Urgent.News

What's breaking now, across thousands of outlets.

Tech

Lawful Agents, Part 1: Your agent is reading an API that was never written for it

If you have ever pointed an agent at an enterprise API, you will know how it goes. The response comes back five levels deep, the model spends half its context working out where the manager's name lives, it retries when a field is not where it expected, and eventually it fills the gap with something it made up. Then you spend the afternoon wondering whether the problem is your prompt. It usually…

This article series focuses on creating trustworthy agents that can interact with enterprise systems, specifically using Workday's Agent-Ready MCP tools. These tools differ from traditional enterprise APIs in several key ways. Firstly, their schemas are flat, meaning there is less structure to navigate, resulting in fewer tokens spent parsing, faster responses, and less room for the model to make up information.

Secondly, every tool describes itself, carrying metadata about its function and expected parameters. This allows the agent to receive error information that helps it fix values on the next attempt. Lastly, these tools follow a standard called MCP JSON-RPC 2.0, which is supported by a single endpoint for each tenant, simplifying the connection process for the agent.

For developers, the interesting aspect is that they can utilize these tools regardless of the underlying platform. Agents built with platforms like Claude, Cursor, Copilot Studio, or Google Agentspace can call the same tools as Workday's Custom Agents. However, registering a third-party agent in Workday's Agent System of Record (ASOR) is necessary, and the registration process involves providing a list of allowed tools in the agent definition.

If a write tool depends on other prompt tools, all of them must be included in the agent definition for the write operation to complete successfully.

The MCP Explorer on developer.workday.com is a helpful resource to find the necessary tools, filter by type, and retrieve the required tool IDs. Alternatively, the ASOR Agent Resource Search API allows for programmatic access to this catalogue using an API call.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in Tech

Guardrail layers and latency

Respuesta corta Los guardrails de IA se implementan en capas superpuestas —preprocesamiento, ejecución y postprocesamiento— cuya integración afecta directamente la latencia final del sistema.

  • Three-layered AI guardrails: preprocessing, execution, postprocessing
  • Execution layer adds 15-120 ms per stage, 78% of total latency
  • Optimization techniques: shared memory, compile-time regex, early checks

More from Wednesday 21 October →