Building for Agents: Resend's Strategic Pivot
Resend CEO Zeno Rocha discusses the critical shift from developer experience to agent experience. Learn how standardizing API primitives, serving markdown pricing pages, and implementing granular permissions are essential for capturing the emerging market of autonomous AI agents.
The Shift from Developer to Agent Experience
The software industry is undergoing a fundamental transition from optimizing for human developers to optimizing for autonomous AI agents. Resend, a leading email service, has identified that traditional developer experience (DX) metrics, such as animated icons and keyboard shortcuts, are irrelevant to agents. Instead, success now depends on "Agent Experience" (AX), which prioritizes machine-readable data, standardized protocols, and deterministic outputs. This shift represents a new competitive landscape where the ability to be consumed by LLMs is as critical as usability for humans.
Strategic Implications for SaaS Infrastructure
Resend’s approach highlights several key strategic moves for infrastructure providers. First, content must be served in formats that agents can easily parse. Resend discovered that LLMs could not interact with JavaScript-enabled pricing sliders, leading to incorrect market positioning. By serving a markdown version of the pricing page via content negotiation, they ensured accurate data retrieval. Second, API standardization is non-negotiable. Proprietary headers for rate limiting or authentication confuse agents, whereas RFC-compliant standards allow for automatic retry logic and error handling. Additionally, publishing OpenAPI specs in JSON within the .well-known directory significantly improves discoverability for agent tools.
Operational and Security Considerations
The rise of agents introduces new operational challenges. Agents generate significantly higher request volumes than humans, necessitating robust rate limiting and granular permissioning. Resend is rethinking its authentication layer to support OAuth and granular scopes, moving away from simple API keys. This is crucial for enterprise adoption, where security and least-privilege access are paramount. Furthermore, the company is building closed-loop systems where agents analyze user feedback and automatically generate code fixes, turning operational data into product improvements.
Conclusion
The future of SaaS lies in building for a dual audience: humans and agents. Companies that fail to standardize their interfaces and data formats will be invisible to the growing ecosystem of autonomous tools. Resend’s strategy demonstrates that embracing these changes early can drive significant growth, as agents increasingly recommend and integrate with services that are easy to consume programmatically. The focus must shift from visual polish to structural clarity and standardization.
Key insights
-
Traditional developer experience features like animations and keyboard shortcuts do not translate to agent experience. Agents require structured, machine-readable data and standardized protocols to function effectively.
Impact: Companies must re-evaluate their DX investments to prioritize AX, focusing on data accessibility and protocol compliance to remain competitive in the AI-driven market.
-
LLMs struggle with dynamic, JavaScript-heavy interfaces, leading to inaccurate data retrieval. Serving static, markdown-based versions of key pages via content negotiation resolves this issue.
Impact: Implementing content negotiation for markdown can improve the accuracy of AI recommendations and reduce misinformation about product pricing and features.
-
Proprietary API headers and non-standard formats create friction for agents. Adhering to RFC standards for rate limiting and authentication allows agents to automatically handle errors and retries.
Impact: Standardization reduces the cognitive load on agents and developers, leading to smoother integrations and higher reliability in automated workflows.
-
Agents and enterprise users demand granular permissions rather than binary access. Simple API keys are insufficient for complex, multi-step agent tasks that require specific scopes.
Impact: Implementing granular OAuth scopes and API key permissions is essential for enterprise adoption and safe agent deployment in production environments.
-
The definition of a 'developer' is expanding to include non-technical users leveraging AI tools. This broadens the total addressable market but requires simpler onboarding and error handling.
Impact: SaaS companies must design for a wider audience that includes 'vibe coders' and non-technical builders, ensuring low friction and clear error messages.
Action items
-
Audit your API documentation and pricing pages to ensure they are accessible via markdown or JSON formats. Implement content negotiation to serve these formats to non-human user agents.
Impact: This will improve the accuracy of AI-generated recommendations and ensure your product is correctly represented in LLM-driven workflows.
-
Review your rate limiting and error response headers to ensure they comply with RFC standards. Replace any proprietary headers with standard formats that agents can automatically interpret.
Impact: Standardized headers enable agents to implement automatic retry logic, reducing failed transactions and improving the reliability of your API for automated clients.
-
Publish your OpenAPI specification in JSON format within the .well-known directory of your domain. Ensure this file is kept up-to-date via CI/CD pipelines.
Impact: This improves discoverability for agent tools and MCP servers, making it easier for them to understand and interact with your API capabilities.
-
Develop granular permission scopes for your API keys and OAuth tokens. Move away from binary access models to allow agents to request only the specific permissions they need.
Impact: Granular permissions enhance security and meet enterprise compliance requirements, making your platform more attractive to large-scale agent deployments.
-
Design your CLI tools to support non-interactive modes with structured JSON output. Ensure that all commands can be executed and parsed programmatically without human input.
Impact: This enables agents to use your CLI as a primary interface for executing tasks, increasing the utility of your product in automated pipelines.
Quotes
“AI agents, they cannot drag a pricing slider.”
“We had to go back to the basics, the CLI, the OAuth layer, the right headers, the right ATP verbs.”
“If you follow the RFC, agents will know, they will know how to auto retry.”