But that also makes it dangerous.
This is the future of AI. We need to understand the different options well to find our own balance between security and productivity. We must determine which specific tools and tasks provide a security-to-results ratio that is worth it.
OpenClaw is the future (and it's scary, too).
It's a free, open-source platform for building your own AI agents and automating tasks.
It's easy to use and highly flexible. You can add hundreds of "skills" created by other users and companies.
So we could say: AI models are still super dumb and can't be trusted, let's do everything with humans.
But they also tested 10,000 humans. 28,5% failed.
Some AI models/tools are more reliable than humans for some tasks. But we have to test them, not decide based on our gut.
We talk a lot about AI risks and mistakes, but are humans really better?
A company recently ran a test with a simple reasoning prompt: "I want to wash my car. The car wash is 50 meters away. Should I walk or drive?"
42 out of 53 AI models failed (said "walk").
Other roles will follow the same pattern soon.
AI agents will probably be the most important tool for our work in a few years (maybe months for some tasks).
Source: https://metr.org/blog/2026-02-24-uplift-update/
The impact of AI agents in coding is so big that researchers are having trouble recruiting coders that want to stop using AI for some of their work.
Even paying them an additional $50 for each hour without AI!
There is clearly a big demand for "AI Personal Assistants" that help you with all kinds of computer tasks. Thousands of people bought a Mac Mini just to try OpenClaw, an experimental and risky system. It's probably going to be one of the big trends in 2026-2027.
You have to find your balance between productivity and security. OpenClaw is free and open hundreds of possible use uses, but can be also quite risky. Commercial alternatives like Claude Cowork are not free and a bit more limited, but also more secure (not 100% though).
You need to try new tools to find new opportunities, but never with sensitive data. Minimize the files and features you give to each AI tool or agent, don't just give access to everything and expect the agent to be "responsible".
You need to understand what each AI system can do and define limits using tool settings or code. Just prompting "do this" or "don't do this" is not a reliable security measure.
AI Agents are powerful but risky. Even for AI security experts.
Recently a Meta AI security researcher started using a trendy agent system (OpenClaw) on her inbox and the agent erased thousands of emails by mistake.
A few takeways:
What are your favourite AI quickwins for nonprofits?
Please share your favourite ideas. Let's make a big list! ๐
And this isn't just for marketing.
Almost every role in a nonprofit has content that can be recycled or transformed: training materials, grants, job offers, meeting transcriptions, etc.
AI doesn't need inspiration. It needs input.
The best way to use AI is not "write me a post about X"
It's: "here is what I already have, now turn it into something new"
Repurposing content is faster and more reliable, Because the raw material is already there: your voice, your data, your story. You're just reshaping it.
Here are some variables you should provide to get a strategy that actually works:
โข Donor Data
โข Tech Stack
โข Budget Reality
โข Brand Voice
When you add constraints, AI gets creative.
When you add context, AI gets accurate.
Stop getting generic advice from AI
"Write a fundraising plan for a nonprofit."
If you prompt like that, you deserve the generic fluff you get in return.
I decided to put my money where my mouth is: I just launched a Google Ad Grants management service at 99โฌ/month (5x less than many consultants).
Thanks to AI, I can now help nonprofits that could not afford a consultant and were wasting their $120,000/year grant.
AI should be making many services cheaper.
If a task takes 80% less time to do with AI, why does the service still cost the same?
Some providers can use that extra productivity to improve their services. But customers should get to choose between better service or lower prices.
If you can't answer that, delete the question.
Paste your draft survey into ChatGPT and ask:
"Review these questions. Flag any that seem actionable or vague. Identify potential bias. Tell me which questions I should cut to reduce respondent fatigue."
Stop collecting "Zombie Data" at your nonprofit ๐งโ๏ธ
We often ask questions because they are "nice to know," not because they drive decisions.
Before adding a question to your next survey, ask yourself:
"If the answer is X, what do we do? If it's Y, what changes?"
Once you build the system, the water flows exactly where it needs to go, instantly and automatically, without you lifting a finger.
Don't just use AI tools. Connect them. Build automated AI workflows using platforms like Zapier, n8n, Lindy or even Claude Code.
Stop using ChatGPT like a bucket. Start building plumbing.
Typing a prompt into ChatGPT is like fetching water with a bucket. You have to walk back and forth every single time.
AI Automation is different. It's about installing pipes.
You want to innovate. But your nonprofit has a risk-averse board.
Here's how AI helps you:
Instead of bringing one big scary idea to the board, use AI to generate 10 smaller experiments. Pick the lowest-risk ones.
Test them. Bring data to your board.
The 3-tier framework for safer nonprofit innovation (use also in your AI prompts):
๐ข Low Risk, High Learning
Experiments you can try next week
๐ก Medium Investment
Ideas that require planning but are worth exploring
๐ด Big Bets
Transformational ideas that need significant commitment
But here's the thing: Your policy needs to balance two competing forces:
A) Too restrictive? Your team innovates in the shadows (or not at all).
B) Too loose? You risk donor trust and mission integrity.
Most nonprofits know they need an AI Policy. Few know how to create a really useful one.
Here's what your policy needs to answer (in plain language):
โข What AI tools can we use?
โข What data can we feed into them?
โข When do we need human review?
โข How do we disclose AI use?
I recently built an AI tool (using Gemini Gems) that analyzes every draft against an inclusivity framework.
It grades findings by severity and explains why a phrase might land poorly.
It's not perfect, but it's a second set of eyes that never gets tired.
๐จ๐๐ฒ ๐๐ ๐๐ผ ๐ฟ๐ฒ๐ฑ๐๐ฐ๐ฒ ๐ฏ๐ถ๐ฎ๐ (๐ป๐ผ๐ ๐ถ๐ป๐ฐ๐ฟ๐ฒ๐ฎ๐๐ฒ ๐ถ๐)
There is a lot of talk about bias in AI tools (which exist for sure, since they are trained on biased content created by people), but no so much on how AI can be used to detect & correct our own bias.
AI Superpowers for your nonprofit.
Discover the best AI tactics to multiply your impact.
nonprofit.top
The real unlock: uploading your own terminology glossary.
Now every translation uses YOUR preferred terms.
Suddenly translations go from 2-hour reviews with 50 manual changes to ready-to-publish in a less than 10 minutes.
