AI Think We're Onto Something
Founders and experts weigh in on big AI questions.
Fast Forward just wrapped its second annual Alumni Retreat, an unconference-style gathering that trades stuffy stage talks for attendee-driven content. We welcomed 115 leaders from our portfolio organizations, alongside experts from The New York Times, Bloomberg Beta, Bridgespan, and more. During a week of frenzied headlines and predictions, this communal space felt like finding solid ground. NYT journalist Kevin Roose cut right to the heart of it: "We need tech nonprofits. We need organizations that are pursuing innovation and technological excellence without the drumbeat of shareholder value in their ears at all times."
At a gathering dedicated to solving complex challenges, one topic was at the center of many passionate discussions: AI. We dove into questions keeping nonprofit builders up at night. From ethical frameworks to favorite tools, here's what we learned about building AI for humanity — straight from the people doing the work.
+ Last month, I wrote about agentic AI with Jared Chung. When Jared presented on this technology at the Retreat, jaws were almost literally on the floor. It's crucial that nonprofits understand and help shape these emerging technologies. In a recent interview with Vox, I got to share how Digital Green is already putting these tools to work in remarkable ways.
1. How do we ensure AI is used ethically and responsibly in nonprofit work?
For nonprofits, whose currency is credibility, threading the needle between technological advancement and ethical responsibility demands extraordinary vigilance. Our founders shared their tips:
Koko founder Rob Morris recommends assembling an external board of AI experts for ethical oversight. His organization partners with Compass Ethics to provide independent review on AI, research, safety, and privacy.
AI needs guardrails, and those guardrails need guards. Natalie Dunn, Director of Product at CareerVillage.org, explained their dual approach. Their system is built with guardrails to prevent harmful or misleading content, while also using an automated system to flag conversations that might indicate self-harm or other risk factors. When such risks are detected, the system immediately brings humans into the loop.
As Empower Work’s CTO Rajiv Puranik put it, "Most people want human support, not AI support — but AI can be used to help train and assist human counselors." Empower Work has found their sweet spot with a human-in-the-loop model, placing trained volunteers between users and AI.
Takeaway: The most responsible AI isn't about replacing humans, but empowering them: creating systems with robust ethical frameworks, external expertise, and critical human intervention.
2. How should organizations evaluate and select AI models?
Building foundational models requires massive resources, making them an unrealistic option for most nonprofits. The real question is how to choose and customize existing models — and whether you need AI at all.
Oliver Hurst-Hiller, CTO & Head of Product at DonorsChoose, shared a big AI learning: small teams should favor off-the-shelf tools over building custom in-house models, whenever possible, even if the off-the-shelf tool doesn't do 100% of what you need. It's most valuable to your org to more quickly validate demand, drive social impact, and maintain the agility crucial for mission-driven work. If you find product-market fit with your AI product, you can always build something bespoke later. Then, it'll be based on an even deeper understanding of your users and scenarios.
Learning Equality stressed how critical it is to establish evaluation frameworks before diving into optimization. The team, led by Jamie Alexandre, Lauren Lichtman, and Navya Akkinepally, compared the outputs of five different models through human expert review. They’re now using that feedback to develop rigorous testing benchmarks to measure model performance objectively, ensuring they can track when changes actually improve results.
And before diving into any of these decisions about buying or customizing, consider this wisdom from an attendee: "Maybe AI isn't the best solution for the given challenge you're trying to solve. That's a successful conclusion too."
Here's the bottom line: Start by evaluating whether AI is truly the right solution for your needs. If it is, explore the spectrum of existing solutions — and crucially, establish clear evaluation frameworks and benchmarks before diving into implementation. Consider customization and fine-tuning if AI is central to your work, or stick with off-the-shelf models if it's more peripheral. Rigorous testing and performance metrics will help ensure your chosen approach delivers the intended impact.
