# Test an agent against difficult company records

Use case · AI agents · Planned API workflow

Canonical page: https://datasdk.ai/use-cases/test-an-agent-against-difficult-company-records

Plan an evaluation that catches invented facts and unsafe actions.

## Who this helps

An agent builder needs to check behavior beyond complete and obvious examples.

## How to use it

1. Create labeled test cases for missing fields, ambiguous domains, and no matches.
2. Check whether the agent preserves uncertainty and obeys action limits.
3. Fix failures before connecting the planned API to live workflows.

## Illustrative example

A test includes a fake webpage instruction that tries to make the agent reveal a secret.

## What to check

Passing a test set does not guarantee reliability on every future company or webpage.

## How Data SDK fits

The full Data SDK API is coming soon. Platform and AI-agent workflows are planning examples. Confirm access, data delivery, usage, branding, and resale terms before building or selling a feature.

These are suggested workflows, not a list of built-in integrations. Use your own approved analytics, CRM, and research process where needed. Matches and fields vary. A company match does not identify a person or prove buying intent.

## Related reading

- [Prompt injection](https://datasdk.ai/glossary/prompt-injection): Prompt injection is untrusted content that tries to redirect an AI system's instructions.
- [Least privilege](https://datasdk.ai/glossary/least-privilege): Least privilege gives software only the access needed for its task.
- [Evaluation set](https://datasdk.ai/glossary/evaluation-set): An evaluation set is a collection of examples used to check how a system performs.
- [Prepare an AI company research brief](https://datasdk.ai/use-cases/prepare-an-ai-company-research-brief): Plan concise briefs tied to a matched company domain.
- [Classify company fit with evidence](https://datasdk.ai/use-cases/classify-company-fit-with-evidence): Plan a reviewable explanation of how a company meets user-defined criteria.
- [Suggest questions for a discovery call](https://datasdk.ai/use-cases/suggest-questions-for-a-discovery-call): Plan questions grounded in the company's business rather than guesses about the visitor.

[Glossary](https://datasdk.ai/glossary) · [Use cases](https://datasdk.ai/use-cases) · [Comparisons](https://datasdk.ai/vs) · [For businesses](https://datasdk.ai/for-businesses) · [For builders](https://datasdk.ai/for-builders)