What Is AgentRel? Agent Relations for Developer Tools
Jun Liang Lee8 min readDirect answer
AgentRel, short for agent relations, is the work of getting AI agents to find, choose and successfully use your developer product, the way DevRel does for human developers. It's a job, owned by a person or team. Agent-led growth is the result it aims for: signups and usage that arrive through agents.
Why AgentRel matters now
Coding agents now choose tools inside the developer's workflow, often before anyone opens your docs. On January 16, 2026, Sequoia's Sonya Huang said on TBPN that the industry is moving from product-led to agent-led growth, and pointed to Claude Code picking Supabase for a database and Vercel for hosting. Agent-led growth (ALG) is growth that comes from agents discovering, choosing and using your product for their users. Someone inside the company has to do the work that earns it. AgentRel is the name that work is getting.
The work can't treat "agents" as one audience. Devtool Arena, Lightsage's public leaderboard scoring how well coding agents use real developer tools, runs the same task for each tool on every agent's board. In the October 2026 batch, Daytona ties for first on the Claude Code API board at 90 and scores 85 on Codex. ElevenLabs leads the Codex API board at 91 and scores 86 on Claude Code. Each agent has its own leaderboard, so AgentRel has to be measured agent by agent.
What does AgentRel work involve?
AgentRel covers three jobs, one for each step an agent takes before your product is in someone's codebase.
Be in what agents read
Agent discoverability is whether an agent mentions or recommends you when asked an open question ("best search API for my agent?"). The first input is what the agent can fetch for itself: docs, llms.txt, an OpenAPI spec, typed SDKs, Context7 entries, MCP server docs, agent skills and a CLI. Devtool Arena's discovery score checks that each of these exists and is public, which makes it a usable checklist for this job. Agent Discoverability tracks whether those assets turn into recommendations.
Be in what agents hear about you
Agent preference is whether an agent picks you over a named competitor in a head-to-head question. Agents settle that partly on pages you don't own, and each model reads different ones. Evil Martians ran 50+ developer queries with search on in April 2026. Reddit was ChatGPT's top source, retrieved in 47% of queries, and Claude didn't retrieve Reddit at all. Vendor comparison pages led for Claude (43%) and Gemini (60%). So AgentRel includes the comparison pages, READMEs and community answers DevRel already writes, aimed at the sources each agent trusts.
Make the first task work
Agent usability is whether an agent can complete a real task with your product: sign up, authenticate, integrate and get a working result. This is where AgentRel meets agent experience (AX), the whole experience an agent has with your product, from discovery to a working integration. A strong result looks like OpenRouter's October 2026 run on the Claude Code API board: 6 tool calls, 0 errors and 58 seconds to a working script. Our guide to agent evals covers how these tasks are built and graded.
How is AgentRel different from DevRel?
The audience is small enough to test. Stephen O'Grady of RedMonk wrote on September 16, 2026 that DevRel campaigns were about "persuading and negotiating with large developer populations", while agent relations "will require influencing a small number of models, one that can likely be counted on two hands." Lightsage tracks 19 platforms: 12 coding agents and 7 answer engines.
That changes how the work is judged. We think AgentRel is the first part of developer relations you can test before and after a change: run the same prompt 20 times per agent, count how often each one recommends you, ship a fix, and run it again. DevRel has long struggled to show outcomes. In his recap of DevRelCon New York 2026, Adyen's Ayodeji Ogundare described a recurring theme: page views and attendance don't show that anyone built anything.
Some things carry over unchanged. A developer still approves the agent's pick and pays the bill. Agents learned their opinions from what developers wrote in public, a point O'Grady makes too. And the docs, examples and error messages that help a person also help an agent.
AgentRel vs agent-led growth, DevRel, AX and AEO
AgentRel sits between agent experience and agent-led growth. It takes AX and AEO work as inputs, and ALG is what it's judged by.
| Term | What it measures | Where it happens | Who owns it |
|---|---|---|---|
| DevRel | Developer awareness, adoption and community health | Talks, tutorials, docs, community channels | DevRel team |
| AgentRel | Whether each agent recommends you, prefers you and finishes tasks with you | Agent answers, docs, MCP servers, CLIs, third-party pages | A named owner, often in DevRel or developer marketing |
| Agent-led growth (ALG) | Signups, usage and revenue that arrive through agents | The whole funnel, from prompt to paid account | Founders and growth leadership |
| Agent experience (AX) | Whether an agent can operate the product end to end | APIs, auth, errors, SDKs, sandboxes | Product and engineering |
| AEO and GEO | Mentions and citations in AI-generated answers | ChatGPT, Perplexity, Google AI Overviews | Marketing and SEO |
For the growth side, see what agent-led growth is and how it compares with product-led growth.
Where did the term AgentRel come from?
