Skip to main content
In this lesson, you will ask your agent to find recently funded AI companies and founder contacts. You will learn to inspect the request, save its run ID, and distinguish a completed run from a useful result.

​
Before you start

Use a coding agent that can run shell commands. You need permission to install Deepline and authorize a workspace. The research request uses provider-backed work and can incur usage. Check access and price before starting it.
For an agent without a shell, complete MCP setup first. For Claude Cowork, use the Cowork setup guide.

​
1. Install and connect

Paste this instruction into your coding agent:
Complete the browser authorization that setup requests. Then ask your agent to check its effective connection:
The response has connected: true. Its workspace identifies the workspace you intend to use. Installation alone does not confirm authorization.
If this check fails, stop here. Follow the diagnostic or retry instructions from setup. Coding-agent setup covers installation recovery.

​
2. Inspect the starter

Ask your agent to inspect and check this maintained Play:
Notice target_count and days. Their defaults are three requested results and a 30-day window. The Play finds candidates, checks funding and AI evidence, and looks for current founders with verified work emails. Ask your agent to explain required provider access and applicable pricing. These inspection commands do not start the research.
You can identify the requested count and time window. The Play check passes, and you understand the access and usage required to run it.

​
3. Start one request

Paste the primary starter prompt into your agent:
Ask it to use the inspected starter. Have it retain the returned run ID and show the result from that run. The prompt is the same one offered after setup. Do not submit it again while the first run is still processing. The run ID identifies this execution. The Play name identifies reusable logic. Keep both, so you can inspect existing work without starting another run.

​
4. Review the recorded result

Ask your agent to inspect the saved run. In a terminal, replace <run-id> with the actual ID it returned:
If the status is still running or queued, inspect that same ID later. A failed or cancelled run needs diagnosis. Do not treat it as an empty result. For a completed run, ask your agent to show the result dataset and check:
  1. The company matches the AI evidence.
  2. The funding announcement falls within the returned window.
  3. The contact is a current founder of that company.
  4. The work email has the returned verification evidence.
The Play returns count, requested_count, and shortfall_count. It can return fewer than three accepted rows. Preserve that shortfall instead of adding unsupported companies or contacts.
You can trace the displayed rows to one saved run. Each accepted row has the required evidence. Missing matches and failures remain visible.
You completed the first workflow: authorize a workspace, inspect a maintained Play, start it once, and review its evidence.

​
Check your understanding

A company has a verified founder email, but its funding announcement is 45 days old. Does it satisfy this 30-day request?
No. A verified email does not establish recent funding. Check the funding evidence against the run’s time window. Keep the row outside the accepted set.
Later, explain the difference between a Play name and a run ID without rereading this page. Ask your agent to correct any gap in that explanation.

​
Continue

Use these after you finish the first lesson. Each prompt starts new work and can incur usage. The menu is generated from the same source as setup.