What is context debt?

Where context debt comes from

A Salesforce org accumulates state faster than it accumulates documentation. Objects, fields, flows, validation rules, permission sets and Apex classes are added by different people across years, often under delivery pressure, and the reasoning behind each decision is captured in a ticket comment, a chat thread, or nowhere at all.

The org itself records what exists but not why. Metadata tells you a validation rule fires on Opportunity; it does not tell you which compliance requirement it came from, or that a similar rule was removed two years ago because it broke a partner integration. That missing layer is what people reconstruct by asking colleagues.

Multi-org and multi-repo setups multiply the problem. Production, sandboxes and scratch orgs drift apart, and knowing which environment is authoritative for a given question becomes its own research task.

Signs a team is carrying context debt

Context debt is rarely tracked, but it shows up in recognisable patterns:

• Estimates are dominated by investigation rather than implementation.
• The same questions about the same objects are asked repeatedly, months apart.
• One or two people are a bottleneck because only they know why things are the way they are.
• Onboarding a new administrator or developer takes months before independent delivery.
• Changes cause regressions in areas nobody knew were connected.
• Documentation exists but is stale enough that people no longer trust it.

Why generic AI assistants do not reduce it

A general-purpose AI assistant can explain Apex syntax or draft a flow in the abstract, but it has no access to the org it is being asked about. Every conversation starts from zero, so the human has to supply the context first - pasting metadata, describing conventions, explaining history.

That transfers the work rather than removing it. The person still performs the archaeology; they now also perform it in a chat window. And because the assistant retains nothing between sessions, the same context is re-supplied on the next request.

The cost is not the assistant’s reasoning quality. It is that the context has no persistent home.

How continuous context reduces it

Reducing context debt requires the understanding of the org to live outside people’s heads and stay current as the org changes. That means reading org metadata and repositories directly, connecting them to decisions and documentation, and updating that map continuously rather than at documentation sprints.

RplAI is built around this: it connects to Salesforce orgs and repositories, maintains a persistent map of metadata, code, documentation and past decisions, and gives its specialist agents access to that map on every request. Documentation and changelogs are produced as work happens, so the record does not drift from the org.