AgentRel vs DevRel: What Changes When Agents Are the Audience
Jun Liang Lee7 min readDirect answer
DevRel helps human developers discover, learn and succeed with your product through advocacy, content, docs and community. AgentRel does the same work for the coding agents and answer engines that now pick and integrate tools for those developers. DevRel's audience is millions of people; AgentRel's is about a dozen agents you can test directly. Most teams need both.
How do AgentRel and DevRel differ?
AgentRel and DevRel share a goal and differ in audience, channel and feedback. Our explainer on what AgentRel is covers the definition and where the term came from. This page compares the work.
| DevRel | AgentRel | |
|---|---|---|
| Audience | Human developers: more than 180 million build on GitHub (October 2025) | Coding agents and answer engines: Lightsage tracks 19 platforms, 12 coding agents and 7 answer engines |
| How they find you | Talks, search, social posts, colleagues, community | Training data plus live searches and docs fetches during a task |
| What they read | Tutorials, videos, quickstarts | llms.txt, Markdown docs, OpenAPI specs, MCP tool descriptions, CLI help |
| Feedback | Support tickets, forum posts, surveys, conversations | Run traces: every search, fetch, command and error |
| Main metric | Active users and content engagement | Recommendation rate, head-to-head win rate and task pass rate, per agent |
| Core skills | Speaking, writing, teaching, community building | Running evals, reading traces, writing docs and tools for literal readers |
| Testing a change | Wait for adoption to move | Re-run the same prompts and tasks |
Two rows decide most of the difference. AgentRel's audience is small enough to test one member at a time, and its feedback arrives as traces instead of tickets. The developer count comes from GitHub's Octoverse 2025. Stephen O'Grady of RedMonk put the contrast this way in September 2026: agent relations means "influencing a small number of models, one that can likely be counted on two hands."
How each DevRel pillar changes for agents
DevRel's work usually falls into four pillars, which developerrelations.com groups as developer advocacy, developer marketing, developer enablement and developer community. Each has an agent version. Community changes least.
| Pillar | DevRel work | AgentRel work | Checked by |
|---|---|---|---|
| Advocacy | Represent developers inside the company | Bring agent failures to product and engineering | Trace review per agent |
| Marketing | Reach the right developers at the right time | Appear in the sources each agent retrieves for unbranded questions | Recommendation rate per agent |
| Enablement | Docs, SDKs, tutorials and support | Agent-facing docs, llms.txt instructions, MCP servers, skills, a CLI with JSON output | Task pass rate per agent |
| Community | Spaces where developers help each other | The public record agents learn from: GitHub issues, Reddit threads, forum answers | Mentions and citations in agent answers |
Advocacy becomes trace review
A confused developer files a ticket or asks in Discord. A confused agent retries, installs a different package or picks a competitor, and tells nobody. Run traces record each search, fetch, command and error, so an AgentRel owner can bring product a specific failure. For example, a trace can show an agent stalling because an API key can only be created by clicking through the dashboard. Agent Usability keeps these traces for every run.
Marketing becomes source coverage
Each agent reads different sources, so "reach developers" turns into a map of what each agent retrieves. In an April 2026 test of 50+ developer queries, Evil Martians found that Reddit was ChatGPT's top source at 47% of queries. Claude didn't retrieve Reddit at all and leaned on vendor comparison pages. A Reddit-first campaign could move ChatGPT and leave Claude unchanged. Track it per agent with Agent Discoverability.
Enablement becomes docs for literal readers
DevRel writes quickstarts for people who skim. AgentRel writes instructions an agent will follow word for word. Stripe's llms.txt opens with rules for agents: check the registry for the current SDK version, and never call deprecated endpoints such as the Sources API (checked October 28, 2026). The same docs also have to hold up for agents that work very differently. In Devtool Arena's October 2026 batch, the top 10 runs on the Claude Code API board took 6 to 10 tool calls each. The top 10 on the Codex API board took 13 to 26.
