On September 30, 2026, more than a dozen pastors, ministry leaders, and nonprofit directors spent a day with us at Trinity Church of Portland for Discern & Build: AI for Ministry Leaders. In the morning we talked through where AI fits in ministry and where it doesn't. In the afternoon everyone built something with a coach next to them.

We promised the room two things: every workflow and skill we used, and notes from the sessions. Here are both, plus what we heard in the feedback survey. It's public, so if you weren't there, you can use all of it too.

Get the workflows and skills

The Ministry AI Starter Kit. Everything from the build lab, in one place and in plain English: 12 ministry workflow guides (a church AI policy, board finance reports, an annual budget, grant proposals, donor follow-up, and more), each with a copy-and-paste prompt that works in any AI chat and the matching skill as a one-click download for Claude. It also includes the 39-prompt Ministry Prompt Library and the take-home pack from the day. It's free; just tell us where to send it.

Get the free Starter Kit →

Session notes

These are summaries, not transcripts. We've left out what people shared about their own ministries during introductions and the build time.

1. The lay of the land: how AI actually works

Clayton Henry, viaForge

  • The family tree. Artificial intelligence → machine learning → generative AI → large language models. Today's major models handle text, images, audio, and video.
  • How a model answers. It learned everything it knows before you ever opened it. It writes one small piece of text (a "token") at a time, by probability. It doesn't learn from your chats. Tokens are also how usage is measured and paid for.
  • The context window is how much the model can hold "on its desk" at once. When it fills up, older material falls off. Knowing when to start a fresh chat is a real skill.
  • RAG beats fine-tuning. Retraining a model on your documents is expensive and rarely needed. Retrieval (giving the model a library to search, like an open-book test) is the practical answer for most ministries.
  • Terms worth knowing: a workflow runs when something happens ("when a form comes in, do this"). An agent gets a goal and boundaries and decides the steps itself. A skill is a saved playbook for a repeatable task. An MCP connector lets AI work directly inside another app.
  • Privacy. On free plans, your data is often part of the product. Check whether your cloud drive's AI features are reading your files by default. For sensitive work, open-weight models running on your own computer keep data on the device.
  • A rule of thumb: AI is strongest at language, synthesis, and organizing messy information. It's weakest at exact math across large datasets. Treat it like an eager junior assistant: give it a role, a goal, context, rules, and the format you want. Before you rely on it, ask three questions: Is it true? Is it appropriate? Who stays responsible?

2. A Christ-centered framework for using AI

Ahshuwah Hawthorne, Brilliance Labs

  • The posture: AI is broadly appropriate for ministry work, with active human oversight. Integrity breaks down when oversight becomes a rubber stamp.
  • Where people stay fully present: Scripture and biblical content, sacred and sacramental moments, crisis pastoral care, discerning God's will, and decisions about people (hiring, discipline, benevolence, membership).
  • Two lanes. AI-adjacent: a person leads, and AI helps with research, prep, and follow-up. AI-central: AI researches, proposes, and builds, and a person approves at every gate before anything is sent, spent, published, or deleted.
  • The discernment flow: start with God's glory (why are we doing this?), bring the real problem, pass it through four questions (glory, calling, resources, readiness), then choose a lane. Afterward, review what was gained and what was lost.
  • Tensions to hold: efficiency vs. presence, stewardship vs. inclusion, speed vs. discernment, individual use vs. communal accountability, and delegation vs. accountability. You still answer for what your agent does.
  • Reality checks: models tend to agree with you (ask "how could this be better?" rather than "is this good?"), they can make things up, and agents can overreach or be hijacked by malicious web content.

Read the full framework, free: A Christ-Centered Framework for Using AI →

3. Panel: fear, headlines, and God's sovereignty

The coaches, moderated by Ahshuwah Hawthorne

  • The room asked about headlines warning that AI could wipe out humanity. The panel's read: fear drives clicks, and some large AI companies benefit from rules that make it harder for smaller competitors to catch up. That's a reason to read these stories carefully, not to dismiss real risks.
  • The deeper answer was theological. God is never surprised by new technology, and his purposes don't depend on any company. That gives Christians steady ground that neither the optimists nor the doomsayers have.
  • The panel agreed the church needs more Christians working in AI, in research, in public conversation, and in ministries, bringing hope instead of fear.

