Last year the question was whether AI could draft your newsletter. This year the vendors are demonstrating something different: assistants that connect to your actual systems — the membership database, the calendar, the finance ledger — and act. Look up the lapsed members, draft the renewal reminders, schedule the send. The industry calls this agentic AI, and the connectors that make it possible have standardised remarkably quickly.
Why academic organisations should care
Because the work it absorbs is exactly the work that drowns small teams: cross-referencing spreadsheets, chasing renewals, assembling the same report monthly, moving data between systems that were never introduced to each other. A society run by volunteers feels this more than anyone. The promise is not fewer people; it is the same people doing the work only people can do.
The part the demos skip
An assistant that can read your membership database is a data-protection question. An assistant that can write to it is an operational one. The failure mode is no longer a bad paragraph; it is three hundred wrong emails sent politely and at scale. The governance question is therefore not “which model” but “which permissions” — and that is a question IT departments already know how to answer, because it is the same one they answer for every staff account.
A pilot worth running
- Pick one workflow, read-only. “Summarise this month’s membership changes” is a fine first agent. It can see; it cannot touch.
- Use organisation accounts and named connectors. The assistant connects through credentials you issue and can revoke — never through a volunteer’s personal login.
- Keep a human on every send. Drafting and queueing is automation; sending is a decision. Keep the decision.
- Log what it did. If you cannot reconstruct what the agent read and wrote last Tuesday, you are not ready to widen its permissions.
- Review after a month. Either the time savings are real and measurable, or you stop. Both outcomes are useful.
The infrastructure question underneath
Agents are only as good as the systems they connect to. An organisation whose member data lives in four spreadsheets has nothing for an agent to act on safely. This is the quiet reason platform tidiness has become an AI-readiness issue: clean data, sensible permissions, and one source of truth were always good practice; now they are the entry ticket. The organisations getting value from agents in 2026 are, without exception, the ones that did the unglamorous platform work first.
