Jentic Tech Talk

Arazzo describes deterministic sequences where each step calls an operation, data flows forward, and the workflow runs to completion. That model covers most of what enterprise integration workflows need. What it cannot yet describe is the moment when a workflow must stop, present a situation to an actor, and wait for a response before it can continue.

The patterns that require this are not edge cases. A buy-now-pay-later checkout finds that three of five items cannot be financed, and the customer decides whether to proceed with the eligible items, switch payment methods, or cancel. A deployment pipeline has validated a change and assessed the risk, but an operator must sign off before any infrastructure is touched, potentially hours later. An OAuth authorisation code flow requires a human to navigate to an authorisation server, authenticate and consent, and the result arrives as a query parameter in a redirect callback, not as an API response.

The formal proposal adds a first-class interaction field on a step with three modes:

• - form structured actor response, described with JSON Schema
• - redirect OAuth-style flows, where a browser or user-agent is in the mix
• - acknowledge simple approval gates

Timeout handling is declared as an ISO 8601 duration with an explicit onTimeout branch defining what happens if no response arrives. Once an actor responds, the validated response flows into downstream steps as normal outputs, with no special casing required. There is a natural alignment with MCP’s Multi Round-Trip Request pattern: an Arazzo executor running as an MCP tool host can map interaction steps to InputRequiredResult and resume statelessly across any executor instance.

The Overlay Specification, the OpenAPI Initiative’s standard for modifying API descriptions without touching the source document, released version 1.2 on September 22nd. Two additions address friction that practitioners running Overlays at any real scale will recognise.

Reusable Actions A new components.actions section lets you define an action once and reference it from any number of targets via $ref. Adding a standard error response, injecting a security requirement, or inserting a missing summary across twenty paths now means writing it once and referencing it twenty times. References can override individual fields from the component, so a shared base action can still be tailored per target without duplicating the whole block.

Portable Document Identity Overlay documents now carry their own URI. When $self is set, relative URIs in extends resolve against that identity rather than wherever the file was retrieved from, so the same Overlay works correctly whether pulled from a CI runner, a local filesystem or a registry. A related tightening: extends and $self can no longer contain URI fragments. Anyone using fragment-scoped extends in 1.1 will need to adjust before upgrading; everyone else changes the version field and is done.

Benjamin Granados (Head of Global AI Community at Santander, MIT Faculty Collaborator) gave a talk at APIAddictsDay26 on October 17th that maps the shift from developer experience to agentic experience in terms worth paying attention to. Developer experience is about reducing time to first call. Agentic experience, in his framing, is about how easily an agent can access, understand, and operate within a product to achieve a user-defined goal. That is not a matter of degree. Developers read documentation and retain context across sessions. Agents execute intent, have no tolerance for ambiguity, lose context between chained calls, and break silently when something does not match their expectation.

The data he cites, from the 2025 Postman State of the API Report, makes the gap concrete: 24% of teams currently design APIs with AI agents in mind; 60% still design primarily for human developers. The five shifts he identifies that agentic consumption demands (discovery, deterministic workflows, agent identity and security, observability, governance) map directly onto the dimensions the Jentic AI Readiness Framework measures.

He names the Jentic scorecard and Arazzo explicitly as the tools for addressing the workflow layer, with a worked Arazzo example in the slides.