Chapter 7: The Maturity Model and Why Experimentation Matters

I want to honor the reader who has decided—so far—not to engage.

Your caution may be wise. AI touches vulnerable people, public truth, and the people you are responsible to form. Some boards are right to move slowly. Some pastors are right to protect the pulpit from experiments they have not prayed through—and some executive directors are right to hold the same line for an all-hands script or a major donor letter until they have tested what they are signing.

And yet: abstinence without bounded experience is not always wisdom. It can be avoidance—a decision made from rumor, fear, or moral performance rather than from hands-on knowledge of what you are refusing.

You cannot responsibly draw bright lines around what you will not automate until you know—at small scale—what the tool does when it meets your voice, your ethics, and your people.

A five-stage map (heuristic, not science)

Enterprise roadmaps from MIT CISR and vendor frameworks like Gartner’s maturity language already describe multi-stage AI adoption for organizations. The stages below are a working map for formation-shaped leaders—aligned with that tradition but split so you can see where caution becomes avoidance. They are not a validated developmental psychology instrument. They are a compass: locate yourself without shame, choose a next step, and expect different domains (communications, counseling and direct care, operations) to sit on different points at the same time—parallel curves, not one scoreboard for the whole org.

In work with teams, I keep seeing the same arc in slightly different vocabulary: from native awareness through structured experimentation with explicit safety rails, then governance and adoption, iterative improvement, and finally public leadership that demonstrates mature use across an organization—still rare today. The labels here split “awareness” into observation versus first hands-on tries so the hinge stays visible: learn before you publish, and document what you learn so the curve is real, not performative.

Stage 1 — Awareness
You know the landscape exists; you may have opinions; you have not run structured, personal experiments. Valid place to be.

Stage 2 — Exploration
You are watching demos, reading policies, asking what peers do. Still mostly observational. Many vendor charts collapse this into “adoption”; I name it separately because consuming AI content is not the same as learning what AI does to your judgment under pressure.

Stage 3 — Experimentation
You try low-stakes uses: outlines, counterarguments, formatting, private drafts—not surprise deployments in your highest-stakes public moments. You log surprises and failures. In organizational settings, this is the phase where many teams adopt explicit moratoriums: no shipping AI-generated or AI-affected content as if it were finished public voice until you have review and disclosure norms; no feeding private donor, counseling-related, or personnel data into tools until policy and IT say how data may leave your boundary. That posture matches what chapter one named as experimentation without publication—speed and accountability kept distinct on purpose.

Stage 4 — Integration
A few bounded practices enter regular workflow with disclosure norms and review. The tool is present; you remain author of public moral weight. At org scale, this is where principles get written down—governance and ethics with the kind of clarity that lets people execute without guessing what “allowed” means.

Stage 5 — Responsible innovation
You help others navigate: playbooks, ethics conversations, modeling repair when something goes wrong. Not “power user cosplay”—stewardship. The capstone is not fluency; it is demonstrable maturity others can imitate without abandoning accountability.

Teams can sit on different stages for different domains. The map is two-dimensional in real life.

Why experimentation is prerequisite to discernment

Heifetz-style adaptive work requires learning in contact with reality. AI adds a twist chapter two already named with care: with ChatGPT-class tools on demanding tasks, people can gain performance while losing calibration—including, in at least one major study, greater overconfidence among those who rate their own AI literacy higher. The point here is practical: try with feedback—co-review, red-team prompts, “delete the best sentence if you cannot verify it,” scheduled retrospectives—so performance does not outrun judgment.

Green / yellow / red as discernment, not permission

Think traffic lights for risk surface, not virtue badges. Chapter twelve picks this vocabulary back up for pace and rotation; for now, the substance is enough to share with a council.

• Green: formatting, brainstorming private notes, summarizing your own transcript with verification.
• Yellow: drafting for public view—always with human review and disclosure conversation.
• Red: crisis intake and ongoing counseling, public accusations, anything where harm is asymmetric and the model cannot carry accountability.

This is not a compliance checklist that replaces prayer. It is a shared language so councils and staffs can move at different speeds without calling each other cowards or cowboys.

Organizations: bring the curve, not only the policy

Nonprofit leaders might add staging: IT pilot, comms pilot, program pilot—each with different disclosure rules. Churches might pair public denominational and ecumenical guides (the wider conversation is already underway—chapter two named a few bodies in play) with local stories so theology does not feel imported from a headline.

One pattern I use in organizational sprints is a Discovery Lab shape: about four weeks, time-boxed, whose main output is documented use cases and lessons, not a rush to new public copy. It is the institutional version of “try small, log surprises, decide what earns integration.” The label is less important than the discipline: evidence before normalization.

Resistance is data. Ask what people fear losing—presence, fairness, jobs, truth—and design experiments that name those fears as hypotheses.

Urgency, FOMO, and the counterintuitive gift of boundaries

The market noise says everyone is already “AI-native” except you. Boards read headlines; staff feel behind; vendors promise transformation by Tuesday. Bounded experimentation is how responsible leaders buy calendar and emotional space—not to freeze forever, but to refuse the simulation of maturity (slides without practice, policies without pilots). Sometimes the tool’s best gift in this season is administrative or drafting relief so a leader can slow down where formation actually happens: conversation, prayer, reading, sleep. Faster output is not always the win; legitimate pace is.

Synthesis

You earn the right to set boundaries by first understanding what you are bounding. That line is not a slogan. It is the difference between ethics and aesthetics.

Next: if experimentation touches language, voice becomes the urgent question—not as vanity, but as the carrier of earned authority—and the risk of the almost-right sentence that ships heresy in fluent prose.

---

Reflection questions

1. Where are you—and where is your team—on the five stages, domain by domain?

2. What is a two-week, low-stakes experiment that would teach you something you cannot learn from debate alone?

3. Who will give you honest calibration feedback so performance does not outrun judgment?