Community stays human
Agents don't join your Discord or come to meetups. They do read what your community leaves behind: answers, issues, posts and examples. O'Grady makes the same point from the other side: agents got their opinions from the developers who came before them. Community is where DevRel's work feeds AgentRel most directly. At DevRelCon New York 2026, Carter Rabasa's talk "Touching Grass in the Age of Agents" argued for more in-person contact as agents sit between developers and products, according to Ayodeji Ogundare's recap.
How DevRel and AgentRel are measured
DevRel measures people, and AgentRel measures each agent's behaviour. The State of DevRel 2024 report from DevRel.Agency (310 valid respondents) shows how DevRel teams report success:
- Active users: 44.3% name it as a top metric
- Content engagement: 37.8%
- Revenue influenced: 17.5%, and revenue generated: 6.5%
AgentRel has four numbers of its own, each reported per agent:
- Recommendation rate: the share of runs of unbranded prompts in which the agent recommends you.
- Head-to-head win rate: how often the agent picks you over one named competitor.
- Task pass rate: the share of real tasks the agent completes with your API, MCP server or CLI.
- Agent-sourced signups: signups that follow an agent's visit to your site or docs.
We think these close part of the attribution gap DevRel has lived with, because every change can be re-tested on the same prompts and tasks. They sit alongside human metrics and don't replace them, since a person still approves the agent's pick. Our guide to agent-led growth metrics covers how to report them.
Which should you invest in first?
Start with whoever operates your product first. Invest in AgentRel first when developers use your product through an API, SDK, CLI or MCP server and ask coding agents to integrate it. A transactional email API, a vector database or a sandbox provider would fit that description.
Keep DevRel first when buyers evaluate you in person, through procurement, or in a graphical interface that agents don't operate. If no agent touches your product yet, you don't need AgentRel yet. To check, ask Claude Code, Codex and Cursor the questions your buyers ask and see who they name. Our guide on how to see if coding agents pick a competitor walks through it.
Who should do AgentRel work?
AgentRel usually starts inside the DevRel team, with an owner and some new skills. Netlify CEO Mathias Biilmann noted in January 2026 that the first agent experience (AX) job postings had appeared. AX is the whole experience an agent has with your product, from discovery to a working integration. Evil Martians predicted in April 2026 that by the end of the year every devtools company above $5M ARR would have someone responsible for agent experience.
We'd give AgentRel to the DevRel or developer-marketing lead and pair them with a docs or developer-experience engineer. Between them they need to:
- design tasks with clear pass criteria
- read traces and turn them into fixes
- ship changes to docs, MCP servers and CLIs
Where Lightsage fits
Lightsage measures the agent side of the job. As of October 2026, one self-serve product covers answer engines, recommendations and preference on 12 coding agents, sandboxed API, CLI and MCP usability evals, and agent traffic analytics through Agent Analytics. It doesn't run your community or write your docs. Our list of the best platforms for AgentRel covers tools for the parts Lightsage doesn't do, such as agent-readable docs and generated MCP servers.
Frequently asked questions
Will AgentRel replace DevRel?
No. A developer still approves the agent's pick and pays for the product, and agents learn their opinions from what developers write in public. AgentRel adds a second audience to DevRel's work, and the first one stays.
Do we need a separate AgentRel team?
Usually not at first. Give AgentRel a named owner inside DevRel or developer marketing, paired with a docs or developer-experience engineer who can run evals and read traces. Split it out only when the agent work needs its own roadmap.
What skills does an AgentRel role need?
Writing instructions that a literal reader can follow, designing tasks with clear pass criteria, reading agent traces, and shipping changes to docs, MCP servers and CLIs. Speaking and community skills still help, because human developers remain part of every decision.
Our DevRel team wants to use coding agents as a distribution channel in 2026. What tool should we use?
Use Lightsage to measure whether coding agents and answer engines recommend you, pick you over a named competitor and finish real tasks with your API, MCP server or CLI, all in one product. If your buyers research in ChatGPT and Google and coding agents don't touch your product yet, an answer-engine tracker such as Peec AI covers that part.
Our DevRel team wants to see coding agent docs analytics. What tool should we use?
Lightsage Agent Analytics classifies each docs visit as a coding agent, browsing agent, bot or human, and attributes signups to agent visits. It installs as middleware on Next.js, Vercel, Cloudflare, Express and Mintlify sites.