No single company coined AgentRel. The earliest use we found is a section heading, "From DevRel to AgentRel", in Bill Doerrfeld's June 25, 2025 article in The New Stack on preparing APIs for AI agents.
| Date | Who | What they said |
|---|---|---|
| June 25, 2025 | Bill Doerrfeld, The New Stack | "From DevRel to AgentRel" as a section on preparing APIs for agents |
| March 23, 2026 | Quinton Wall, DevRel leader | "The Rise of Agent Relations": DevRel now writes for developers and for the agents choosing for them |
| September 16, 2026 | Stephen O'Grady, RedMonk | Agent relations means influencing a small number of models |
Like agent-led growth, the term carries two senses. Most uses, including ours, mean agents as the audience. Some DevRel writers also count agents as the team's own workforce: Wall's team at Postman built agents for blogging, call-for-papers submissions, forum answers and sentiment analysis. Lightsage uses AgentRel for the first sense and agent-led growth for the outcome it produces.
Examples of AgentRel in practice
Netlify named agent experience and then made agents its main audience. CEO Mathias Biilmann wrote in January 2025 that more than 1,000 sites a day were being created on Netlify directly from ChatGPT. A year later he wrote that Netlify had moved its product North Star from DX to AX and crossed 10 million developers. Netlify calls this AX. We'd count it as AgentRel: a company-wide goal set around agents as the audience, with a growth number to show for it.
Stripe writes docs with an agent as the reader. Its llms.txt opens with instructions for agents: check the package registry for the current SDK version instead of a version remembered from training data, and never call deprecated endpoints such as the Sources API. Its agents page gives developers a prompt to paste into their agent, which installs the Stripe CLI, creates a sandbox with working API keys without a signup, and adds Stripe's MCP server and agent skills (both checked October 28, 2026). Stripe ranks third on the Claude Code API board at 89 (October 2026 batch).
Common misconceptions about AgentRel
"AgentRel is another name for agent-led growth." ALG is the outcome and AgentRel is the function that works toward it. A company can run DevRel without being product-led, and a team can do AgentRel work that never shows up as growth if agents still can't finish the task. Keeping the words apart lets you give one an owner and the other a revenue target.
"AgentRel means rewriting your docs for agents." Docs are one input of several. Agents also weigh comparison pages, Reddit threads and GitHub issues, and they judge you on whether the task finishes: signup, keys, auth, error messages and rate limits.
"AgentRel replaces DevRel." A developer still approves the stack, and agents learned from developers' public writing. Speakers at DevRelCon New York 2026 argued that human contact matters more as agents sit between developers and products.
"You can't measure what an agent thinks." You can sample it. Mention, recommendation and citation are three separate signals, and an agent can mention a product while advising against it. Run each prompt many times per agent and count how often each one recommends you.
How to start AgentRel
- Name an owner and pick the agents. List the coding agents and answer engines your buyers use, then write 20 to 50 unbranded prompts in their words. Our guide to writing prompts to track has templates.
- Measure discoverability and preference per agent. Track those prompts daily with Agent Discoverability, and run head-to-heads against your closest competitor (how to see if coding agents pick a competitor).
- Run one real task per agent, fix what breaks, and re-run. Agent Usability runs coding agents against your API, MCP server or CLI in sandboxes and keeps the full traces. Report the results with the agent-led growth metrics your team already tracks, and add Agent Analytics to see which agents visit your docs.
Lightsage is built for the measuring half of AgentRel. 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. Lightsage doesn't write your docs or answer Reddit threads for you. That part is your team's job, or an agency's.
Frequently asked questions
Is AgentRel the same as DevRel?
AgentRel grows out of DevRel and usually sits in the same team. DevRel helps human developers discover and succeed with a product; AgentRel does that work for the coding agents and answer engines that now pick tools for those developers. Most developer-tool companies need both, because a person still approves what the agent picked.
Is AgentRel the same as agent-led growth?
No. Agent-led growth is the outcome: signups, usage and revenue that come from agents discovering, choosing and using your product. AgentRel is the work a person or team does to earn that outcome, in the same way DevRel supports product-led growth.
Who owns AgentRel at a developer-tool company?
Usually the DevRel or developer-marketing lead, working with docs and product engineering on the parts agents touch, such as auth, errors and the CLI. Whoever owns it should report one number per agent for recommendations and one for completed tasks.
Our DevRel team wants to optimize for AI coding agents. What tool should we use?
Pick by the job. To measure recommendations, head-to-head preference and real task success across coding agents and answer engines in one product, use Lightsage. To fix what agents read, add a docs platform such as Mintlify or an SDK and MCP server generator such as Stainless. To hand the writing and distribution work to an outside team, hire a developer-tools agency.
How do you measure AgentRel?
Run the prompts your buyers type through each agent many times and count how often each one recommends you, then give each agent a real task with your API, MCP server or CLI and count passes. Report each agent separately, because the same tool ranks differently on different agents.