Increase your agent reliability by catching failures at setup and runtime | Ampersand Changelog

Increase your agent reliability by catching failures at setup and runtime

The difference between a vibe-coded demo and trustworthy software is how reliably it runs in production, and reliability starts with error handling.

Ampersand now gives your agentic product two layers of protection: Sampling catches configuration problems during setup, and Notifications alert your system to failures at runtime.

Catch installation failures at setup with Sampling

Many builders are finding that their customers make a successful connection to Salesforce, but then realize it doesn’t work. The first time data is pulled is often the most critical moment, and it’s where an installation fails to pull data.

Our customers/builders asked: “Can we catch these issues during setup?” What if we could catch these errors before customers finish their configuration, rather than after the customer has moved on to other tasks?

Introducing Sampling. Sampling runs a test read to confirm that permissions and field mappings are correct before the customer finishes configuring their installation. Now configuration issues surface at setup, before the customer even leaves the page.

Without sampling: connect account → configure fields → finish setup → queue data sync → the first sync hits a permission error, and the customer has already moved on.

With sampling: connect account → configure fields → a test read runs → the error is caught so the customer can fix it right there, before leaving the page.

Sampling is now on by default. It runs automatically in the InstallIntegration component (@amp-labs/react v2.13.5+). You can also opt out with skipSampling when creating an installation via API.

Get notified of failures at run time with Notifications

Even with Sampling in place, integrations can still fail at run time. If a customer’s integration stops working, or never started in the first place, you want to know as soon as possible.

Previously you had to check the Ampersand dashboard or poll the Ampersand API to find out something was wrong. Now you can set up notifications on specific errors, so your system is alerted the moment a common integration error happens.

The best way to improve your customers’ experience is to build your system or agent to listen for these notifications, so you aren’t scrambling at run time.

Ampersand supports these notification types to enhance your agentic integration experience:

GroupEventNotifies you when
Connectionconnection.createda customer connects an account
connection.error a connection breaks (for example, expired or invalid credentials)
connection.refresheda connection is re-authenticated and healthy again
connection.deleteda connection is removed
Readread.schedule.paused a scheduled read is paused, usually due to an error
read.triggered.errora triggered read fails after retries are exhausted
Subscribesubscribe.create.error a real-time subscription fails to set up
Writewrite.async.donean asynchronous write completes
Installationinstallation.createdan installation is created
installation.updatedan installation's configuration changes
★ Recommended starter set. If you’re setting up, start with connection.error, read.schedule.paused, and subscribe.create.error. They cover the failures most likely to silently break a customer’s integration. See the full list of notification events →

Build reliable agents with Sampling and Notifications

If you’re leaning on on-call engineering to respond to integration errors, the highest-leverage improvement to your system is better error handling. By setting up both Sampling and Notifications, you get reliable integrations with your customers’ data on the first try, and agents that can debug in real time.

Wired together, an agent can even self-heal: catch a connection.error, prompt the customer to re-authenticate, and clear the error the moment a connection.refreshed event arrives.

A new take on native product integrations