Customizing Digital Workers: Instructions, Skills, Tools, and Triggers

Digital worker templates from the Marketplace are a strong starting point, but most organizations need to customize them to match their specific processes, terminology, and systems. Agentic Studio provides several layers of customization, from simple instruction changes to adding new capabilities and integrations.

Templates are typically around 60% applicable out of the box. The customization layer is where you close the remaining gap.

Changing Instructions

Instructions are the core policy of a digital worker. They define what the worker does, how it reasons, what format it returns, and how it handles edge cases. Because instructions live in the prompt, even small wording changes can reliably shift the worker's downstream behavior without requiring any changes to skills or tools.

For example, if a digital worker currently produces a free-form summary, you can update its instructions to require a structured JSON output with specific required fields, null-handling rules, and targeted follow-up questions for missing data. The same inbound event will now produce a machine-readable result instead of a narrative response.

Start with instruction changes before adding new skills or tools. Often, adjusting how the worker reasons or formats its output is all that is needed.

Adding Skills

Skills are modular capabilities that a digital worker invokes when it detects the right intent. Rather than embedding every behavior in the instructions, you can keep instructions lean and extend the worker's capabilities through targeted skills.

When defining a skill, the most important configuration is the invocation parameters, specifically the prompts or intent patterns that should activate the skill. Well-defined invocation criteria ensure the skill is called reliably when needed, and not called when it should not be.

Example skills and their typical triggers:

• Get Work Order Details: triggers on requests such as "what is the status of WO-12345?"

• Search Knowledge Base: triggers on questions such as "how do I reset the controller?"

• Create Dispatch Recommendation: triggers on requests such as "who should we send?"

If the worker cannot find the required information to call a skill, it should ask targeted follow-up questions rather than proceeding with incomplete data.

Adding Tools

Tools give a digital worker real execution power, meaning the ability to take actions beyond reasoning and text generation. In Agentic Studio, tools can come from three sources:

• Connector tools: API operations imported from a connected system such as IFS, ServiceNow, or a third-party application.

• Out-of-the-box tools: pre-built capabilities provided by IFS Loops.

• Custom tools: tools you build to match specific organizational processes.

Depending on the connector, tools may be exposed as traditional APIs or as MCP-based capabilities, allowing the digital worker to reliably discover and invoke the right function at the right time.

Adding Triggers

Triggers start a digital worker automatically in response to an external event from a connected system. The event payload carries key context, including IDs, timestamps, and changed fields, which the worker uses to determine its next step.

Example triggers:

• Service Work Order Created (IFS): sends the work order ID and key fields. The digital worker auto-classifies priority, flags missing information, and recommends next steps.

• Incident Status Changed (ServiceNow): on escalation, sends incident context. The digital worker summarizes impact, retrieves relevant knowledge base articles, and drafts the escalation update.

Configuring Human-in-the-Loop (HITL)

• HITL adds an approval or decision step to a digital worker's configuration so that sensitive, ambiguous, or high-impact actions are reviewed by a person before the execution continues.

• HITL can be routed to Microsoft Teams, Slack, or other collaboration channels. Configuration adapts to the chosen channel. In Slack, for example, you select the target workspace and channel where requests should be posted.

• HITL can also be made conditional, triggering only when specific rules are met, such as low confidence, missing required fields, an exception condition, or a high-value transaction, while keeping the rest of the process fully automated. This allows different humans or teams to handle different decision points within the same configuration.

Structured Output

Structured output ensures a digital worker returns results in a predictable schema, for example a named JSON object, so that downstream systems, dashboards, or integrations can reliably parse and use the response.

To configure structured output, define an explicit output contract that maps each required field to its source in the conversation or tool results. Specify required versus optional fields, allowed values, and what to do when data is missing (for example, set the field to null and ask a targeted follow-up question). Adding this contract to the digital worker's instructions makes outputs consistent and integration-ready.

Concurrency

Concurrency controls how many parallel tasks a digital worker can run simultaneously. This matters particularly for high-volume scenarios where instructions like "for each entity, do this" can trigger a fan-out into many parallel executions.

Unbounded fan-out can significantly increase token usage and drive spikes in system resource consumption. Setting a concurrency limit caps parallel fan-outs, paces execution, and keeps the process predictable while still benefiting from parallelism where it matters.