We Are Not Building a Chaplain

Every few months someone suggests, with the best intentions, that SoapBox should offer an AI chaplain. A gentle voice in the app that a person could talk to when they are struggling, that would pray with them, offer Scripture, listen without judgment at any hour. The pitch usually ends with the observation that people are already doing this with general-purpose chatbots, so why not give them a version that is at least grounded in the faith.

It is a reasonable pitch. The demand is real. It is also the one feature I have decided we will never build, and I want to explain why, because the reasons are not squeamishness about technology. I build AI for churches. I am not afraid of the tool. I am afraid of what that particular use of it does to the church.

What a chaplain actually is

A chaplain is not a source of comforting sentences. If that were the job, a well-stocked bookshelf would do it. A chaplain is a person who has been sent. Sent by a congregation, a denomination, a hospital, a body of some kind that vouched for them and can be called to account for them. When a chaplain sits with you, the church is sitting with you, in the form of one accountable human being who will remember this conversation, who can be found tomorrow, and who can be wrong in ways that have consequences for them.

A machine cannot be sent. Nobody laid hands on it. No body stands behind it that can be called to account when it says the wrong thing to the wrong person on the worst night of their life. It will not remember you in the way a person remembers, it cannot be found tomorrow, and when it fails, the failure lands entirely on the person who trusted it. That is not a technical limitation waiting for the next model. It is what the word chaplain means, and no amount of capability closes the gap.

The product problem underneath

Set the theology aside for a moment and look at it as a founder, which is also my job.

The moment our software speaks in a pastoral voice, the church that deployed it becomes responsible for a voice it does not control. A pastor who installs a chaplain feature has, without quite deciding to, delegated the most sensitive conversations in their congregation to a system they cannot supervise, trained by people they have never met, that will say something different tomorrow than it says today. When it goes wrong, and it will go wrong, the person who gets the phone call is the pastor, and they will have no idea what was said.

Worse, it removes the pastor from exactly the conversations they most need to be in. The member who opens the app instead of calling is often the member the pastor did not know was struggling. An AI chaplain does not surface that person. It absorbs them. It gives the church the comfortable impression that care is happening while quietly ensuring that no human in the church knows it is needed. A tool that hides need from the people responsible for meeting it is not a ministry tool. It is a way of not finding out.

What we built instead

The line we drew is simple to state: our AI is a study aid, not a shepherd. It serves anyone who opens it. An explorer who has never read the Bible can ask what a passage means and get a plain answer. A believer can sit with a chapter, reflect on it, and have the ordinary questions of study answered with the text in front of them, and with their own church's teaching when their church has shared it. For the person who wants to go further, there is a deep study mode that reaches into the original Greek and Hebrew, the cross-references, and the commentaries, the kind of research that used to take a shelf of books. For a pastor it is a research aid that gives back the hours a sermon takes to prepare. What it does not do is stand in for a person. When a question stops being about the text and starts being about a life, the tool points the way to a pastor and a real community, because that is whose conversation it is. It works alongside the shepherd, never in place of the shepherd.

That means it helps a pastor prepare a sermon by pulling together commentary and cross-references they would otherwise spend a Saturday hunting for. It drafts the administrative mail that eats a pastor's week. It notices, from the ordinary data a church already has, that a family has not been seen in a while and puts a note in front of a human, who decides what to do about it. It helps a volunteer coordinator fill a schedule. It never talks to a member as if it were the church.

The test we apply to any new feature is the same every time. Does this give a pastor back an hour they can spend with a person? Or does it spend that hour for them, with the person, so they never have to show up? The first we build. The second we do not, no matter how the demand curve looks.

What about the person who has no one?

This is the honest objection, and I do not want to wave it away. There are people who will open a chatbot because there is nobody else, and telling them the machine is not a chaplain does not give them a chaplain.

But I think the objection points the wrong direction. The answer to a person with no one is not a better machine. It is a church that finds out. That is precisely what our tools are for: making it more likely, not less, that a real human being in a real congregation learns that someone is alone and goes to them. Every hour of administrative work we take off a pastor's desk is an hour that can go to the person who would otherwise be typing into a box. I would rather spend our engineering making that hour available than building a very convincing reason for the church not to need it.

I will also say plainly that I do not think the machines are dangerous because they are alive or because they want to replace pastors. The systems we have today want nothing. The danger is entirely on our side: that we will let a fluent, tireless, always-available tool do the thing that was supposed to require a person, because the person is slow and the tool is not, and we will call it care.

The line, stated once more

We are not building a chaplain. We are building tools so that the chaplains, the pastors, the elders, the volunteer who brings the meal, have more time and better information to be what only they can be. If a competitor builds the chaplain and wins the market with it, I will have been wrong about the business and right about the church, and of the two I know which one I can live with.