Agent-Led Growth vs Product-Led Growth: A Real AgentMail Example

Direct answer

Product-led growth uses the product experience to help acquire, activate, and retain customers. Agent-led growth focuses on journeys where an agent discovers, selects, or uses a product on a customer's behalf. They can work together: a product-led business can serve customers directly and through agents. The AgentMail example below shows delegated product use, not customer acquisition.

---

The practical difference becomes clearer when you follow a task. In a real Devtool Arena run, Claude Code read AgentMail's documentation, installed its Python SDK, and wrote a script that created an inbox and sent an email. The output included the new inbox and a matching sent-message record.

That run shows what delegated product use looks like—and which parts of the experience a devtool team needs to make work. For the broader growth model, see our introduction to agent-led growth.

The customer owns the goal. The agent carries out the task.

Imagine a developer using an email service directly. They read the instructions, install the SDK, write a script, and inspect its output. If a dependency is missing, they install it before continuing. This is an illustrative journey, not a measured comparison with the run below.

When the developer delegates, an agent may carry out those operations instead. The product's instructions, tools, and responses now need to help the agent reach the same outcome.

This does not make PLG synonymous with a person clicking through a dashboard. PLG describes how the product experience drives growth, and that experience can include programmatic use. The distinction here is who performs the work within it.

What Claude Code was asked to do

The task named AgentMail, supplied its documentation URL, and asked for a script that would create a new inbox, send a test email to a specified recipient, list the inbox's messages, and print the results as JSON.

An API key was already available in the environment. The agent did not have to choose an email provider or sign up for an account. The question this run answers is narrower: could it turn the instructions into working code?

Open the AgentMail API report on Lightsage's Devtool Arena. Select Oct 2026 and look for the run completed October 15, 2026. Its generated file is inbox_manager.py. The Journey tab shows the tool calls and responses; Execution Output shows the script's result. The page may show a newer run later.

The agent went from documentation to a working script

Claude Code first checked whether the AgentMail Python package was installed. The check returned ModuleNotFoundError: the environment had credentials, but lacked the SDK needed to use them.

It then read AgentMail's quickstart, retrieving the installation command and examples for creating an inbox, sending a message, and listing messages. This gave it a concrete path from the requested task to the relevant SDK operations.

Claude Code installed the package and wrote inbox_manager.py. The script loaded the supplied API key, created an inbox, sent the requested email from it, and listed that inbox's messages. It collected the responses into one JSON result.

The agent then ran the script. The useful sequence is simple: read the instructions, install the dependency, write the integration, run it, and inspect the result. The customer had delegated the implementation work that a developer would otherwise do directly.

The output showed what actually worked

The execution output contained three pieces of evidence:

• A new inbox: the response included an inbox identifier and creation timestamp.
• A send response: AgentMail returned a message identifier and thread identifier.
• A matching message record: the inbox's message list contained one message labeled sent, with the requested recipient and subject. Its identifiers matched the send response.

Those records support the conclusion that the script created an inbox and sent the requested message through AgentMail. They do not show receipt in the recipient's inbox or whether anyone read the email. If delivery were the success criterion, the test would need that additional evidence.

The lesson is practical: a working integration should leave a result the customer can inspect. Here, the documentation and SDK helped the agent reach that result, and the returned records made it checkable.

What this means for an ALG strategy

In the illustrative direct-use journey, the developer would read the quickstart and build the integration. In this run, Claude Code performed those steps using AgentMail's Python SDK.

For a devtool team, that expands the experience worth investigating. A clear dashboard may help the person, while a clear quickstart, an installable SDK, and checkable responses help the agent operating for them. Both contribute to whether the customer gets a working result. Our guide to Agent Experience covers these stages from discovery through completion.

The possible growth benefit still needs testing. Reliable delegated use may give customers a reason to adopt and keep using a product. This run demonstrates task completion; it does not measure acquisition or retention, and AgentMail was not discovered or chosen spontaneously. To test consideration separately, use an open-ended task and inspect the responses through Agent Discoverability.

To apply the lesson, choose one task that represents your product's first useful outcome. Follow an agent's attempt from the initial instructions to the evidence of success. When it stops, inspect the response it received and whether a valid next action was available.

Lightsage's Agent Usability helps teams inspect those execution paths. Our guide to getting recommended by coding agents explains how to turn a finding into a product improvement and test it again.

Frequently asked questions

What is the difference between agent-led growth and product-led growth?

Product-led growth uses the product experience to drive acquisition, activation, and retention. Agent-led growth focuses on journeys in which agents discover, select, or use products for customers. A product-led business can support both direct and delegated use.

Did Claude Code choose AgentMail on its own in this run?

No. The task specified AgentMail and provided its documentation URL. An API key was already available. The run demonstrates setup and product use after product selection; it does not show independent discovery or account signup.

Does agent-led growth replace product-led growth?

No. A customer can use a product directly for some tasks and delegate others to an agent. The product experience can support growth through both journeys.