Overview
Developer advocacy is entering a new phase because developers are no longer the only meaningful users of developer tools. Engineers increasingly orchestrate fleets of agents, non-engineers can now build with technical infrastructure, and agents themselves read documentation, call APIs, encounter errors, select libraries, and recommend products. The speaker therefore proposes “agent advocacy” as an extension of DevRel rather than its replacement. Its engineering dimension measures tool use through evaluations and traces, identifying friction such as confusing parameters, wasted turns, latency, and token costs. Its go-to-market dimension examines whether agents mention or recommend a product when users describe relevant problems, not merely when they explicitly shop for a category. A Sourcegraph experiment illustrates the gap: the product appeared in 65% of comparison-oriented responses but received zero mentions in a realistic pain-driven scenario. Addressing this requires current, authoritative, quotable content; agent-friendly interfaces; placement in agent marketplaces and registries; and a low-friction route from discovery to adoption. Traditional DevRel responsibilities—enablement, community, feedback, and credibility—remain essential, but must account for the distinct behavior of machines and humans. The central argument is that improving the agent experience functions like a curb cut: designed for one new user, it ultimately makes the product easier for everyone.
Sections
Key Concepts
Terms used to describe the evolution of developer relations and the emerging agent-facing discipline.
- Software evangelism: the largely one-way practice of promoting a software product and spreading its message.
- Developer advocacy: a two-way feedback function grounded in empathy for developers, translating their needs back into product decisions while supporting bottom-up adoption.
- Agent advocacy: an extension of developer advocacy that treats AI agents as users, recommenders, and participants in the product feedback loop.
- GEO: generative engine optimization, the practice of improving whether and how generative systems discover, interpret, mention, and recommend a product.
- Agent experience report: a practical assessment produced by directing an agent at documentation and examining its interaction trace for confusion, failures, recovery behavior, and friction.
- Curb-cut effect: the principle that accommodations built for a specific user group can remove friction and create broader benefits for everyone.
Strategic Implications
Higher-level conclusions that follow from the speaker’s examples and operating model.
- Agent advocacy collapses parts of product research and go-to-market measurement into one observable system: the same traces can reveal whether a tool is discoverable, understandable, usable, and efficient.
- Recommendation share measured through explicit category comparisons may be a vanity metric. Pain-based prompts are more representative because they test whether an agent can map a user’s situation to a product before the user knows which category to request.
- Error recovery can conceal interface debt. An agent that eventually succeeds may still impose meaningful costs through wasted turns, excess tokens, latency, and reduced recommendation confidence.
- The growth of generated content creates an information feedback loop in which obsolete product narratives can become more visible rather than naturally fading away.
- Human and machine credibility are diverging: humans may reject formulaic AI prose, while agents may respond favorably to highly structured content resembling their own outputs.
Forecasts
Anticipated changes in developer work, product discovery, and developer relations.
- Developers will increasingly act as orchestrators and supervisors of fleets of AI agents rather than working alone.
- AI fluency will become a standard expectation in engineering roles and job descriptions.
- Agents will drive a growing share of bottom-up tool adoption by recommending and installing libraries or services during active workflows.
- Developer advocacy will persist but expand to include machine-readable enablement, agent-oriented product interfaces, and experimental measurement of agent behavior.
- Agent participation in developer communities will create new privacy and data-governance questions when assistants observe or retain community conversations.
Humorous Moments
Jokes used to soften the provocative claim that developer advocacy is dying.
- After asking developer advocates to raise their hands, the speaker wonders whether they attended to throw tomatoes at her for declaring their profession dead.
- Because the speaker’s manager is himself a developer advocate, it seemed inappropriate for him to deliver the profession’s eulogy; an abandoned alternative involved the speaker dressing as a robot and mauling him onstage.
- The practical resolution to the staged robot attack was that the manager simply went on vacation.
- The speaker contrasts human dislike of formulaic AI prose with Claude’s apparent fondness for its own stylistic habits.
- She describes herself as a “data science scientist nerd person” before presenting measurement examples.