Ecommerce Task Automation: Tasks, Workflows, and Paz Routines

A task states the result needed: identify missing capacity details in priority bottle descriptions. A workflow connects the steps: select products, compare source facts, identify gaps, and produce a correction. An AI agent can choose intermediate steps or tools based on the evidence it encounters.

A deterministic workflow follows predefined rules. An agent can interpret less predictable inputs, but still needs a defined task and permitted actions. Neither a saved prompt nor a successful tool call proves that the business goal is met. A consumer shopping agent acts for a shopper; the commerce task here serves the merchant business.

In Paz, Routines save what to check, the relevant scope, and when to run. A generic task describes the business job; a Routine is the saved recurring capability used to check it. Some Routines run on a chosen schedule, while system Routines follow existing syncs or events.

A supported finding can lead to a shared follow-up with evidence, progress, and next checks. This is not the same as a Routine run. Overview brings these investigations together with your personal briefing and connected signals. Inbox holds questions and decisions that need attention; marking a briefing read does not resolve the shared investigation.

• Goal: the question to answer or product-data condition to improve.
• Trigger and scope: the supported schedule or event, products, market, and channel.
• Inputs: the authorized sources and business context needed for a reliable result.
• Action: a finding, proposed correction, or supported destination update.
• Completion: the evidence that the output meets the goal.
• Exceptions: what happens when facts conflict or a source is unavailable.

A weekly catalog check can return no missing facts, detect a gap, or fail because a source is unavailable. Those are different results. The ecommerce automation guide provides a reusable specification for all three.

A custom Paz Routine checks a selected collection each week for descriptions that omit verified material or capacity. Its finding identifies the affected product and source fact. A supported optimizer can prepare a correction for review; saving the Routine does not authorize publication. An already complete description needs no action; conflicting source values produce an exception.

Business context explains why a focus collection takes priority over a phase-out range with similar gaps. If the workflow applies a correction, action permissions and approval policy define the destination and fields it may change.