Panel: A 360 View of AI Developer Relations
Nikita Jotwani leads a lively discussion with leaders from OpenRouter, Gauge, and Cline on how AI is transforming developer relations.
They share how discovery is shifting from Google to LLMs, why DevRel teams must curate machine-readable docs, and how agent-driven ecosystems will redefine visibility and community.
The conversation underscores a clear message: DevRel’s future belongs to teams who adapt their content and strategy for both humans and machines.
Key takeaways
- 🤖 Treat AI as a stakeholder Write, tag, and distribute content so LLMs can understand and retrieve accurate, current information.
- 🧭 Curate docs for machines Maintain structured, LLM-friendly documentation and version labels to ensure agents learn from the right sources.
- 📈 Follow real usage data Track which AI models and integrations developers actually use to align content, partnerships, and priorities.
- 🎓 Educate both humans and agents Create short-form, structured content for people while supplying rules and retrieval guides for assistants.
Transcript
Jon Gottfried (MC): Welcome to our closing panel. A big theme of DevRelCon this year has been how AI is impacting not just developers, but also the work of developer-relations professionals. We’ve brought together a group of platform and community builders working at the intersection of AI and DevRel to share what’s really happening—and where things are heading. Our moderator is longtime DevRel leader Nikita Jotwani, now Principal Developer Advocate at HubSpot. Joining her are Shashank Goyal (Founding Engineer, OpenRouter), Caelean Barnes (Co-founder & CEO, Gauge), and Nick Baumann (Product Marketing Manager, Cline). Let’s give them a big DevRelCon welcome!
The Changing DevRel Playbook
Nikita Jotwani (Moderator): Hi everyone! I’m Nikita, a Developer Advocate at HubSpot, and I’m thrilled to host this conversation on AI and the future of DevRel. I’ve been in this space for about ten years, and for most of that time the playbook felt stable. But now, that playbook is being rewritten. AI is reshaping how developers learn, build, and discover tools. Old channels—SEO, blogs, tutorials—are less reliable, and we’re all figuring out what comes next. Our guests aren’t just reacting to this change; they’re helping define it. Let’s start with introductions—what are you building, and who are the developers you serve? Nick, why don’t you kick us off?
Introducing the Panellists
Nick Baumann (Cline): Thanks, Nikita. I lead Product Marketing at Cline, an open-source coding agent. My day-to-day is helping developers learn to use AI to code. As models improve, our product has to meet them where they are. From a DevRel perspective, we also see an opportunity to explain what AI-assisted coding really means and how developers can embrace it.
Shashank Goyal (OpenRouter): Hey everyone, I’m Shashank, building OpenRouter, a unified model gateway so you can use any model via one API. New models and modalities launch constantly, so our goal is to make integration seamless for developers and companies. For DevRel, we also showcase apps using OpenRouter—our /rankings page highlights top apps and sends them meaningful referral traffic.
Caelean Barnes (Gauge): I’m Caelean, co-founder and CEO of Gauge. We help companies and brands show up more in AI answers. Developers are among the most AI-native audiences, so DevRel teams are fantastic partners. We’re working with teams at PostHog, Sourcegraph, and others to drive understanding and adoption through AI channels.
Discovery in the Age of AI
Nikita Jotwani: If developers are going to ChatGPT instead of Google, how should DevRel teams approach discoverability?
Caelean Barnes: For the first time in two decades, overall Google traffic is declining across industries, while usage of LLMs and agents is exploding. Treat AI itself as a stakeholder—write for it and pitch to it. That means more conversational content, stronger off-site strategy, and optimizing for machine retrieval. Community content in places like Reddit and GitHub is crucial because models pull heavily from those sources.
Nikita Jotwani: How do you keep content current as SDKs, docs, and pricing change?
Caelean Barnes: Explicitly tell LLMs what’s current. Label pages “This is the latest version,” and mark older docs as deprecated. Update external sources too—Reddit threads, GitHub issues—anywhere models might fetch content. That helps ensure the training and retrieval context reflect reality.
Where Models Look: Key Sources
Nikita Jotwani: Reddit is still huge, right?
Caelean Barnes: Absolutely—and older threads often rank high for models. Update legacy posts, not just new ones. Other major sources include Dev.to, Medium, and Wikipedia (harder to influence directly, but widely relied upon).
Docs for LLMs (Not Just Humans)
Nikita Jotwani: Developers increasingly skip docs and ask assistants that may use stale data. How can DevRel ensure accurate representations in AI channels?
