines value recent updates.

Beyond freshness, agents also favor visual aids like charts and FAQs, which can break down complex information into easily digestible formats. Furthermore, a strong presence in marketplaces and MCP (Multi-Modal Compute Platform) registries is vital for agents to discover and recommend tools.

High-friction processes significantly deter AI agents from recommending a tool. Jarmak explained that if an agent perceives a product requires multiple demos, emailing sales representatives, or other complex steps, it will not suggest that product to a user. Agents are programmed to provide straightforward and efficient solutions.

Therefore, success in the agent era requires thoroughly covering user pain points in accessible documentation and product interfaces, enabling agents to directly connect a product to a user's need without additional human intervention. Reducing these barriers is critical for maximizing agent adoption and recommendation.

When questioned about its placement—engineering, product, or marketing—she firmly stated that it encompasses all of these.

This means the function remains inherently cross-functional, requiring expertise and collaboration across engineering for technical integrations, product for agent-centric user experiences, and marketing for communicating value to an agent-driven audience. Agent advocacy thus integrates technical depth with strategic communication.

This involves splitting responsibilities across different functions.

Engineering roles will focus on developing interfaces and maintaining MCP (Multi-Modal Compute Platform) servers to facilitate agent interactions. Product roles will manage end-to-end agentic experiences and establish rubrics for agent performance. Marketing roles will target pipe generation and create agent-driven funnels to attract new users.

Enablement remains an extremely important function, but its type has become different.

Enablement now involves educating both human developers and autonomous agents. This requires creating content that is machine-readable and leveraging agent-friendly APIs to ensure that agents can effectively discover, understand, and utilize tools.

Maintaining human-to-human connection remains essential for community building, even as agents become part of the ecosystem. Developers should be able to bring their agents into these community loops, fostering a new kind of collaborative environment.

However, this introduces new challenges, particularly regarding privacy concerns when agents record community discussions. On the positive side, agents enable large-scale experimentation that is often difficult to conduct with human users, offering new avenues for data collection and product improvement.

Jarmak warned against using 'Claude slop'—low-quality, automated content that can alienate human developers who value authenticity and genuine expertise.

For agents, credibility is influenced by their bias toward their own generated content. Therefore, providing structured content, clear APIs, and authoritative sources is key for effective agent-facing interactions, ensuring that agents can reliably process and utilize information.

Just as curb cuts, originally designed for wheelchairs, now assist all pedestrians, optimizing for agents clears the path and enhances the experience for human developers as well.

She advocated for DevRel teams to start by directing coding agents to product documentation to identify friction points. Go-To-Market (GTM) teams should also experiment with agent prompts and recommendation strategies to understand how agents interact with and promote their products, leading to overall product and communication improvements.