What Agent Playground Shows Your Team

Custom Checks let a firm build its own review logic in plain English. Agent Playground is the part of Structured AI that lets a team watch that logic actually run, in real time, against their own drawings.

The Problem It Solves

Building a custom check is one thing. Trusting it is another. If a QA lead writes a check meant to catch missing fire dampers on a mechanical set, the natural next question is: how do I know it's actually looking in the right places?

Without visibility into how an agent works, a custom check is a black box that either returns results you have to trust blind or doesn't return results you can debug. Neither is good enough for something a firm plans to run on every project.

What Agent Playground Actually Shows

Agent Playground displays the check as it runs: what it's reading, what it's searching for, and where it's looking on the drawing set. Instead of writing a check and waiting for a final report, a team can watch it work through a set page by page, the same way they'd watch a junior engineer walk through a review out loud.

That visibility does two things. First, it builds trust in the check before it's rolled out firm-wide, because the team has actually seen how it reasons through the drawings, not just what it concludes. Second, it makes debugging fast. If a check is missing something it should catch, watching it run usually shows exactly where the logic breaks down, which is a lot faster than guessing from a wrong result alone.

Where This Fits Alongside Prompt Lab

Agent Playground pairs naturally with Prompt Lab, which lets a team test two different approaches to the same check side by side on their own drawings to see which performs better. Prompt Lab answers "which version works better." Agent Playground answers "why."

Together, they turn custom check-building from a one-shot guess into something a team can iterate on with actual evidence, rather than trial and error against a live project.

Why This Matters for Firm-Specific Standards

Every firm has standards that go beyond the code. A structural firm might have its own connection detail conventions. An MEP firm might have specific clearance requirements that exceed code minimums. These aren't things a generic plan-check tool would ever catch, because they're not written into any code book.

Custom Checks exist to encode that firm-specific knowledge into something that runs automatically on every project. Agent Playground is what makes it realistic to trust that encoding, since the team can see it working correctly before it's relied on.

FAQ

Do I need to know how to code to use Agent Playground? No. Custom Checks are built in plain English. Agent Playground shows how the resulting agent behaves, it doesn't require writing code to use or watch.

Can Agent Playground be used to debug a check that's already live? Yes. If a check that's already in use starts missing things or flagging too much noise, running it in Agent Playground against the drawings in question is a direct way to see what's going wrong.

Is Agent Playground only for QA/QC checks? It's used anywhere a custom agent is running, which includes QA/QC checks but also other agent-built workflows a firm sets up in Structured AI.