Vibe Coding and Tertius of Iconium: An AI Experiment

“Vibe coding” caught my attention recently as a new addition to the AI lexicon, coined about a year ago by researcher Andrej Karpathy. The term describes using large language model (LLM) AI tools to produce code and then running with that code, putting it into production in public-facing apps, without a human coder building (or sometimes even reviewing) what’s “under the hood.”  I was both intrigued and horrified to learn of such a thing.

To better explore my intrigue and horror, I figured I’d give it a shot. Behold, Mission Moment, a fully vibe-coded web app from Mission Central!

The app displays a brief quote from a blog post (written by yours truly, of course) and then links back to the full post. A new quote appears each day, or so I’m told. I created it (if “created” is the right verb) the same day I’m writing this draft, so I haven’t had a chance to see what the app does tomorrow yet. The version I’m linking to as of this post represents 15 minutes of work on my part, if that.

So, first, intrigue. This fresh and shiny web app is worthy of some appreciation.

Oooh, Shiny

For starters, I’m genuinely impressed with both the efficiency and the elegance of the output I got from Claude, the AI app I employed for this experiment (for the power users out there, specifically model Sonnet 4.6). I was able to go from imagining the app to publishing it in the space of a few minutes, and it doesn’t look half bad. After my initial prompt, I only gave Claude three more responses to refine the app into its current form, which puts Mission Central’s brand-standard colors and typefaces into an appealing minimalist aesthetic.

This kind of result is so new to me, I find I’m still capable of wonder; seeing the final product makes me say, “Wow!” I don’t know much about coding beyond the html shenanigans I got up to twenty years ago in the MySpace era (although my site was actually on the lesser-known Xanga). But given the simple concept for this app, I imagine it’s not too complicated to put together on the back end. My hunch is that making an app like this might be a good final project for an introductory coding class—or at least, it would have been before the advent of AI LLMs.

A few years ago, if an idea for an app like this had occurred to me, my options would have been to go enroll in an introductory coding class or (more likely for me) to find a coder friend and ask or pay them to build it for me. Now, I can just type a few sentences, and voila!

It’s easy to see how AI is democratizing coding, empowering non-coders to realize custom digital projects by dramatically lowering the bar of entry. Creating the app took considerably less effort than writing this blog post about it. The ease reminds me of the breakthrough in the early 2000s with the rise of drag-and-drop website editors like Squarespace and Wix, which democratized professional website design. AI is doing for apps what those platforms did for websites.

In terms of real-world utility for this project, I’ve never been much of a “Quote of the Day” person myself, but I know that many others are. My mother is a long-time aficionado of paper spiral verse-a-day calendars, which seem to enjoy fairly stable market demand! I don’t think there’s a crowd clamoring for Mission Central blog quotes in app form, but it’s not hard to imagine an app like this providing a new format for, say, the Henri Nouwen Society’s daily e-meditations.

Oh, the Horror!

But now, the horror. Even that idea of using a web app like this for meditation selections gives me pause. It was so easy to create the app, as I marveled at how it took care of the interface coding, I almost forgot that I had also outsourced a more important feature: the curation of quotes. The app is “picking” lines from my blog posts that its algorithm deems quotable. In the past, to make an app like this work, some human judgment was required. Someone would have to read (or at least skim!) through posts and choose which sentences to pull out as a taste to pique the reader’s intellectual palette. Curation is an art in itself, a creative endeavor that reflects the values and perspectives of the curator.

I don’t personally know the staff of the Nouwen Society, but I trust that they choose each daily meditation from Nouwen’s works judiciously—that some charitable human thought and, indeed, prayer goes into the choice of text and the crafting of the reflection question. If I were to discover that they had outsourced selection and question creation to an AI for the sake of “efficiency,” I would be appalled.

Don’t Forget Tertius of Iconium

This preference for human curation is not merely functional, as though it’s simply a question of whether an LLM can pick quotes “as well as” a human being can. The presence of a human being doing the picking is part of the meaning of the exchange.

