I was in a coaching session with Ilias, walking him through how to build a YouTube monitoring agent for his investment research.
The setup was straightforward: watch 20 AI-focused YouTube channels, get a summary of new videos delivered to Slack once a week. Instead of spending hours keeping up with what’s happening in the AI space, the agent does it automatically.
We were deep into the mechanics — which channels to track, how to structure the summaries, how to filter for signal versus noise — when I mentioned something I’d added to my own version of this workflow.
One extra prompt. One extra output. Almost zero extra cost.
The Second Prompt
Here’s what I told the agent, in addition to summarizing the video:
“If there’s a sponsor mentioned in this video, tell me who it is and log the company name to Google Sheets.”
Now, every week, alongside my AI research summaries, I get a running list of companies that are actively spending money to sponsor AI content.
That list goes straight to my podcast agency. I forward it with a simple note: I want to reach out to these.
The Productivity Show is always looking for the right sponsors. These companies have already demonstrated they’re willing to pay for AI-adjacent audiences. Finding them used to require manual research. Now the agent surfaces them as a byproduct of the research workflow it was already running.
The Principle: Secondary Intelligence
Most people design AI workflows around a single output.
The agent does the thing you built it to do, and that’s the end of the story. Which is fine. The primary job is worth doing.
But once an agent is reading content — processing a video transcript, scanning an email thread, reviewing a meeting recording — extracting a second signal from that same content is nearly free. The agent is already there. You’re just asking it to notice something else while it’s working.
A few examples of how this applies:
A meeting summary agent that already captures action items can also flag recurring objections your team hears from prospects.
An email monitoring agent that already drafts responses can also log unusual patterns — clients who suddenly stop responding, pricing questions that come up more often, complaints that appear in multiple threads.
A competitor monitoring workflow that already summarizes what competitors are publishing can also extract the tools, integrations, and partnerships they mention.
In each case, the second output costs almost nothing extra. The agent is already processing the input. You’re adding one more question to the same pass.
Why This Matters for How You Design Workflows
When you’re building an AI agent, the natural instinct is to define the job and ship it. The agent does X. Done.
The better habit is to pause before you finalize the design and ask: what other intelligence is already sitting in this content?
With YouTube videos, the answer was sponsor data. With emails, it might be sentiment signals or recurring topics. With meeting transcripts, it might be deal-risk indicators or the questions customers ask most often.
None of that requires building a separate workflow. It requires one more prompt in a workflow that already exists.
How to Implement the Sponsor Tracker
If you run a podcast, newsletter, or any content business that relies on advertising revenue, here’s the specific implementation:
Set up a YouTube monitoring agent for channels in your niche — the ones your potential sponsors are already buying. (Lindy, Make, and Zapier all support this kind of monitoring workflow.)
In the summarization step, add a prompt: “If a sponsor is mentioned in this video, extract the sponsor’s name and add it to a Google Sheet called [your sheet name]. If there is no sponsor, skip this step.”
Review the sheet weekly. Forward the names to your sales team or agency with context on which channels featured them.
The channels you’re monitoring are doing the market research for you. They’ve already sold ad slots to companies with budgets. Those companies are now visible to you, every week, automatically.
The Bigger Takeaway
Ilias was focused on the research angle — which was the right starting point. You build the primary workflow first, make sure it works, then layer in secondary extractions.
But the secondary extractions are often where unexpected value lives.
The primary job of the workflow was intelligence about AI developments. The secondary job turned out to be a weekly pipeline of sponsorship leads.
Build the thing you need. Then ask what else is already in there.
I help founders and operators design AI workflows that do more than one job. My AI consulting and workshop programs are where we build these kinds of layered systems. If you want to see what this looks like for your specific business, reach out.
