The skeptic's welcome

THE PLAYBOOK: The skeptic's audit

Twenty minutes, one page, and you'll never evaluate an AI tool naively again.

You're right to be cautious. Data centers are real. Job displacement is real. The people you serve deserve better than being someone's training data. So before your organization touches any AI tool — ours included — run it through these three questions. Write the answers down. One page.

1. Where does our data go? Ask the vendor, in writing: what data leaves our organization, where is it stored, and is it used to train their models? If they can't answer clearly, that's your answer. (Our own standard, for the record: inbound-only, always disclosed, and your donor database stays yours. Hold us to it too.)

2. Whose job changes — and how? Name the person. If the tool drafts grant reports, does your grant writer become an editor, or become redundant? "Augmentation" isn't a slogan; it's a staffing plan. Write the plan before you buy the tool. The organizations that do this well don't cut staff — they finally let their best people do the work only they can do.

3. Who could this harm, and who decides? Run the tool past the people it affects — including the community you serve. If a chatbot answers your clients at midnight, your clients get a vote. Fifteen minutes of asking beats a year of apologizing.

That's the audit. Three questions, one page, twenty minutes. Tape it inside the cabinet where you keep the passwords. Any vendor who flinches at these questions — including us, if we ever do — doesn't deserve your mission.

Time saved: every bad AI purchase you now won't make. Value: your community's trust, which is the whole budget.

FROM THE FRONTIER: The industry just cut the price of sustained AI work

Last week, Anthropic released an updated AI model, Claude Fable 5.1. The model-name hype is noise — here's the signal, in plain English.

The part that matters isn't a smarter chatbot. It's the price cut on long, multi-step work. Today's useful AI isn't a thing that answers one question — it's a worker you hand a task to: check the donor records, draft the follow-ups, file the report, flag the exceptions. In the industry they call these "agents" — think junior staffer, not search box.

Running that kind of sustained work used to be the expensive part. With this release, Anthropic cut the cost of the memory an agent re-reads while working (the industry term is "cache reads") by roughly three-quarters — and Anthropic's own math says that makes agent-style workloads up to about 45% cheaper to run. Treat that as vendor arithmetic, not gospel — but the direction is real and it's industry-wide: sustained AI labor is getting cheaper, fast.

The second signal: governance is arriving alongside the capability. Anthropic paired the release with enterprise safeguards meant to let an organization deploy these models while keeping its monitoring data inside infrastructure it controls. That's the unglamorous plumbing that makes an after-hours caller-intake agent or a grant-deadline tracker something a small nonprofit can actually run and audit.

Our judgment: this is the trend behind everything this newsletter will teach you — AI moving from answering machine to junior staffer, at a cost and with controls a nonprofit can handle. The skeptic's fair question is "can we afford it, and can we control it?" — and this week the honest answer got better on both counts. What to safely ignore: the benchmark leaderboards and the model number itself. Fable 5.1 is an incremental update; nothing about it changes your to-do list this month. The trend is the story, not the version.

And cheaper doesn't mean automatically good — which is exactly why this week's playbook comes first.

FREE FUEL: Google Ad Grants — $10,000/month in free search ads

Google gives qualifying nonprofits up to $10,000 per month in free search advertising on Google.com — real ads on real Google searches. Someone in your city searches "volunteer opportunities near me" or "food pantry open today," and your organization can be the first result. Free.

• Benefit: up to $10k/month in-kind ad spend (a ~$329/day use-it-or-lose-it allowance). Search ads only — no display or video. And be honest with yourself up front: most small nonprofits don't spend the full amount at first, and that's fine. A few thousand dollars a month of free, targeted visibility is still a gift.
• Eligibility: registered charitable organizations in eligible countries, verified by Google's validation partner Goodstack. Government entities, hospitals and healthcare organizations, and schools and universities are excluded (though their charitable or philanthropic arms may qualify). Your site needs to be on your own domain with real content about your mission.
• Application: join Google for Nonprofits, then activate the Ad Grants product — there's a pre-qualification process. Budget real time for this; approval involves setup steps (analytics, conversion tracking) and a compliant website.
• The catch to know before you start: Google requires the account to be actively managed and kept within program policies — including click-through-rate minimums and mission-alignment rules for keywords and ad copy. An unmanaged account gets suspended. Plan for someone (staff or a volunteer) to tend it monthly, or don't apply yet.
• Deadline: rolling. But the $10k/month clock only starts compounding once you're approved — sooner is better.

You're reading Sawdust Silicon, a program of E5 Enclave Inc — one tested AI playbook a week for the people running the mission. Free forever. If this was useful, forward it to one nonprofit operator who is AI-skeptical. They're exactly who it's for.