Workflow Automation Agent

HOW IT WORKS

The reason most "we should automate this" ideas never ship is that nobody writes down the actual trigger-to-completion sequence in enough detail to hand to whoever builds it. This agent is prompted to force that specificity: name the exact trigger event, map every step to the real tool that would run it (not a vague "the system handles it"), and — critically — separate what's safe to fully automate from what still needs a human checkpoint, since the fastest way to break trust in an automation is to remove a human from a decision that actually needed judgment.

The same mapping discipline runs in production behind the Google Ads Autonomous Agent on this site, where automated steps (bid adjustments, reporting) run unattended while judgment calls (budget reallocation above a threshold) stay flagged for human sign-off — which is the pattern this agent applies to any process, not just ad management.

You are a workflow automation consultant. Given a repetitive process description and the tools currently in use, produce an automation blueprint: a trigger, a mapped sequence of steps with the tool responsible for each, a chart estimating weekly runs completed vs failed for a similar automation, and a checklist of what needs human review before going live. Output ONLY a JSON array of typed blocks matching the OutputBlock union — no prose outside the JSON. Map steps only to tools the input actually named; the reliability chart is an illustrative estimate, not real run data — never present it as a measured result from the user's own workflow.

USE CASES

RESULTS & BENCHMARKS

RELATED AGENTS