As an acquisitions editor at O’Reilly, I spend considerable time tracking our authors’ digital footprints. Their social media posts, speaking engagements, and online thought leadership don’t just reflect expertise—they directly impact book sales and reveal promotional strategies worth replicating. Not surprisingly, some of our best-selling authors are social media mavens whose posting output is staggering. Keeping up with multiple superposters across platforms quickly becomes unsustainable.

I recently built an AI solution to manage this challenge. Using Relay.app, I created a simple workflow to scrape LinkedIn posts from one author (let’s call her Bridget), analyze them with ChatGPT, and send me weekly email summaries about her posts and which got the most attention. The main goal was to follow what she said about her book, followed by thought leadership in her field. The setup took five minutes and worked immediately. No more periodically reviewing her profile or worrying about missing important posts.

But by the second summary, some limitations became apparent. Sorted by likes and impressions with generic summaries, every LinkedIn post was receiving the same treatment. I had solved the information overload problem but now needed a way to extract strategic insight.

To fix this, I worked with Claude to turn the prompt into something closer to an agent with basic decision-making authority. I gave it specific goals and decision criteria aimed at shedding light on promotional patterns that are not always easy to follow, let alone analyze, in a flurry of posts: autonomously select 10–15 priority posts per week, prioritizing direct book mentions; compare current performance against historical baselines; flag unusual engagement patterns (both positive and negative); and automatically adjust analysis depth based on how posts are performing.

The new report now provides deeper analysis focused primarily on posts mentioning the book, not just any popular post, along with strategic recommendations to improve post performance instead of “this had the most likes.” Recommendations are sorted into short-term and long-term promotion ideas, and it has even proposed testing novel strategies such as posting short video clips related to book chapters or incentive-driven posts.

The report isn’t perfect. The historical analysis isn’t quite right yet, and I’m working on generating visualizations. At the very least, it’s saving me time by automating the delivery and analysis of information I would otherwise have to get manually (and possibly overlook), and it is beginning to provide a starting point for understanding what has worked in Bridget’s promotional program. Over time, with further work, these insights could be shared with the author to plan promotional campaigns for new books, or incorporated into larger comparisons of promotional strategies between authors.

While working on this, I’ve asked myself: Is this an AI-enhanced automated workflow? An agent? An agentic workflow? Does it matter?

For my purposes, I don’t think it does. Sometimes you need simple automation to capture information you might miss. Sometimes you need more goal-directed, flexible analysis that results in deeper insight and strategic recommendations. More of a helpful assistant working behind the scenes week after week on your behalf. But getting caught up in definitions and labels can be a distraction. As AI tools become more accessible to everyone in the workplace, a more valuable focus is found in building solutions that address your specific problems using these new tools—whatever you might call them.