The consumers you know about are not the list
Changing an endpoint is a decision made against a mental list of consumers, and the mental list is always short. The internal service everyone remembers, the mobile client, perhaps a partner. The integration built two years ago by a team that has since dissolved is not on it.
The real list sits in two places: the specification history, which says what the contract was, and the access logs, which say who used it and how recently.
Breakage is not uniform either. A removed optional field breaks the consumer that reads it and nobody else. Knowing which is which turns a broad announcement into three specific conversations.
Producing the list before the change is cheap. Producing it afterward is an incident.
How the ActionFlow runs on the canvas
Specification files are read by a GitHub node at both versions. An Extract Data node reduces each to its surface: endpoints, parameters, required flags, and response fields.
A Generate Object node compares them and returns the change set as typed differences (removed, renamed, newly required, type changed) rather than a textual diff.
An HTTP Request node pulls consumer identity and call patterns from your gateway or analytics for the period you set. A Uniq node reduces them to distinct consumers.
A Loop node walks those consumers. For each one, a Generate Object node returns which of the typed differences apply to what that consumer actually sends and reads.
Comparing against observed calls rather than against the specification is the decision that makes the output true. Most consumers use a fraction of an endpoint.
A Filter node drops consumers nothing touches, which is usually most of them. A Sort node orders the rest by severity.
A Create Excel node produces the impact table. A Linear node opens the migration task.
An Email node in the Human in the Loop group holds the consumer notice for your approval. Nothing goes to an external developer without a person releasing it.
Nodes this ActionFlow uses
- GitHub (Integration): specification files at both versions.
- Extract Data (Data): reduces each version to its surface.
- Generate Object (AI Core): typed differences, not a textual diff.
- HTTP Request (Util): consumer identity and call patterns from the gateway.
- Uniq (Control): distinct consumers out of raw call records.
- Loop (Control) with Generate Object (AI Core): impact per consumer, against observed calls.
- Create Excel (Data): the impact table, ordered by severity.
- Email (Human in the Loop): external notices wait for a person.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.