Medialyst Agent

Our most capable column type for personalization. Unlike a simple prompt, agents search the web, read articles, and perform multi-step research before generating output.

Medialyst Agent vs Use AI

Capability	Medialyst Agent	Use AI
Web search	✅ Can search Google News	❌ No web access
Read articles	✅ Can fetch and read full articles	❌ Only sees data in your table
Multi-step reasoning	✅ Plans and executes research steps	❌ Single prompt, single response
Tool use	✅ Multiple tools available	❌ Text generation only
Best for	Deep research, hand-matched angles, competitive analysis	Simple transformations, summaries, formatting
Credit cost	Higher (doing more work)	Lower

When to use which

Use AI when you already have all the data in your table and just need to transform or summarize it.

Medialyst Agent when you need to find information - research a journalist's recent work, check their publication's focus, or find connections to your story that aren't in your table yet.

How to add an agent

Open any workflow table
Click Add Column → Medialyst Agent
Write your research prompt
Run on 5-10 rows first to test
Refine the prompt, then run on your shortlist

Writing effective prompts

Use this structure for reliable outputs:

Task: Describe what you want researched

Inputs:
- Journalist: {{Name}}
- Publication: {{Publication}}
- Recent Articles: {{Recent Articles}}
- Story: {{Story}}

Output Format:
- Summary of findings
- Key insights
- Recommended approach

Constraints:
- Be specific
- Reference evidence from recent coverage
- Avoid generic statements

Example: Coverage style classifier

Research this journalist's recent work and classify their writing style.

Inputs:
- Journalist: {{Name}}
- Publication: {{Publication}}

Output Format:
- Primary style (investigative / feature / breaking news / opinion)
- Evidence (2 specific article examples)
- Best pitch approach for this style

The agent searches for the journalist's recent articles, reads them, and provides a classification with evidence. Evidence-backed, not guesswork.

Chaining agents

For complex research, split the work across multiple agent columns:

Recent Focus agent - summarizes what the journalist is currently covering
Story Gap agent - identifies angles they haven't covered yet
Pitch Opener agent - writes a personalized first sentence using both outputs

Smaller, focused agents are more reliable than one mega-prompt trying to do everything.

Save prompts to library

When you've written an agent prompt that works, save it for reuse:

Open the prompt editor
Click Save to Library
Give it a descriptive name (e.g., "Coverage Style Classifier")
The prompt is now in your library

Next time you create a Medialyst Agent column, insert saved prompts instead of writing from scratch. Your team can access shared library prompts too.