Arazzo.
The open standard that teaches APIs to work in sequences, not just endpoints, and why payroll, scheduling, and timekeeping are about to run on it.
OpenAPI describes doors.
Arazzo describes the route through the building.
OpenAPI tells you what each endpoint does. It says nothing about which endpoint to call first, what to do with the token it returns, or how to know the whole job succeeded. The Arazzo Specification — an open standard under the OpenAPI Initiative — fills that gap: a machine-readable map for chaining endpoints into an actual business outcome.
An Arazzo document spells out ordered steps, the dependencies between them, runtime expressions that carry data from one response into the next request, success criteria evaluated against real responses, and branching logic for when things go off-script. “Book the flight.” “Onboard the employee.” “Reconcile the timecard.” The outcome is the unit of description — not the verb.
The spec reached v1.0 in 2024 and has moved quickly since. Version 1.1.0, released in June 2026, added AsyncAPI support — so one workflow can call a REST endpoint, publish to an event bus, wait for an asynchronous acknowledgment, and feed the result into the next step — plus chained sub-workflows, a Selector Object for JSONPath/XPath extraction, and a formal grammar for runtime expressions. On the roadmap: gRPC, GraphQL, MCP and A2A step types, loops, and actor-in-loop execution.
HR systems don't have endpoints.
They have errands.
Nobody integrates with a payroll platform to call one endpoint. They integrate to get something done: pay people correctly, keep shifts covered, prove a punch happened. Every one of those jobs is a chain of calls across several APIs — and until now, that chain lived in someone's head, a wiki page, or a tangle of integration code.
Take the UKG Pro suite — HCM APIs for employee data, payroll, benefits, and personnel actions; WFM APIs for time, scheduling, accruals, and attendance — serving more than 80,000 organizations. Or the classic cross-system mess: pull commissions from Salesforce, match them to employee records in BambooHR, submit tax data to the payroll platform. Arazzo turns these errands into documents: versioned, testable, and executable by humans and machines alike.
Payroll is a chain, not a call
A pay-statement review means validating the employee, reading the current rate, pulling statement history, and reconciling direct deposits. Four calls, one answer. Miss a step and someone’s paycheck is wrong.
Scheduling decisions need three systems at once
Coverage checks weigh the location schedule against one employee’s shifts and their accrual balances. PTO approval branches on whether the balance covers the request, then surfaces conflicting shifts.
Timekeeping must prove itself
A clock punch isn’t done when it’s submitted — it’s done when you read the day’s punches back and find it there. Verification is part of the workflow, and Arazzo makes that explicit.
Employee data lives in two truths
The HCM is the system of record; the WFM roster is where the work happens. Reconciling the same person across both — before payroll runs on mismatched data — is a workflow worth writing down once.
Twelve Arazzo workflows for UKG Pro,
written to be read — and run.
We authored twelve Arazzo documents against the UKG Pro HCM and WFM APIs — covering timekeeping, payroll, scheduling, leave, and employee data. Every step spells out its request inline, so the flow can be read and executed without opening the underlying OpenAPI description. Pick one and watch how it moves.
An ecosystem that already writes,
tests, and runs these documents.
A spec is only as good as its tools. Arazzo's bench has filled in fast — only a handful of engines actually execute workflows end-to-end today, and knowing which is which is most of the battle. Here is the landscape as it stands.
Run & test
The engines that execute a workflow against a real server and return a verdict.
Author & visualize
Write workflows by hand or by drag-and-drop, and see them as diagrams.
Lint, validate & convert
Keep documents honest, and move between formats.
The reader Arazzo was always waiting for
turned out to be a machine.
Left alone, an AI agent asked to “approve this timecard” will guess which endpoints to call, in what order, and what to do with each response. It will burn tokens figuring that out, and some percentage of the time it will guess wrong — with someone’s pay attached.
Arazzo gives agents a deterministic track to run on. The sequence, the data mapping, the branch conditions, and the success criteria are declared up front, so the model’s reasoning is reserved for judgment calls — not for reconstructing plumbing. SmartBear’s framing is the one we use with clients: expose value, not verbs. Hand the agent the workflow-level capability, not a bag of raw endpoints.
The economics are real, too. When the model already knows what each step is, when to use it, and in what order, it stops generating tokens to figure that out — which cuts latency and LLM cost on every run. Usage- and SLA-based metadata can be enforced at the processing layer, keeping agent consumption inside budget.
And the twelve UKG workflows above? They are exactly the shape an HR agent needs: “check PTO and surface the conflicts” as one deterministic tool, instead of four endpoints and a prayer.
The document is the tool
MCP servers can expose an Arazzo-defined workflow as a single tool. The agent calls “pto-balance-and-schedule”; the server executes four API calls deterministically underneath. Less reasoning loop, fewer hallucinated calls.
The runners are agent-native
Jentic’s Arazzo Runner was built for autonomous execution — auth, dynamic servers, runtime expressions, branching — with a CLI an agent can drive. arazzo-cli ships an MCP server outright; HAPI MCP turns Arazzo workflows into MCP tools without hand-writing server code.
Code generation grows up
Feed an AI coding assistant an OpenAPI spec plus its Arazzo companion and it generates context-aware client code — token from step one into the header of step two, retries where the spec says so. Microsoft’s Kiota is building Arazzo workflow support for C#/.NET as a reference implementation, generating higher-level workflow SDKs straight from the documents.
Trust is testable
Agents acting on payroll systems need guardrails, not vibes. Respect and Specmatic execute the same documents against real servers and return pass/fail verdicts — the workflow the agent runs is the workflow you tested.