Latent Space

Zeno Rocha on why agents need your API, not your app

Zeno Rocha· Co-Founder and CEO of Resend at Resend
·~86 min·English·Latent Space
AgentsBusiness StrategyAI InfrastructureOpen Source
TL;DR

The Resend founder on reinventing email deliverability by trusting senders by default, going down to the metal on infrastructure, and making the whole product formless — API-first, MCP-covered, and readable by agents that can't run your JavaScript.

01Deliverability Philosophy

Born Free, Not Born in Prison

<strong>The established email platforms make a new sender prove they aren't a spammer before a single email goes out.</strong> Resend inverts it — trust every sender by default, then block fast the moment the signals go bad.

it's almost like if you were born in a society where you first go to prison and then you have to spend like 18 years in prison. you got to prove that you are a good citizen and then you get out of prison and now you can do the things you want to do.

— Zeno Rocha, Latent Space
Key Insight
Trust-by-default is only survivable because the detection got fast and automated. The gate didn't disappear — it moved from the front door (human, slow, blunt) to continuous signal-watching (bounce and complaint rates, automated, precise). The naivety was the feature: with no preconception of how it is done, they could ask why anyone is blocked before they have sent anything at all.
02Inbox Physics

Do the Exact Opposite of an Abuser

<strong>A legitimate sender does the exact opposite of a spammer</strong> — warm up slowly, ramp consistently, and send to the most engaged readers first.

you you need to do exactly what an abuser wouldn't do. It's the opposite of that.

— Zeno Rocha, Latent Space
Key Insight
Because Gmail throttles a big send and reads the first batch's engagement — opens, archives, deletes — before deciding how to treat the rest, the order you send in changes deliverability: front-loading the most engaged readers sets a higher ceiling for everyone after them (Outlook and Yahoo each behave differently).
03Product Philosophy

Formless: Give the API, Bind Nothing

<strong>You cannot predict which interface wins, so make the product formless — expose everything through the API and let the client be anything.</strong>

I want my product to be formless. whatever they want to do with the product, it should be possible.

— Zeno Rocha, Latent Space
Key Insight
The agent era rewarded this by accident of history: Resend was API-first long before MCP existed, so when agents arrived there was nothing to retrofit. The warning sits in the mirror — frameworks that bind data straight to the page make the performant default the one with no public API, which is fine for a human dashboard and a dead end for an agent.
04Infrastructure Arc

Start on Rented Infra, Then Own the Metal

<strong>Begin on someone else's infrastructure to move fast, then own the metal once scale makes the hidden costs hurt.</strong>

we're building the infrastructure and going like down to the metal now.

— Zeno Rocha, Latent Space
Key Insight
The arc isn't unique to email — it recurs across developer-experience companies. A thin layer on someone else's infra is often the right start (you ship, you learn, you scale) and, past some scale, the wrong end, because you inherit their limits and a sprawling bill: SES alone pulls in SNS, CloudWatch, SQS and your own webhook plumbing. You earn the right to own the metal once that hidden sprawl costs more than replacing it.
05Agent-Era Distribution

The Official MCP Is the New Buy Signal

<strong>For Zeno, a vendor-owned MCP in the connector store has become a real purchase decision, not a nice-to-have</strong> — and Resend's own MCP calls roughly double every month.

I need the owner of the software to own this piece of integration.

— Zeno Rocha, Latent Space
Key Insight
This inverts a decade of SaaS distribution. Being ranked by Google used to be the growth channel; being an owned, official entry in an agent's connector list is the new one. Zeno moved real spend — dropping Plausible for PostHog — on provenance alone: who maintains the integration closed the deal, not what the product's features were.
06Agent API Design

The 100% Coverage Bet

<strong>Resend runs counter-consensus: 80-plus MCP tools and growing, not a minimal two.</strong> Getting from 90% to 100% coverage — billing, provisioning, deletion — is where agent-API design actually gets hard.

Getting to 100% coverage of what your dashboard can do is actually very very difficult because now you're doing usage billing team invites account provisioning deletion should that be exposed as an API or not?

— Zeno Rocha, Latent Space
Key Insight
The two camps aren't really arguing about tool count — they're arguing about who does the reasoning. A couple of tools (search plus execute) push the reasoning onto the model; dozens of typed tools push it into the API surface. Resend bets that for actions with real side effects — billing, deletion — an explicit tool is safer than hoping the model composes the right call.
07AEO in the Agent Era

Agents Can't Run Your JavaScript

<strong>An LLM read Resend's pricing off a JavaScript calculator, couldn't run it, and called the pricing “obscure”</strong> — so Zeno spelled the numbers out in plain Markdown and the model got them right.

JavaScript widgets they're not going to be executed so I spell out the same data in text so that uh LLMs can actually read that thing.

— Zeno Rocha, Latent Space
Key Insight
Answer-engine optimization is partly just accessibility for non-human readers: anything gated behind client-side execution — widgets, charts, images — is invisible to a model that only fetches and reads. The deeper, less controllable lever is recency plus ubiquity: newer tools get recommended more, and tools talked about in many places get fetched more. You can author the first (Markdown, FAQs, llms.txt) far more easily than the second.
08The Values Decision

$10M Where My Daughter Plays With Me

<strong>Zeno would rather run a $10-million company where his daughter still wants to play with him than a hundred-million one where she doesn't.</strong>

I would rather live in that reality than live in a reality where we're hundred million dollar company and my daughter doesn't want to play with me.

— Zeno Rocha, Latent Space
Key Insight
What keeps this from being a platitude is that Zeno frames it as a concrete, acknowledged gamble — a $10-million company his daughter is close to, chosen over a hundred-million one she isn't — and admits he cannot know if it's right. It reframes the number every founder is told to maximize, revenue, as the one he is willing to cap.