Integration is where Salesforce architecture choices show up first. The wrong pattern works fine for 100 records and dies at 100,000. This page is the practitioner reference for choosing the right pattern up front, with the failure modes that help you predict where the next outage will come from.
04 steps · 04 FAQs
“Every Salesforce integration outage I've seen was an architecture choice from 18 months earlier finally hitting scale.”
Practical steps
Source of truth, direction, sync pattern, frequency, error path. Write it down before you build. Contract changes mid-build are the cost surprise.
Every callout has a logging hook. Every event has a counter. Every error path has an alert. Skipping observability is borrowing time you'll repay with interest.
Run 10× expected volume through it. The integration that works at design volume is not the integration you're shipping.
When (not if) the integration breaks, the on-call needs the recovery steps in 30 seconds. Document them. Test them. Update them.
From the library
Score
The 4 permission layers and how they really resolve, 10 named anti-patterns, the SOQL query bank, real-world breach stories, and a cleanup sequence that won't break access. Spring '26 current.
Dev workflow
How working Salesforce developers are using AI to ship Apex, LWC, and integrations 3–5x faster, without giving up code review discipline or production safety.
Score
100 copy-paste AI prompts across 10 admin categories: user management, formulas, flows, debugging, reports, LWC, SOQL, Apex, data cleanup, and documentation. Battle-tested on real Salesforce orgs.
Frequently asked
Platform Events for fire-and-forget notifications where the receiver doesn't need to ack. Callouts for synchronous request/response with the external system as source of truth. Mixing them produces architectures that are hard to reason about and hard to monitor.
Bulk API for batches over 200 records, REST for individual writes, Platform Events for change-data-capture broadcast. The mistake is using REST for everything because it's familiar, your governor limits will let you, until they don't.
Three layers: (1) retry with exponential backoff for transient errors, (2) dead-letter queue for retries-exhausted records, (3) alerting on DLQ depth so you know before customers do. Skip any layer and silent data loss is your default mode.
Async by default. Sync only when the user is waiting on the result. Sync integrations couple your Salesforce uptime to the external system's uptime, and the external system always has lower SLA than you do.
More for developers
Hand the actual work to Clientell AI your AI agent for Salesforce integration patterns, flows, data ops, and user management.
Unlimited messages · No credit card required