In an earlier era, I might have received selected quotes in the form of a mailed newsletter. There’s an epistolary quality to such exchanges, in which a human interaction is mediated through a communication technology, be it ink and paper or bits and bytes. Sometimes when Katie is reading a book in the evening, she’ll pause and read a passage to me that she thought was worth sharing. The e-meditations of the Nouwen Society are a digitally mediated form of such a moment: A person who has themselves been shaped by Nouwen’s work pausing to say, “Take a look at this part here!” Even if this person is a stranger, their part in the drama matters.

Last Wednesday, I attended Evening Prayer at my church, which is going through some of the Daily Offices together during Lent. The New Testament reading was chapter 16 of Paul’s epistle to the Romans, and I was struck by verse 22: “I, Tertius, who wrote down this letter, greet you in the Lord” (NIV). Romans is Paul’s magnum opus, a thrilling and world-shaping theological account of God’s action to save the world—both Jews and Gentiles—through Jesus Christ. Although Paul is the writer of this magnificent letter, he did not physically write it down himself. Tertius was his amanuensis or scribe, setting stylus to papyrus as Paul dictated. Tradition calls him Tertius of Iconium, an early Christian bishop.

It’s not accurate to call Tertius a curator, but there’s a helpful analogy here. Like a curator, Tertius’ role was not essential. Paul could have simply written down his epistle himself (as he did at other times, at least for addenda to his work; see Galatians 6:11). Likewise, I can pick up a book from Henri Nouwen myself and simply read it, without consulting anyone at the Nouwen Society about which quotes stood out to them. But Tertius’ involvement was meaningful. If it wasn’t, verse 22 would not have come down to us as part of the gift given to us in Holy Scripture. Tertius, too, greeted the Roman believers in the Lord. That greeting matters. His exchange with the Romans, although instrumental to Paul’s words, was itself a human exchange, an interaction among persons.

Weighing the Trade-Offs

Likewise, the human curator of the Nouwen Society e-meditations plays a meaningful role—meaning that is entirely missing from my Mission Moment app’s efficient auto-selections. If you click into the vibe-coded app, it’s true that you’ll get a snippet of something I really wrote. But you won’t get the mediated meaning of a person saying, “Take a look at this part here!” The exchange itself is no longer an interaction among persons. As such, it is significantly less valuable than the “inefficient” work of a human curator.

That’s not to say that auto-curation is always wrong. There are situations where AI is preferable to human curation because of its efficiency, despite the loss of human judgment: Technical textual analysis comes to mind. If I simply need to know how often a word occurs in a set of texts, it’s fine and dandy for a computer to tell me. Insights gleaned from tools like Google’s Books Ngram Viewer would not be possible without automated pattern recognition. 

But I would suggest that selection of spiritual reading is better left in slower, human hands. For a real-life project like this, I might use AI to build out the interface, but I would want a thoughtful human reader to choose the text snippets that get displayed each day. Speaking of text curation projects, Henri Nouwen himself was once asked by his publisher to put together quotes from his earlier works for a daily reader. Famously, he refused and said he would write all new material, a labor of love resulting in Bread for the Journey.

At many junctures in our work, we face a similar trade-off between meaning and efficiency. That was true before the AI revolution, but the trade-offs will become more stark as depersonalized work becomes easier and easier to do. We have to weigh the trade-offs before leaping ahead with the fastest option. Allow me to propose a few “weighing questions” to ponder when considering the use of AI at work:

• What gives this project meaning?
• Is there a person whose presence in this work adds value just by being a person? How can we keep them in the mix?
• What do the end-users get out of what we’re offering? Are they looking for something human?
• Is this use of technology freeing me up to add the human touch where it matters? Or is it taking the human touch away by depersonalizing a human exchange?

I realize these questions are abstract. Meaning can be hard to quantify; it emerges to our view not through measurement, but through discernment, which itself is an inefficient human art. But in the age of AI, such discernment is more vital than ever. Let’s take time for it.