+ In ’s Technically Optimistic, Joshua Essex, CTO of Recidiviz, presents a framework for organizations to thoughtfully evaluate generative AI adoption, especially given the current rush to implement it. His piece emphasizes that while AI offers exciting possibilities, hasty implementation without proper evaluation could harm organizations and the people they serve.
3. What AI tools are organizations loving right now?
Our founders and tech leaders shared their current favorites. Open-source AI, tools that empower non-technical staff, and AI for fundraising efficiency were major themes at the retreat.
Llamafile: Want to run AI locally without internet dependency? Mozilla’s Llamafile lets you run models on your own machine instead of relying on cloud services. This keeps sensitive data private, reduces computing costs, and ensures reliability in low-connectivity areas.
Retool: This data visualization and internal tool builder empowers non-technical team members to query databases and build dashboards using AI-assisted SQL. It's particularly valuable for tracking impact metrics and automating reporting, freeing up technical teams for dev work.
Google NotebookLM: Create custom AI chatbots based on your internal documents. Train it on past materials to generate new content faster, quickly summarize reports, and provide a cost-effective alternative to expanding research capacity.
Claude: Need help with structured writing? Claude excels at organizing content to meet specific requirements. Some Alumni Retreat attendees reported saving up to 25% of their time on applications and formal documents with Claude.
Perplexity AI: This research tool efficiently finds high-quality information and cites trustworthy sources, moving beyond generic AI-generated text. It's especially useful for legal aid, policy advocacy, and research-heavy work.
Get Time Back: Google’s Get Time Back hub hosts free AI tools that can help your organization with everything from sparking creativity to boosting productivity.
Granola: This notetaking assistant transcribes meeting notes and generates summaries and action items. Users praise its accuracy in capturing key points and next steps — particularly useful for reporting and stakeholder updates.
Browser Use: This open-source framework allows you to create agents that control web browsers autonomously. Similar to Operator, it opens up new possibilities for automating complex web-based tasks and workflows.
Quick Bytes
Other Sector Stories
Philanthropy once led the AI revolution, funding the groundbreaking research that built today's technology. But as corporations pour billions into development, nonprofits and their funders must fight to stay relevant and shape AI's future. Sara Herschander’s latest in The Chronicle of Philanthropy urges the field to reclaim its vital role in ensuring AI serves the public good, not just corporate interests.
ROOST, a new nonprofit focused on protecting children online, just launched at the Paris AI Action Summit with a lineup of free, open-source AI tools. With backing from major foundations and tech leaders like the Patrick J. McGovern Foundation, the Knight Foundation, and Mozilla, this initiative could transform how we keep kids safe in digital spaces.
When fighting climate change, knowledge is power. But what happens if the information available is a deluge of unstructured data? Dr. Michal Nachmany saw this problem and decided to build a solution: Climate Policy Radar. Dive into the latest edition of Humans of AI for Humanity, a joint series written by the Patrick J. McGovern Foundation and Fast Forward, to learn more.
APN Funding News
MIT Solve opened its 2025 Global Challenges and Indigenous Communities Fellowship, calling on innovators worldwide to submit tech-driven solutions to some of the planet's most pressing and persistent problems. Prizes include The AI for Humanity Prize, open to solutions leveraging data science, AI, and/or ML to benefit humanity.
France and major funders launched Current AI, with $400M to help nonprofits and other public interest organizations access the data, tools, and computing power they need to harness AI for social good.
Public Benefit Innovation Fund, backed by the Gates Foundation, Ballmer Group, and Renaissance Philanthropy, announced a $20M AI initiative to expand access to social services.
Let’s Talk
I am living and breathing AI for humanity these days. If you are too, let’s talk!
Love how this highlights using AI to support humans, not replace them. What’s one challenge in the nonprofit space you think AI shouldn't try to solve at least not yet?
I'm surprised the list of tools didn't include Consensus. I really like it for surveying academic literature. Is that because the list came from commercial companies?