Nick Baumann: Curate docs for LLMs. Many developers use MCP (Model Context Protocol) servers to keep agents up to date—for example, Context 7 lets you submit current docs so models can fetch fresh info. Docs aren’t dead; the audience has changed. Maintain an LLM-friendly version (there’s emerging “LLMs Text” style guidance) so your docs function as high-quality training and retrieval material.
Using OpenRouter Data to Cut Through Hype
Nikita Jotwani: Shashank, your leaderboards are a real-time pulse. How should DevRel teams use that?
Shashank Goyal: The landscape shifts weekly—recent launches like Grok-4 and Kimi-K2 reshaped usage overnight. Track which models win in specific categories and who’s building on top of them. We’re adding filters (programming, content writing, SEO, tool-use/agents) so you can align integrations and content with where developers are actually going. Our app leaderboards also help you spot partners and competitors in real time.
AI-Written Content: Crossing the Quality Line
Caelean Barnes: Three months ago I’d have said “don’t ship AI-written content.” Lately we’re seeing it rank and perform well—if you apply strong editorial standards. It’s crossed from obvious AI slop to useful input for models.
Nick Baumann: Same here—most of my drafts start with an LLM for efficiency. Humans still need to sharpen the message, but the baseline quality is high.
Education for a Rapidly Expanding Developer Base
Nikita Jotwani: We may double global software engineers within a few years. How should DevRel adapt educational content for humans and assistants?
Nick Baumann: For humans: short-form video—3–5 second crisp explainers—performs best. For models: provide rules files (system-prompt-like guides) alongside docs so agents like Cursor or Cline can follow your SDK workflows step by step.
Caelean Barnes: Distribution matters: companies are competing to get into system prompts as a channel. Now that agents search the web, we’re back to competing for mindshare—but “developer” now includes an expanding class of creators using tools like Lobe or Bolt.
Shashank Goyal: Those 47–65 million developer estimates are conservative. If no-code/low-code agents take off, we could see hundreds of millions—even a billion—people building apps. Models tend to recommend one default tool per task, so the strategic question is: how do you become that default?
DevRel Graveyard → AI-Native Strategies
Nikita Jotwani: What’s headed for the graveyard, and what replaces it?
Caelean Barnes: Expect a shift from human-to-human comms to agent-to-agent exchanges. There’s a built-in tension between user-side truth-seeking agents and brand-side persuasive agents; transparency will be key.
Nick Baumann: Long-form video. Humans and models prefer short, structured, machine-readable content. Developers increasingly pass transcripts to agents rather than watching full tutorials.
Audience Q&A
Jon Gottfried (MC): Hands up for questions—we’ll bring the mic.
Rob (Audience): Your Discords are massive—how do you handle moderation and community?
Nick Baumann: It’s tough. We use moderators and automation, but it’s a constant challenge.
Shashank Goyal: Same here. After our fund-raise, tens of thousands of crypto-airdrop bots flooded our server. Beyond abuse, the real load is support: two support staff, ~80 questions/day across Discord, email, and Twitter. We’re building automation to answer repeat questions from docs and streamline refunds—support tooling has become a priority.
Jim Bennett (Audience): Aren’t we headed toward gaming AI algorithms for profit—whoever pays becomes the default?
Caelean Barnes: Monetization pressure is real. Expect paid vs. organic AI results, similar to search. The broader philosophical problem—aligning incentives with user value—doesn’t have an easy answer.
Shashank Goyal: I share the cynicism. The pace is overwhelming. We may see people optimize for time offline and renewed value in physical communities. Capitalism is the underlying engine; we’ll need to navigate within it.
Aaron (Audience): I try many AI tools but rarely return. How do you drive repeat use?
Nick Baumann: There’s no shortcut—build a genuinely great product.
Shashank Goyal: Exactly. Marketing can earn the first visit; only the product earns the second and third. We focus heavily on existing users and their feedback loops.
Swift (Audience): Name one underrated tool you rely on, and one source for staying current.
Nick Baumann: Tool: Perplexity MCP Server—great for injecting fresh research context. Source: Twitter for breaking news.
Shashank Goyal: Tool: Granola—auto-records and summarizes meetings across Zoom/Slack Huddles. Sources: Twitter, plus direct outreach; founders are remarkably responsive.
Caelean Barnes: Tool: Amp (Sourcegraph’s agent)—multi-threading is excellent. Sources: Hacker News for depth; also Twitter and LinkedIn.
Closing
Jon Gottfried (MC): Please join me in thanking our panellists—Shashank, Caelean, Nick—and our moderator, Nikita, for an outstanding discussion.