Guide

Chapter 17: AI, Formation, and Leadership

Josh Shepherd3 min read

Accurate is not always formative.

Picture formation time after Easter—youth room, adult class, or nonprofit volunteer huddle: a leader pastes an assistant-generated discussion guide into the room’s thread—clean transitions, plausible questions—and halfway through a teenager says quietly, “Did anyone live this week, or did we just read about it?” The line is not anti-tool; it is formation’s line. A flawless FAQ about spiritual disciplines can still malform if it substitutes consumption for community, information for obedience, screens for presence. Formation asks a harder test than content:

Does this process, over time, make people more like Christ in verifiable ways—together?

Generative AI can resource that process; it cannot be the process.

Formation vs. information (precision without sentiment)

Jack Mezirow’s transformative learning and related adult-education theory help explain why dissonance + reflection matter—but do not treat those citations as permission to brand your eight-week cohort as a validated psychological battery. The Four Necessities language some formation designs use—dissonance, action, reflection, community—is a design heuristic grounded in plausible theory, not a lab-proven instrument.

James K. A. Smith’s cultural liturgies remind us that habits form desire beneath argument—so the habit of “click, receive, feel inspired” competes with the habit of showing up late to rehearsal because someone needed a ride.

Keep screen harms and LLM harms analytically distinct: TikTok timelines and chatbots both tug attention; they are not the same failure mode. Collapsing them weakens your formation diagnosis.

When AI helps formation infrastructure

Accessible readings for low-literacy neighbors—human curated.
Language support for multilingual communities—human reviewed.
Scheduling and logistics so volunteers sleep.
Draft discussion guides—teacher edited for local context.
Use-case logs—not glamour work, but memory: what you tried, who reviewed it, what you refused, so the next cohort is not reinventing the same mistakes. (Private documentation is different from publishing voice you have not owned; chapter seven’s maturity map keeps that distinction explicit.)

When AI undermines formation

These are extensions of the refusals named in chapter thirteen; here the lens is formational outcome, not only ethics.

Simulated one-to-one care marketed as companionship.
Personalized “spiritual direction” without oversight.
Auto-generated discussion-group curricula nobody has lived.

Leadership formation: the mirror test

Leaders are high-leverage and high-risk users. The performance–metacognition gap Fernandes et al. report in Computers in Human Behavior (2026, https://doi.org/10.1016/j.chb.2025.108779)—sketched more fully in chapter seven—means tool-assisted gains can coincide with miscalibrated self-assessment. That is especially dangerous when followers cannot see your rough drafts: they infer competence from output, not from the chain of review behind it.

Practices that help:

Named mentors who can say “that post sounded unlike you.”
Sabbath from publish rhythms.
Public disclosure defaults for high-trust channels (the disclosure craft in chapter ten applies here as integrity infrastructure, not only comms hygiene).

Churches: discipleship rooms

Parish and nonprofit delivery rhyme but are not identical—logistics, liability, and volunteer culture differ. In my own notes I treat churches as strongly formation-shaped while the full institutional playbook for church-specific delivery is still being written; the moral bar in this chapter does not wait on a finished manual.

Youth: AI-assisted study tools alongside adult relationship—never replacing adult circle leaders.
New members: translation help yes; substitute community no.
Counseling triage: if it is not a human with office hours, it is not ongoing counseling—it is triage at best.

Parish vignette (Wednesday): A new-member lunch runs long; someone asks a newcomer from another country what “calling” meant in their former congregation. Whoever is leading could let a phone summary answer—or could close the laptop, ask a follow-up question only a human who has watched the family for three weeks would ask, and carry silence without filling it. AI might later help draft a welcome email; it cannot host that silence without stealing something only bodies trade.

Programs are not “eight weeks because science said so”

Eight weeks (or any arc) is a product choice—defensible formationally, not because a journal proved eight beats seven.

Movement leaders who design learning

Even if you never lead a local congregation, you architect formation through books, cohorts, and platforms. The same bar applies: does your stack increase embodied practice or only increase downloads?

Reflection questions

Name one habit your digital tools have trained in you—virtue or vice.
Where does your content explicitly send people toward tables, walks, and silence together?
Who checks whether your “formation product” still forms you?