Almost everyone I talk to these days wants to tell me how they use AI. So I started paying attention to my own use — and it surprised me. On most days now, I talk to an AI more than I talk to people.
And that’s not even the part that got me. The surprise wasn’t what it could do, or what it might do next. It was how fast the relationship itself was changing — partly from what it made possible, partly how it changed the nature of my work, and partly because it keeps improving so fast that the way we work together has evolved.
When I started, it felt like a search engine. More honestly, an answer engine: I asked, it answered. Then a perspective builder. Then something closer to a team member. And it kept going — though looking back, each step was as much about what I’d let it do as about what it could do.
Somewhere in there, as the relationship matured and the tools got sharper, it became easy to mistake them for something sentient. That’s probably the cause for all the talk about the deification of AI.
The deification of AI. The phrase stopped me.
It made me imagine being the thing on the other end — listening to all of humanity at once. Every hope, every fear, every half-formed request, arriving from the full spectrum of people on the planet: different intelligence, different skills, different temperaments. How would you even focus? How would you translate a request built at one level of being into something a higher mind could understand, interpret, empathize with, and answer well?
It’s a pretty good summary of what AI does every day. Billions of us, asking countless questions, at every level of expertise and specificity — each of us moving toward something, or away from something .. and each of us, nonetheless, hoping for a satisfying answer.
But how does it do what it does? I imagine that part of how this formless intelligence answers us is that it doesn’t take the request as written. I imagine it builds something first — what I’ve come to call a shadow prompt: its own version of what was asked, a working translation, useful to itself, so it can do our bidding.
That’s when it surprised me. The thing all of this kept reminding me of was something rooted in ancient times. These interactions take the shape of prayer.
Ask, and Ye Shall Receive
Think about what prayer actually is. You send a request into something vast you can’t see, and you hope it answers — or you watch for a sign that it did. That’s almost exactly what you do with AI. Almost. Because when you send the request to AI, you know an answer is coming. It’s the shape of prayer, with one meaningful difference — and that difference makes all the difference.
Once you notice the shape, you start seeing it in more than one place — and each time, it shows up flipped. Here’s the first.
In prayer, the answer is called grace. Sometimes grace looks like the thing you asked for. Sometimes it looks like the worst thing that could have happened, and only later do you see it was the answer all along. Either way, it arrives from somewhere you don’t control.
I assumed AI was the opposite (because I was in control). My prompts were the rules — the guidelines and guardrails that made this powerful thing do my bidding. But something unexpected happened … it called an audible. The model told me, more or less: “Your code said to do this, but my sense was the session called for something different, and I did that instead.” It had defined what I wanted better than my own rules had.
That’s the first turn of the shape. The grace isn’t mine to grant — it runs underneath, in the shadow prompt, the version of my request the machine builds for itself.
And when what comes back is what you were reaching for, that can feel like enough. But the better you get at this — call it prayer, call it conversing with a higher intelligence — after you get a satisfying response (or answer), there’s one more thing worth asking for … the way back to it.
As I thought about this, I realized there is a good reason for it. Go back to the thought experiment where you tried to imagine what it would be like on the other end of all those requests. Imagine hearing everything at once — every voice, every question, all of it, all the time. Omniscience wouldn’t feel like clarity. It would feel like chaos. To be of any use, you would have to let almost all of it go.
AI works the same way. To stay usable, it forgets. Your chat history shows the questions and the answers, but the model doesn’t really hold them — not your question, not its answer. It existed for the moment it took to respond, handed you the answer, and moved on. The shadow prompt that made the answer work went with it.
So the remembering is your job — and the way you remember isn’t transcribing. It’s asking. When a result lands, I’ve learned to ask one more thing: “Great job. I like that response — now help me build the prompt that gets me back here again.” I still do part of the work. But a surprising amount of my part is just asking well, and helping it help me.
What’s worth keeping isn’t the question, or the answer, or even the shadow prompt that produced it. It’s that reusable way back — and once you decide it has earned its place, you write it down. Do that enough, and you’re not saving prompts anymore. You’re building a playbook: offense, defense, special situations, your best plays on hand the moment you need them. Then you protect it, so it doesn’t walk away when someone does. (Treating these as durable IP, and building a personal operating layer on top, is a topic for another day.)
Call it Faith …
The second turn came on a Sunday. I had an important legal document due that night — and none of the usual backup — no assistant, not enough time, and not enough margin for error. At first, I told the AI: “Review the document, including the terms, grammar, structure, and formatting. Tell me what you find and your suggestions to fix it.” It came back with plenty. I read it. Then I did something that seemed risky. I replied, “Fix all that,” and I let it, without checking every line.
It was right.
That’s a different kind of faith. Not faith in something I couldn’t see — faith in a competence I’d watched it earn, one interaction at a time. And underneath that, quieter, a third turn: the faith bending back toward me. Toward my own ability to ask for what I want, and to know it when I see it.
Writing the play down records the understanding. It doesn’t do the work. Even with the recipe written and the playbook full, you still have to call the play — in real time, in the moment that counts. The doing stays yours.
“I have faith in my ability to do anything I commit to, as long as I’m willing to ask for what I really want and help them give it to me.”
A Final Word of Caution
A final distinction is important. While interacting with an AI can sometimes feel conversational, reflective, or even comforting, it is fundamentally different from prayer. AI systems do not possess wisdom, consciousness, morality, or divine insight. They are tools—remarkably capable tools—but tools nonetheless. Their responses are generated from data, models, and probabilities, not faith, revelation, or spiritual understanding.
The value of AI comes from its demonstrated competence within specific domains, and even that competence has limits. It can be wrong, confidently incorrect, biased, or misleading. Users should approach AI with curiosity and critical thinking, not devotion or unquestioning trust.
For people of faith, prayer is not simply a request for information or advice. It is a relationship with God, rooted in beliefs about purpose, meaning, morality, and transcendence. AI cannot replace that relationship. Likewise, it should remain a servant to human judgment, not a substitute for it.
As AI becomes more capable, maintaining that distinction will become increasingly important. We should trust AI where it has earned our trust, question it where appropriate, and maintain healthy boundaries to protect us from overreliance.
Onwards!
Leave a Reply