How Arts Organizations Can Think About AI Connectors

Start with what you want it to help you do. I wouldn’t connect everything just because you can. Pick a useful workflow, involve your IT team or technology partner, and give the tool the access it needs for that work.

Understand whether it can read, write, or both. Read access lets AI retrieve information. Write access lets it take actions that change something in the connected system. A connector that can read your project management system can help you understand your day or identify upcoming deadlines. Giving it write access might let it create tasks directly in that system. Both can be useful, but they’re different permissions with different consequences.

My recommendation is to start with read-only. Personally, I would keep Email, Google Drive, SharePoint, calendars, and similar systems read-only. I’m not comfortable giving AI permission to send emails, change meetings, edit shared files, or change who has access to information. Whether it misunderstands what I asked or makes an unintended change, the potential consequences are too high for my comfort.

I’m more comfortable with write access in internal project management systems like Asana or Hive, where creating tasks and organizing work can be really useful. I’d still look carefully at which actions we’re enabling. Creating a task is different from deleting tasks or changing everyone’s deadlines.

Make those decisions at the organizational level. Decide which capabilities are allowed, enable those, and disable the others. These boundaries should be enforced through settings rather than left to individual employees to manage through prompts.

Organizational plans can offer centralized controls, but the options vary by platform, plan, and connector. There may also be permissions to review in both the AI tool and the system it’s connecting to. If a connector doesn’t offer the controls your organization needs, I wouldn’t enable it.

Read-only still deserves a careful look. It prevents changes to the connected system, but it still lets AI retrieve information. For arts and culture organizations, that could include donor and patron information, artist contracts, personnel records, or board materials.

Think about which information the connector actually needs to reach. It’s also a good time to review existing permissions. Just because someone technically has access to something doesn’t mean that access was thoughtfully set up in the first place—and AI can make information that was previously conveniently buried much easier to find.

The risk goes beyond AI misunderstanding you. Connected content can contain malicious instructions that try to get AI to do something unintended. That’s called prompt injection. You don’t need to become a technical expert, but it’s another reason to be thoughtful about which connectors you trust and how much access you give them.

You still need to check the work. A direct connection doesn’t guarantee that AI pulled the right dates, used the right filters, or found all the relevant information. Especially when the results inform a decision or report, verify what it used and how it arrived at the answer.

Someone should be responsible for each connection. Start with a small workflow, test what the connector can and cannot access or change, and know how to revoke access. Revisit the settings as capabilities change and remove connections that are no longer needed.

I see a lot of opportunity here. With a useful purpose and a thoughtful setup, connectors can make AI much more helpful in the day-to-day work of an arts organization, especially in managing large sets of data or operational systems that would otherwise take up a human’s time with tedious, manual processes. That’s time that can now be (better) spent doing more rewarding, creative work.