Designing AI-Agent Interfaces: Lessons from Resend’s Email Platform

Machine-readable pricing drives selection

Resend’s pricing page includes a slider that, when “.md” is appended, returns a markdown table in a single request. This format lets language models query price points directly, a trick the podcast notes as “the kind of change I liked in our Supabase interview.” By exposing structured data, Resend increases the likelihood that agents such as Claude Code will surface it first in recommendation lists.

Tool proliferation and context management

The host observed that expanding the tool set does not automatically reduce the size of responses; a single call can still exceed the model’s context window. Rocha suggests a simple experiment: expand a routine call’s result and measure how much the model actually consumes, informing decisions about tool granularity.

Post-send considerations and abuse mitigation

Beyond selecting an SDK, Resend tackles click-tracking safety. Shared tracking domains can be blocklisted if abused, harming all customers. To avoid this, Resend offers custom tracking domains and maintains a free tier despite the risk of spam sign-ups. The company backs this with a $3 million safety commitment, underscoring a willingness to absorb abuse costs to preserve deliverability.

Our read: Resend’s strategy shows that exposing clean, machine-friendly data and carefully curating tool outputs can tip AI agents toward a smaller provider, while a pragmatic safety budget mitigates the downstream risks of open-access features.