# How Mandy Hornaday built a Claude Code agent that scores podcast guests on credibility and audience

Mandy Hornaday hosts Growth Activated, a B2B marketing leadership podcast. She wanted to stop finding guests by gut feel. Over a weekend she treated as a hackathon, she used Claude Code to build a guest research agent. The agent scores candidates on two things: B2B marketing credibility, and whether they own an audience that could promote the episode. It researched and scored 10, then 50, then 100 potential guests in about 90 minutes from setup to output. She skipped automated outreach to avoid API costs and drafts emails in Claude chat instead. She also built a post-production workflow for episode SEO and promotional content, which she says she had not tested yet.

| | |
|---|---|
| **Who** | Mandy Hornaday, host of Growth Activated, a B2B marketing leadership podcast ([growthactivated.com](https://www.growthactivated.com)). LinkedIn: [linkedin.com/in/mandyhornaday](https://www.linkedin.com/in/mandyhornaday) (verified 2026-10-19: author of the post, and linked from growthactivated.com) |
| **Problem** | Finding the right podcast guests. Her manual process was "slow, inconsistent, and I knew I was missing people" |
| **Stack** | Claude Code, Visual Studio Code, Claude Pro (upgraded to Claude Max), web research, Google Drive, Claude chat for outreach drafts. For the untested post-production workflow: Claude's research function |
| **Confidence** | Medium (first-person, two of her own sources; no outreach, booking or promotion results stated) |
| **Last verified** | 2026-10-19 |

## The problem
She says the most time-consuming part of running the podcast "isn't recording - it's finding the right guests." Her old process was to "scroll LinkedIn, check a few conferences, Google around, ask for referrals," then build a shortlist "based on gut feel and whatever I happened to come across." She says it worked, but it was "slow, inconsistent, and I knew I was missing people."

The trigger was an intro call with Liza Adams, a guest she was booking. Liza Adams showed her a coordinated team of AI marketing agents. Around the same time, Claude had a release at the end of January that people on LinkedIn were discussing. She decided to learn to build agents herself and "treated my weekend like a hackathon."

## The blueprint (as she describes it)
1. **Pick a task where you already know what good looks like.** She chose guest research over her post-production workflow because it was "something a little easier." She had personally researched dozens of these guests before. That meant she could tell right away whether the output was on target.
2. **Set up Claude Code with a chat window.** She used Visual Studio Code, which she says lets non-developers use Claude Code "through a typical chat window." She connected her Claude subscription. She says you need Claude Pro, "$20 a month," and that she upgraded to Claude Max "within 24 hours" at what she thinks was "about $100 a month."
3. **Describe the criteria in plain language.** She told it what she looks for in a guest:
   - B2B marketing leadership credibility (awards, speaking events, community presence).
   - A strong point of view: "not just a title, but actual thought leadership."
   - An audience they could promote the episode to (newsletter, podcast, LinkedIn following, their own community).

   She also told it where she usually validates speakers: "look at B2B marketing awards, look at different communities, see if they host their own podcast." She wanted it to look at credibility markers and also to prioritize guests who have an audience of their own.
4. **Ask for platform risks up front.** She asked it to flag security or platform risks. It warned that LinkedIn has "really strict rules around automation and scraping" and that she could put her own profile at risk. She chose "no LinkedIn data for now." It mentioned that a Sales Navigator license gives more flexibility, but she did not want to risk it while testing.
5. **Let it build the scoring model and run the research.** It built "this really intricate scoring model," relied on web research, scored everyone, and put the results in a Google Drive folder. Fields include:
   - name, title, company and LinkedIn URL
   - podcast appearances, speaking events, community affiliations and awards
   - whether they own an audience, and how big
   - research source, notable topics and custom details
6. **Scale in steps.** She asked for 10 first, then 50, then 100. She says she can "always go back and tell it to add more."
7. **Keep outreach manual and free.** She wanted it to write personalized outreach emails automatically. Claude Code told her that would need API calls, which she describes as costing "essentially a fraction of a cent per email." She told it to "do whatever you have to do to make it free." Now she prompts it when she is ready to personalize emails for a specific set of guests. She uses Claude chat "to draft outreach in real time based on the info it pulled."
8. **No schedule yet.** The workflow does not auto-run. There are "No automated refreshes, no rescoring on a schedule." She calls that the eventual goal.

**Secondary: the post-production promotion workflow (untested).** She says this one "took significantly more time to build" and that she had not tested it yet. She planned to test it on the same episode.
- **Scope:** "SEO research, episode title creation, episode description, chapter markers, transcript cleanup, and then all of the promotional content. LinkedIn posts, blog posts, all of it."
- **Research before building:** She used Claude's research function to run deep research first, before building the agent. The topics were podcast optimization (what Apple and Spotify prioritize for discoverability), blog best practices for turning podcast content into reusable blogs, and LinkedIn posts made from podcast content.
- **Load the research into the workflow:** She uploaded all of that research into the workflow. Her research told her that Apple does not look at episode descriptions and that Spotify looks at everything, including the description. These are findings from her own research, not verified facts.
- **One agent, for now:** It is currently "one agent doing all the tasks." She wants to try running it through "a team of specialized agents" next, such as an SEO agent, a content creation agent and a social media agent.

## Stated results
- Guest research workflow: "about 90 minutes" in the post, and "about an hour and a half" from setup to finished output in the episode.
- "Zero to 100 scored, researched potential guests." She says this will keep her going "for the next couple of months."
- People already on her wish list showed up, "alongside names I'd never come across who were equally qualified." In a comment she adds that "it surfaced names I had yet to find through my usual process."
- Outreach replies, bookings or recorded episodes from the list: None stated.
- Post-production promotion workflow: not tested at the time of either source. None stated.

## Caveats
- Both sources are hers: a LinkedIn post that promotes her episode, and the episode transcript on her own site. The episode points listeners to her CMO Operating System waitlist landing page on growthactivated.com, which is a product of hers.
- The post lists LinkedIn profiles among the outputs, and the field list includes LinkedIn URL. She also says she chose "no LinkedIn data for now." The sources do not explain how these fit together.
- The Apple and Spotify platform claims come from her own deep research. She does not cite a primary source.
- The pricing figures are hers. She is unsure about the API cost question: "Do I actually need to pay? No idea."
- The 90 minutes covers building and running the workflow. She does not give before-and-after time per guest.
- She also built a Webflow developer agent. She says it took longer than building the page manually.
- Handles: LinkedIn linkedin.com/in/mandyhornaday (verified 2026-10-19: author of the post, and linked from growthactivated.com). X, Threads, Bluesky: not checked.

## Sources
- https://www.linkedin.com/posts/mandyhornaday_the-first-ai-workflow-i-built-took-about-activity-7435330832631877633-y0Qu (primary: her first-person LinkedIn post, 2026-03-05, plus her comments)
- https://www.growthactivated.com/blog/ai-agents-marketing-leaders-weekend-hackathon (primary: transcript of her solo Growth Activated episode, dated February 11, 2026)

---
*Want agents like these working your distribution? [See the Agent Stack Teardown](/agent-stack-teardown). Get new teardowns weekly: [subscribe](/subscribe) (double opt-in).*