AI for Good at Mimecast
On October 1, I volunteered at AI for Good, a Mimecast skills-based volunteering event at our Lexington, MA office. Employees coached nonprofit staff on using Claude, and I worked with a youth mentoring nonprofit.
Most nonprofits already use AI, but few get much value from it. Typically, the barrier isn’t access to these tools. It’s training, time, and confidence. That made clear how we’d tackle our time together: help the staff use AI on work they already do, so more of their time goes to the mission.
A repeating event
Every quarter, the staff plan an event for their mentors and mentees. It comes around four times a year, so the planning ought to be repeatable too.
A Claude project they can reuse
I helped them build a Claude project, explaining what it is, what it does, and how to give it context. If you haven’t used one, it’s a saved workspace: you add instructions and reference files once, and every new chat starts with them. We loaded this one with examples from past events and the nonprofit’s own guidelines.
We broke the instructions down into a simple framework: role, voice, context, and format. Once we nailed the context and format, we wrote a problem statement and started iterating. The staff tested the project as we went, tightening the instructions each round. By the end, it produced a polished first draft of the next event’s show flow, in the same format they’ve used before.
Now, for each quarter’s event, the staff can start a new chat in the project. Each chat starts with the past events and guidelines already in hand, so planning is faster and the quality holds steady from one quarter to the next.
Why a project and not a prompt
A good answer from a single chat is easy to lose. A project keeps the examples and guidelines in one place, so the result doesn’t depend on who’s planning that quarter. It’s the same idea as a good runbook in support.
It also gets better with use. After each event, the staff can add that plan and what they learned back into the project, so the next quarter starts from a stronger place. And if someone new takes over the planning, they don’t have to start from scratch. The context is already there when they open it.
Coaching, one on one
We talked through use cases, which models to pick, what effort means, and how to give Claude good context, so the staff can bring it into their everyday work, not just this one event.
Cleaning up context first
Good context often lives in files full of names and email addresses, and for a youth mentoring nonprofit, that matters. We talked about what shouldn’t go into Claude, then made a quick mockup: a single HTML file that runs entirely on their computer, so nothing leaves the machine. You paste in text, and it redacts names, emails, and other values. That way, the staff can clean up files and text before using them as context for Claude later.
What surprised me
What surprised me most was how much potential these organizations have with AI that they don’t use yet. Once we started looking, there were so many things they could automate. They didn’t even know they already had access to other frontier models, so we talked about how to translate this work there. The role, voice, context, and format framework carries over, whichever tool they open.
We only had time to flesh out one idea fully, but we built the foundation for them to succeed with the others.