Getting Started with AI in Rock
While Rock's documentation covers how AI Agents work, it doesn't answer the questions that decide whether the rollout goes well, which can ultimately define its success at your organization. Our team has an approach that we’ve developed in partnering with leading churches, and we think it can help you too.
Here are the three critical stages to a successful rollout:
- Ask first. Find out what people expect before you configure anything.
- Roll out in stages. Don't open access to everyone at once.
- Watch and adjust. Once it's live, pay attention to what's actually happening.
Before You Turn It On
It’s easy to skip straight to configuration when you’re rolling out a big new feature. But this approach can leave staff confused and lead to partial or incorrect adoption.
As one of the churches with early access to Rock IQ, LCBC developed an approach that is worth borrowing directly. Before building anything, they sit down with the people who'll use it and ask a question, the way a product manager would ask it before writing a single line of code: if you had an AI agent right now, what would you expect it to do for you? It’s the same principle behind outlining any agentic workflow before you build it: get the real problem and a real example first. LCBC runs it at the organization level instead of the task level.
That conversation surfaces two things:
- The requests you can't fill. Usually, based on a misunderstanding of what the tool does. Address that early, before someone is already disappointed.
- The requests you can fill easily. Sitting right there in what people already asked for. Build those first. An early win tied to a real request does more for adoption than any announcement email.
The Rollout Ladder
Next, we recommend a laddered rollout in three stages instead of turning it on for everyone at once.
First, pick your agent type on purpose. A Chat Agent and an MCP Agent carry different costs and different security surfaces to manage, so don't just default to whichever one you clicked first. Rock Cast's AI Summit Q&A episode covers how model choice and MCP connections change what you're billed for. It’s likely that most people will use a mix of both, but you’ll want to be well-versed on the differences between them.
Rung 1: Just you. Rock's documentation calls this out under its own heading: “New agents start locked down.” By default, every new agent restricts access to the Rock Administrator role only, automatically, before you configure anything. Nobody else can see it until you deliberately open it up. So, the first stage happens automatically: you're the only one using it until you decide otherwise.
Rung 2: A small group of early adopters. Once you are ready, open access to a handful of people. Rock's agent security documentation suggests a role like “RSR - Staff Workers” as a typical starting point for an internal agent. Go smaller than that if you can. A few people who are already curious will find any real concerns and raise what will be common questions so you can get ahead of them.
Rung 3: Real security roles, mapped to your staff. This step is critical, so don’t skip it. Make sure your security roles are set up accurately in Rock. Rock checks four things before an agent acts:
- Who can see the agent at all
- Which skills they can use inside it
- Which specific tools inside those skills
If a person can't see a record in Rock, the agent can't show it to them either. The agent runs on the permission system you already have. Keep a volunteer-facing agent read-only for most tasks. Strip a public-facing agent down to a few essentials: service times, event details, campus locations, nothing more.
If your church has already written down what data never goes into an AI tool, health information, counseling notes, financial details tied to a name, apply that same list here. Our free AI Use Church Policy Template covers exactly that list if you haven't built one yet.
The bigger obstacle here, though, is usually hesitation. Rock Cast's episode on what's really blocking church AI adoption traces resistance to risk aversion, perceived cost, and a staff that's already at capacity. We recommend starting with people who are already curious, giving them a small, bounded win, and letting them bring the next person along does, which is exactly what Rung 2 is for.
After It's Live
A rollout doesn't end when access is granted. It ends when you actually know what people are doing with it and is a step we commonly skip when we’re busy.
Kingdom City's approach is the piece most rollouts miss entirely: once the agent is live, look at what staff are actually asking it to do. Query the session history and read a sample of real requests. They rarely match what you imagined people would ask. That single habit shows you three things.
The gaps. If three different people ask the agent to do something it wasn't built for, add the missing skill or tool. If people keep asking questions that the documentation already answers, close that gap the way we recommend for any AI tool: required, mapped to the actual role, and repeated on a consistent rhythm. Rock Cast's episode on the hidden costs of untrained staff dives into what happens when you skip that step: lost productivity, missed connections with your congregation, and software nobody remembers why they're paying for.
The coaching. As staff begin using the tools, they may not always get it right. This is an excellent time for coaching conversations. Rock Cast's episode on keeping AI out of its own blind spots covers this exact failure mode in detail. A quick, specific correction beats a general reminder in a staff meeting. For something more structured, Triumph's AI Fluency Rubric gives you a five-level framework for naming where someone's AI skill sits today and what growing it looks like next.
The wins. Find the person using the agent thoughtfully, catching its mistakes, getting real value out of it, and say so to the team, by name. That does more for adoption than a mandate, because it's proof from someone your staff already trusts.
The Point of All of This
Rock did the hard technical work already. Ultimately, though, the success of the rollout depends on the steps you take to guide your staff through the process.
If you'd like help thinking through your own rollout, our AI Agentic Workflows Cohort walks a small group of churches through building and holding one real workflow together. If you'd rather talk it through directly, reach out. We're glad to help either way.