Last semester, a colleague of mine at the CUNY Graduate Center started showing up to conversations with a lot of energy about AI. Stephen Zweibel, Digital Scholarship Librarian, had been making the case that librarians needed to get serious about it—not in the “here’s how to use ChatGPT for your reference questions” way, but in the “we should actually be building things” way. It’s a different argument than the ones that tend to dominate at a large university, where the AI conversation mostly revolves around faculty research support, plagiarism policy, and enrollment tools. Workflow automation, in-house tool development, reducing vendor dependency—that stuff doesn’t always have a natural home. My boss and I started thinking about how to channel Stephen’s enthusiasm into something structured. A peer mentoring cohort seemed like the right fit.
We’re three weeks into a 16-week program exploring agentic AI and its implications for library work, and I’ve been trying to write something about it before too much time passes and I lose the early impressions.
The cohort builds on something we’ve been doing for a few years through our Alma Extensibility Task Force, a group of CUNY librarians who use Alma’s REST APIs to extend functionality based on systemwide needs. That work has produced small but meaningful things: automations, integrations, configuration improvements that Alma doesn’t natively support. The idea here is to expand that model beyond people who already identify as programmers, and beyond Alma itself.
Thirteen library faculty and staff from nine CUNY campuses are spending the semester building AI-enabled tools to address real workflow challenges. Stephen is leading the cohort, guiding participants through development while keeping a consistent focus on critical evaluation: understanding APIs, interrogating model limitations, assessing risk, and figuring out what’s actually technically feasible versus what’s well-marketed.
The projects reflect the kinds of problems libraries are genuinely grappling with: automating faculty publication tracking for institutional repositories, building accessibility checkers for course content, enhancing discovery systems, streamlining course reserves workflows, developing CUNY-wide tools for data access. People’s technical backgrounds range from complete beginners to experienced developers, which makes the dynamic interesting.
A few of the projects I’ve been wanting to pursue for a while—including externalizing Alma letters into version-controlled infrastructure and standardizing access model descriptions across institution zones—are taking shape within this cohort. These were ideas that had been sitting on a back burner for longer than I’d like to admit. With structured time and tools like Claude Code, they’ve started moving.
My role is mostly scaffolding: structuring the cohort, handling logistics, setting up the shared collaboration space in Microsoft Teams, trying to create conditions where people can experiment without feeling overwhelmed. The Teams space is already more active than I expected: participants troubleshooting installation issues together, sharing resources, asking questions about APIs and workflow automation. Several people are juggling a lot right now, professionally and personally, and they’re still showing up.
We’re only three weeks in, but it’s already interesting to watch.