AI as Mission Infrastructure
When Expertise Becomes Cheap
~4 min read
The title of this chapter contains an exaggeration. Expertise is not becoming cheap. Access to fragments of expert-like assistance is.
That distinction explains one of AI’s most important organizational effects. A small ministry historically encountered hard capability boundaries. Translation required a translator. Custom software required a developer. Design required a designer. Data analysis required an analyst. Large organizations bought expertise; smaller ones waited, improvised, recruited volunteers, or abandoned projects.
Generative AI moves some of those boundaries. A non-designer can generate candidate layouts. A ministry worker can prototype software. A missionary can ask sophisticated questions of a spreadsheet. A pastor can receive a first-pass explanation of unfamiliar technical material.
The user has not become an expert. The organization can nevertheless attempt more.
When Assistance Gets Cheap
The same phenomenon can improve equity. A remote church can receive first-pass legal, design, technical, or language guidance that helps it identify when professional help is truly necessary.
A small mission organization can prototype an idea before seeking funding. The strategic gain is not that experts disappear. It is that scarce expert attention can be concentrated where the consequences or complexity justify it.
Productivity Is Not One Number
Workplace evidence cautions against turning this into a universal productivity law. In one large study of customer-support workers, AI assistance improved average productivity and benefited less-experienced workers especially strongly. The likely mechanism included making patterns associated with stronger workers available to weaker ones.1
In another setting, experienced open-source developers working in complex repositories they already knew became slower with the early-2025 AI tools tested by METR. Later evidence using newer agents pointed in a more positive direction, while selection effects made precise estimates difficult.2
The durable conclusion is not “AI makes workers 14 percent faster” or “AI makes developers 19 percent slower.”
It is that productivity depends on task, user, tool, workflow, review burden, and time period.
A translator may spend less time producing routine first drafts and more time resolving difficult ambiguity.
A developer may generate less boilerplate and spend more time on architecture, security, and maintenance.
A theological educator may spend less time repeating elementary explanations and more time mentoring students.
A designer may produce candidate variations rapidly and spend more time understanding the people for whom the product exists.
This is not guaranteed. In some settings review overhead can exceed generation savings. Organizations need to measure their own workflows.
If AI makes junior work easier, organizations may hire fewer juniors. Over time, where do senior experts come from?
This supervision pipeline problem matters in translation, software, research, and ministry training. Experts usually become experts by performing lower-level tasks and receiving correction.
Organizations should not optimize away every developmental task without deciding how future competence will be formed.
AI can help create new training pathways, but the pathway must be intentional.
Productivity studies frequently measure task completion time. Ministries should also measure quality, correction burden, learning, and downstream consequences.
If an AI-generated report takes half the drafting time but requires leadership to correct subtle inaccuracies later, the apparent saving may be false.
If AI translation doubles output while overwhelming reviewers, the bottleneck has moved rather than disappeared.
The Bottleneck Moves
AI often does not remove a bottleneck. It moves it. When content is scarce, production matters. When content is abundant, curation matters. When coding is scarce, prototyping matters. When prototypes are cheap, maintenance and security matter.
When answers are scarce, explanation matters. When answers are abundant, trust and judgment matter.
Part IV follows these moving bottlenecks through software, resources, administration, and mission intelligence.
What Becomes More Valuable
The more durable economic shift is the falling cost of candidate generation. A system can now produce a translation draft, a design draft, a code draft, a research synthesis, a teaching outline, or a policy draft in moments. As candidate generation becomes cheaper, the value of deciding whether a candidate deserves use rises. Someone still has to know whether the translation is faithful, the code is secure, the statistic means what the dashboard implies, and the theological explanation belongs to the church’s tradition. AI does not remove expertise. It redistributes where expert time has highest leverage.
A small ministry should welcome expanded capability without claiming expertise it does not possess.
“I can prototype this” is not “I am now a software engineer.” “I can generate a translation” is not “I can approve this translation.” The distinction is particularly important in mission because teams may work far from specialist support. AI can help them cross the first barrier. It should also make escalation points explicit.
A professional role combines many skills. AI may automate some components and leave others untouched.
A translator researches, drafts, negotiates terminology, understands culture, manages projects, and reviews. A developer codes, designs architecture, secures systems, communicates with users, and maintains deployments. A pastor studies, teaches, counsels, leads, administers, and bears relational responsibility.
Saying “AI can do translation” or “AI can code” hides this bundle. Organizations should decompose roles into tasks before deciding how staffing should change. As production becomes abundant, the credible institution becomes more valuable. Readers will have access to endless theological explanations. They will still need to know which church, school, translator, or scholar stands behind the material.
Cheap expert-like output can therefore increase the value of identifiable human and institutional trust rather than reduce it.