The problem
Before a first sales call, a rep needs to know what the prospect does, whether they need the product, and what to say. That research is spread across many tabs, and the notes rarely say where each claim came from. Parallel sells web research APIs for AI agents, so I built the prep tool on its own Task API.
What I did
- Wrote the brief as a Task API output schema: 13 fields, with fixed choices for the agent signal and the best-fit API so briefs compare cleanly.
- Passed in what Parallel sells, its six APIs with prices, checked against parallel.ai on October 22, so the research agent can match a prospect's need to a specific API.
- Showed each field with numbered sources and the API's confidence rating, sharing one source list across the brief.
- Explained failures in plain language: a rejected key, an account out of credits, or a request Parallel didn't accept.
- Added a dry run that prints the exact request without calling the API, so it can be checked for free.
- Wrote six offline tests against an invented company, so they cost nothing and need no key.
Results
It hasn't run against the live API yet; that needs a Parallel account and key. The sample below is invented to show the format, and says so.
What I'd do differently
Get an API key on day one. I built and tested everything offline, so the first live run is still ahead, and that run is the real test of the schema.
Artifacts
Source on GitHub ↗ · Sample brief ↗

