Newsletters

Welcome to our October 2026 Newsletter! First, an apology & an explanation. Normally I would have sent this Sunday night (so sorry that it’s late), but I wound up going … Read more

AI incidents unfolding–contact you representative and your senator and tell them AI safety needs legislative oversight.

I’m seeing more organizations recognize that they need a public statement about how they use AI.

This is not the same thing as an internal AI policy, and I think the distinction matters. An internal policy tells staff things like recommended tools, and where they’re not allowed to use AI. A public AI statement is for your board, your supporters, your donors, and your community. It answers a different question: not “what are we allowed to do with AI” but “what are we doing with AI, and why.”

I’ve been watching prompting advice age out faster and faster, and I think that’s actually a useful signal.

A year ago, “give AI a persona” was standard advice — something like “you are an expert California labor attorney” before asking your question. The idea was that framing the AI as a domain expert would improve accuracy. There was early evidence it helped. But Wharton’s Generative AI Lab recently tested persona prompting across current frontier models and found it doesn’t reliably improve factual accuracy anymore.

I’ve been thinking about what actually makes people successful with AI. After running dozens of workshops and talking to hundreds of nonprofit staff, I’ve noticed the same five things come up over and over.

Curiosity is the foundation. People who get the most out of AI are the ones who try things, break things, and ask “what happens if I do this?” You don’t need to be a technologist. You just need to be willing to experiment.

Updates on AI for Community We’ve been busy! May has been a full month of trainings, and we’re continuing to add more. Hot Off the Press Tip of the Month … Read more

When I have a research report that really matters to me, I run multiple “deep research” queries on all three AI services (ChatGPT, Gemini, Claude), then feed those results to Claude’s Opus model (highest-end currently), and ask Claude to combine & synthesize one final report. I ask it to identify and resolve any contradictions between the reports via additional deep research. And if, at the end, it has unresolved issues, to call those out at the end of the report as items for me to dig into.

A blog post I came across recently offered a reframe that I think is genuinely useful: instead of thinking of AI as a coworker or assistant, think of it as an exoskeleton.

The idea is that an exoskeleton doesn’t replace you, it amplifies what you can already do. Ford uses exoskeletons on assembly lines so workers can hold tools overhead for hours without the physical strain. BMW uses them so workers can carry heavy parts without injury. The exoskeleton makes the person more capable, but the person still has to know what they’re doing.

[The] report identified a few key reasons for why this was happening, including task expansion and more multi-tasking. Both of which I’ve definitely experienced over the past year. I can do so much more when I’m leveraging AI, but it also feels like I’ve got 20 spinning plates, and I have to keep them all going or I’m somehow failing.

Something I talk about in training is how useful it can be to have free-form text fields in survey forms. It’s now possible to have AI do a reasonable job of analyzing this text and extracting insights, whereas in the past nobody wanted to slog through reading a bunch of forms – which is why everything was all yes/no or rate on a scale of 1 to 5. In a similar manner, it’s now possible to have AI extract useful insights from archival data