Case study · Agentic systems
I built an order workflow across email, FTP, PDFs, and existing systems.
An AI agent kept choosing allowed actions until the work finished or stopped. Code checked the order. Operators approved exceptions and actions that could affect fulfillment.
My role: I assessed the workflow, made the product and system decisions, built the integration, and handled the release.
System flow
Controlled order workflow
Email, PDFs, and structured files → allowed agent action → code checks → operator approval when required → fulfillment systems. Monitoring shows each run. Evaluation cases can be rerun.
Project details
How I built the system.
- The problem
Orders came in as structured files, email, and PDFs. They still had to pass through the operational, warehouse, transportation, and print systems already in place.
- What I did
Before I wrote anything, I mapped the rules and each system’s job. That showed what the agent could do and where an operator had to approve the work.
- What I built
- The workflow accepted structured files, email, and PDFs. The agent read the unstructured parts.
- It kept choosing allowed actions until the work finished or stopped. Code handled validation and business rules.
- Monitoring showed each run. I could rerun the same evaluation cases to check expected behavior.
- The workflow paused until an operator approved an exception or an action that could affect fulfillment.
- During release, the old and new paths ran side by side. Validated orders still triggered work in the existing fulfillment systems.
- What happened next
After the assessment, the client hired me for a separate project to build the integration.