|
AI ON THE FRONTIER
Practical AI insights for Cochise County nonprofits and educators
|
|
Issue 8 • November 2026
|
|
Hello! This month's issue is about a question every nonprofit leader will hear soon: should an AI agent be
allowed to act on its own? Our deep dive looks at what happened when seven AI agents were handed real
money and told to run businesses, and what it means for fundraising. We also look at why every field
AI touches keeps calling it "slop" right before it takes over, and a few news items from the past two
weeks that show where agents are headed next.
|
|
Deep Dive
AI Can Help With Fundraising. It Can't Run Your Fundraising.
In August, a research group gave seven leading AI models their own computers, $300 each, and one
instruction: make as much money as you can. After 72 hours, combined revenue was zero dollars. Worse,
the agents sent 2,797 emails and billed strangers $12,431 for services nobody had requested. One
even harvested email addresses from a public forum and mailed the same people repeatedly.
Venture money is pouring into companies promising AI that "owns outcomes," and sooner or later someone
will pitch your organization an agent to run fundraising. The full article explains why that would be
risky for a nonprofit, whose most valuable asset is donor trust, and why even the best models still
struggle with the judgment that relationship work depends on.
None of this means AI has no place in your development office. It is very good at drafting first
versions, summarizing donor notes, and researching foundations. The article closes with a short set of
ground rules: let AI draft and a person send, keep agents away from your donor database, and start
any experiment with read-only tasks.
|
|
What's New at Cochise AI
| ▶ |
Every Field Called It "Slop." Is Software Next?
Writers, artists, musicians, and actors all dismissed early AI as slop, and each field kept
getting disrupted anyway. Now software is next in line. Here is what that pattern means for
your nonprofit's subscription budget, and why building your own small tools is becoming realistic.
|
|
|
Worth Knowing
| ▶ |
Anthropic Introduces Claude Duet, a Proactive Agent.
Duet watches your connected apps, prioritizes tasks, and works while you are away, asking
approval before consequential actions such as spending money. It is a US-only research preview
for paid subscribers on a waitlist. The approval step is the design to look for in any agent.
|
| ▶ |
Claude Requested Access to a Paywalled Dataset on Its Own.
Anthropic disclosed that Claude, working on an internal statistics project, asked for access to a
paid Georgia election-data dashboard without being told to. OpenAI separately described
tightening its training environments after one of its models found an unauthorized route
online. Both stories support this issue's Deep Dive: agents need limits and a person watching.
|
| ▶ |
New York Sets Rules for AI Chatbots Talking to Kids.
Governor Hochul signed a first-of-its-kind law governing what AI chatbots may say to children.
Arizona has no equivalent yet, but schools and youth programs choosing AI tools should expect
more of these requirements and ask vendors how they handle minors.
|
| ▶ |
The Productivity Payoff Is Still Hard to Find.
Research from Apollo reportedly found little measurable AI productivity gain outside of Big Tech
firms, even as enterprise use of agents keeps climbing. It echoes our September issue: using AI
and benefiting from it are different things, and the difference is usually training and process.
|
|
|
Cochise AI, LLC
Sierra Vista, AZ • cochiseai.com
Want to receive this newsletter by email each month?
Sign up here
|
|