DevRel in the AI Era

Between 2023 and 2026, LLMs, coding agents, and agent-mediated discovery changed what Developer Relations does, who it serves, and how teams measure the work. This section covers those changes, the evidence for them, and the disputed claims around newer practices.

The central claim

Modern DevRel serves two distinct audiences:

1. AI agents. Cursor, Claude Code, Copilot, ChatGPT, Perplexity, Gemini, and background coding agents use docs, sample code, schemas, and sometimes llms.txt files to produce code on behalf of developers. They are the execution audience and perform better with complete, structured, machine-readable material.

2. Human developers. They may encounter a product through X, Bluesky, YouTube, Twitch, Reddit, Hacker News, Discord, podcasts, conferences, or an AI answer before reading its documentation. They decide whether to evaluate, adopt, and recommend the product, and need credible explanations, demonstrations, reasons to care, and accountable technical judgement.

Different materials serve each audience. Reference documentation and schemas support retrieval and execution; talks, demonstrations, comparisons, and authored explanations help people evaluate a product. See Humans and agents.

What changed between 2022 and 2026

Dimension	2022	2026
First-touch discovery	Often Google search → docs	Search, AI assistants, social, video, and peer recommendations
Code authorship	Developer types code	AI agent writes code under developer supervision
Doc reader	Primarily human	Human readers plus retrieval and coding agents
Documentation surface	Web page	Web page + llms.txt + MCP server
Integration interface	REST/GraphQL API + SDK	Same + MCP server + agent-friendly tool schemas
Onboarding metric	TTFHW for human	TTFHW for human + agent task success rate
Discovery channel	Google + Twitter + HN	Google + ChatGPT + YouTube + LinkedIn + Bluesky + Reddit + podcasts
Evaluation evidence	Stars, follower counts, recommendations	AI citations, signed technical work, peer recommendations, working examples
Distributed formats	Blog posts and documentation	The same, plus transcripts, streams, podcasts, and agent-readable schemas

Each of these shifts is treated in detail in subsequent files.

What this section covers

Foundations

• Humans and agents: Why DevRel now has two audiences and how to serve both.
• LLM-mediated discovery: How developer research moved from Google to LLMs.
• Human inspiration: The role of identifiable people on YouTube, X, podcasts, and conference stages.

Agent-facing surfaces

• llms.txt: The llms.txt convention: origin, adoption, evidence, scepticism.
• MCP for DevRel: Publishing and operating an MCP server for a developer product.
• Agent experience: Agent Experience as an emerging discipline alongside DX.
• Agent-facing documentation: Writing docs both humans and machines can use.

Optimisation and measurement

• GEO and AEO for DevRel: Generative / Answer Engine Optimisation tactics.
• AI-era DevRel metrics: Share of Model, prompt feedback, citation tracking, AI-mediated activation funnels.

Identity and culture

• Vibe coding and AI engineers: New developer identities; Karpathy’s “vibe coding”; swyx’s “AI Engineer.”

The debates

• AI-era DevRel debates: Named voices and what they argue.
• Critical views: “DevRel is dead because AI” arguments and the rebuttals.
• AI-era DevRel signals: Concrete, observable patterns of what works in 2026.

Five positions in the debate

Discussion of AI-era DevRel contains five recurring positions:

1. The Dual Audience. Practitioners design separately for human evaluation and agent execution. This position appears in work from Anthropic, Mintlify, Vercel, AngelHack, and DocE-AI from 2025 to 2026.

2. The Strategic-Reframe Camp. Argues that AI expands DevRel’s role. Angie Jones writes, “Plot twist: we’re not dead. We’re standing on the biggest stage of our careers” in How DevRel is Leading AI Adoption (2025). In this account, DevRel models credible AI-augmented practice, compares developer perception with telemetry, and curates trust signals that AI assistants cannot produce.

3. The Restructuring Camp. Argues that the 2022 to 2024 contraction exposed DevRel programmes built mainly around content volume and that AI accelerated their reorganisation. MB Consulting’s RIP DevRel 2010 to 2024 (2024) is one example.

4. The Agent-Experience Camp. Focuses on designing APIs, documentation, MCP servers, prompt templates, and errors for use through AI agents. AI-product and infrastructure companies that publish MCP servers provide current examples.

5. The Sceptical / Empirical Camp. Cautions that many AI-era DevRel claims remain unverified. llms.txt may have negligible measurable impact, AI-citation studies show inconsistent results, and productivity surveys can diverge from telemetry. Signals.sh’s Does llms.txt actually work? (2026) is one example.

Teams can combine these positions: maintain separate material for people and agents, use AI in DevRel operations, build agent-facing interfaces where the product requires them, and measure newer tactics instead of assuming they work.

What to read first

If you have time for only three of the files below:

1. Humans and agents
2. Agent experience
3. AI-era DevRel signals

If you have time for one, read the dual-audience thesis first. The other pages develop its two sides in detail.

See also

• AI and LLMs in DevRel: The earlier, briefer treatment of AI’s impact on DevRel.
• AI and ML companies: AI companies’ DevRel programs.
• Documentation as product: Documentation strategy generally.
• DevRel signals to watch: Broader future patterns.