4. Case study: Ask Ligonier

Ted Hallum and John Cobb, Ligonier Ministries

  • Ask Ligonier is an AI teaching assistant grounded in more than 50 years of Ligonier teaching: over 25,000 resources. New content flows in automatically, and confessional documents are weighted more heavily.
  • Getting Scripture right. The model never quotes the Bible from memory. It inserts a marker, and ordinary code fetches the exact verse text from a Scripture service. That closed a misquotation rate they measured at around 16% in raw model output.
  • Testing. 13 automated evaluators check doctrinal fidelity, policy, care and safety, and transparency. The assistant hands off to a person for support questions, discipleship needs, and mental health crises.
  • Their guiding principles: faithfulness to Scripture, doctrinal reliability, and trustworthiness. Ligonier has published the assistant's instructions openly; their work is at labs.ligonier.org.

5. Case study: six months of building at Brilliance Labs

Ahshuwah Hawthorne

  • From March to September 2026 the lab built about 41 projects with Claude, for roughly $1,000 in usage plus a few nonprofit seats.
  • Examples: a full website rebuild by a non-developer, a client portal for our Frontier Finance consulting work, a Bitcoin for Missions funding map, five years of grant spreadsheets merged into one database with a live dashboard, an in-house event registration system, a monthly bookkeeping close that went from days to about three hours, and a private search tool over a 240,000-page theological library that runs on our own computer.
  • Habits that helped: start in chat (voice dictation works), move to Claude Code for bigger builds, use "handoff" documents to carry context between sessions, rehearse before going live, and have one AI check another's work.
  • The pattern under all of it: break every complex task into small ones. Discern, plan, verify, build, test.

6. Real workflows from a bivocational pastor

Thomas Terry, Trinity Church of Portland

Thomas's principle: less admin means more time for people. Shepherding, counseling, preaching, and presence can't be automated. He showed three workflows:

  • Weekly newsletter assistant. It pulls the preaching calendar, announcements, and budget from a Google Sheet and events from Planning Center, drafts the newsletter, and sends a test to staff. Nothing goes out until a person approves it. He ran it live and the issue was done. He estimates it saves the church about $6,000 a year.
  • Sermon research assistant. Two weeks before each sermon it gathers commentary from his Logos library and past sermon notes: hard words, where commentators disagree, cultural background. A printed packet is waiting Sunday morning. Two hard rules: every citation links back to the source so he can verify it (early versions invented quotes), and the AI never writes the manuscript.
  • Motion graphics and video editing. A dedicated agent, given a style guide, produced the explainer animation for a documentary episode for a few hundred dollars in usage, against a typical quote of about $25,000 for a motion designer.

7. Live build: a board finance report in two minutes

Clayton Henry

  • Clayton pointed Claude at a folder of seven messy (made-up) church spreadsheets. In under two minutes it produced a board-ready report: income, expenses, giving, households, budget vs. actual, key takeaways, charts, and citations back to the source files.
  • The lesson: people want answers, not more software screens. Design for the person reading the report.
  • It's the board-finance-report skill in the Starter Kit, ready to run against your own numbers.

8. Getting skills running

John Buckingham

  • John walked the room through installing skills: download a skill's ZIP file, then upload it in Claude's skills settings. Type / plus a skill's name in Claude Code, or use the + button in the Claude app.
  • Skills are plain-text instructions, so most also work in other AI tools. A few need Python or Node.js; the dev-setup-mac and dev-setup-windows skills install those for you.
  • A note on plans: the free Claude plan is too limited for this kind of work, and the $20 plan runs out quickly. Anthropic's nonprofit pricing for Team plans is worth a look.

9. The build lab: what the room made

The rule for the afternoon: ask Claude first, then raise your hand for a coach. By the report-back, the room had:

  • A tool that takes decades of community health training lessons and helps a team update them with current health guidance and adapt them to a new country's language and culture.
  • A volunteer scheduling system for a large church production team, including gentle nudges to leaders when a volunteer is over-scheduled.
  • A fundraiser's spreadsheet turned into a live web app with sign-in, in about ten minutes, on free hosting.
  • Next year's budget, built with the annual budget skill connected to the books.
  • A year of records moved from documents into a shared list, with a person reviewing before anything was written.
  • A website redesign, an email sorting setup, an expense report helper, and the start of a newsletter and social media pipeline.

"In real time, Thomas ran his prompt; his newsletter was done. I am reviewing my schedule to see what is taking my time that Claude can just do with a prompt."
— Stuart Smith, Love INC of Clackamas County

10. Local AI, privacy, and security

Ronald Bynoe, Reckless Systems

  • Ronald demonstrated a silent, offline machine (128 GB of memory, no internet connection) running several AI models at once, generating images and answering questions with nothing leaving the box. Claude set the whole thing up.
  • When local AI makes sense: board meetings, counseling, calls with workers in restricted countries, anything with sensitive personal data, and big overnight jobs like transcribing a sermon archive.
  • A creative example: a modern-English audio adaptation of Paradise Lost. Several frontier models competed passage by passage on the text, and local models produced more than 1,200 images, 26 voices, and thousands of sound effects for a 12-hour audio play.
  • His security cautions: don't treat mainstream messaging or meeting apps as safe for truly sensitive conversations in hostile settings. Consider self-hosted video calls (such as Jitsi), short-lived channels, and keeping separate tools for separate purposes. Plan now for encryption that may weaken in the years ahead.
  • Every model reflects the values of the people who made it. Choose your tools with that in mind.

11. Q&A: grant writing with AI

  • Use AI for the whole grant cycle, not just the narrative: finding opportunities that fit your ministry, building a deadline calendar, reusing strong answers from past applications, and tracking spending against a grant budget.
  • Connect it to a folder of your past grants so it learns your voice. A good rhythm: you write, AI tightens and fits the word count.
  • The Starter Kit's grant-proposal skill is a good place to start.

What we heard from you

Thank you to everyone who filled out the feedback survey. Six of you did, so read these as signals, not statistics.

  • The day averaged 8.5 out of 10. Coach support averaged 9.2.
  • Confidence using AI for real ministry work rose from an average of 4.3 to 6.3 out of 10.
  • Four of six left with something working; the other two left with something partly working.
  • The most helpful parts: the hands-on build lab, with a coach right there, and Thomas's real workflows.

And what we'll change:

  • Tracks. The room ranged from first-timers to people running dozens of agents. You asked for tracks by skill level, and nonprofit leaders asked for tracks by ministry type, because church-specific examples didn't always fit.
  • More time to build. Five of six said two days would be better than one, with time to practice overnight and come back with questions.
  • Know what to build before you arrive. Several people spent part of the afternoon deciding what to make. We'll help with that before the event.
  • A deeper conversation on ethics. One attendee told us the Christian framework and the "doom" questions got raised but not fully worked through, and that they'd welcome thoughtful skeptics at the table. We agree.
  • Pace and polish. Some of the advanced demos moved too fast for newer users, and the screens were hard to read from the back.

"I left with a prototype I can now share with my team and test."
— Tim Cowley, Medical Ambassadors International

What's next

Much of this feedback is already shaping Hopeful 2027: two days of hands-on AI workshops with tracks for new users, power users, founders, and leaders, February 17–18, 2027 in Tampa Bay, Florida. Brilliance Labs is one of its partners, and saving a seat is free.

Thank you to our coaches, who gave a full day to serve this room: Clayton Henry, Ted Hallum, John Cobb, Thomas Terry, Ronald Bynoe, John Buckingham, and Andrey Gorban. And thank you to Trinity Church of Portland for hosting us.

If your church or ministry wants help deciding where AI fits, writing an AI policy, or building tools your own staff can run, Ignite AI Consulting is how we do that work.

Get the free Starter Kit →