There’s so much happening broadly in public and political discourse right now that it’s easy to forget that we are living in a watershed moment in the course of humanity’s technological progression. ChatGPT was first released to the public in late November, 2022, and with it the capabilities of Artificial Intelligence (AI) suddenly became accessible to a wide swathe of humans for the first time.
In early 2023, I remember speaking with a long-time friend and a personal hero in the emerging tech space, Dave Guarino, about use cases for this new technology (Dave’s passion is improving our country’s safety net programs). Apart from wisely urging me to engage with AI as soon as possible (”just jump in and experiment!”) he also assured me that we were experiencing a significant milestone in human achievement that he considered analogous to the automobile or television or personal computer. AI emergence was that kind of leap that would be defining as humanity gazed back over its shoulder in the eras to come.
I was really intrigued and excited, and I remember encouraging others in the nonprofit sector to also follow Dave’s suggestion. Somewhat predictably, however, nonprofits were hesitant to embrace this new technology for reasons I cannot help but empathize with: the environmental impact, the implicit biases in training data, and concerns about plagiarized intellectual property.
Fast forward to early 2026, and the landscape of AI is evolving at a breathtaking pace. About half of US adults report using AI regularly for either work or personal tasks, and roughly 80% of global companies have integrated AI into their operations. Close to 2 billion people worldwide are now said to use generative AI on a recurring basis.
But here’s what most people outside of technical circles don’t realize: experts are no longer talking about AI as a helpful tool with a wide array of applications, but instead they are beginning to insist that we are in the infant stage of Artificial General Intelligence (AGI) where AI agents are proficient experts across many domains and can support detailed analysis or critical problem solving. The conversation among engineers and AI researchers has shifted dramatically in the past few months (more on this below). This is human history altering stuff, and the people building and experimenting with it are largely talking only to each other.
And yet, I have found that despite increasing mainstream adoption of AI like ChatGPT and Claude, many nonprofits are still resistant to becoming AI-integrated organizations, and feel paralysis about where to begin. This worries me deeply—not just because they’re missing efficiency gains, but because the rules governing how AI decisions get made, how resources get allocated, and how people collaborate with AI etc. are being written right now. If organizations that are so adept at advocating for historically under-resourced communities are sitting out these conversations, their perspective will be largely absent at a moment when it matters most.
I again want to iterate that I do understand the instinct from the nonprofit sector to be wary, but I would implore us to overcome that inertia and “just jump in and experiment,” because the compounding cost of inaction will prove to be far greater than most leaders currently realize.
The Landscape We’re In
AI development is obviously dominated by the tech industry. I don’t believe this is inherently negative, as we have seen authentic demonstrations by industry leaders and professionals with an earnest desire to advance humanity and create a path toward “AI for good.” But we have to be honest about the structural reality of AI’s trajectory: market-driven enterprises don’t always easily find the intersection between societal benefit and profitability. When these priorities are at odds, the market almost never loses.
Meanwhile, government is woefully out of step. Regulatory frameworks are being left in the dust by the pace of development, and people attempting to write laws about AI often don’t understand the basics of how it works. By the time legislation passes, the technology is already a thousand miles ahead. As Hyperdimensional recently wrote, “It is quite possible [the AI] policy debate America will have in 2026 is already antiquated.”
This matters because the values being embedded into AI systems right now, like what gets measured, who gets prioritized, what “success” of AI means, are calcifying into the downstream defaults. They’ll become the invisible architecture of how services get delivered, how access gets determined, how need gets assessed. Once those defaults are set, this trillion dollar ship will prove too large to steer.
Why These Organizations, Specifically
Nonprofits, government agencies, and community institutions have always occupied a unique position in our society. They have proximity to affected communities that tech companies generally don’t. Historically, they’ve served as the infrastructure of care—filling gaps, advocating for the overlooked, and holding systems accountable when they harm the people they’re supposed to serve.
The communities most likely to be impacted by AI-driven decision-making such as people navigating public benefits, seeking social services, and facing algorithmic assessments are the same communities these organizations exist to serve. A community health nonprofit understands barriers to access that won’t show up in any training dataset. A state agency administering SNAP benefits has insight into client behavior that no Silicon Valley product team possesses. But if mission-driven organizations fail to bring these perspectives, they may not be considered at all.
What’s Actually Happening Right Now
I am fascinated (and worried) about what isn’t making it into mainstream discourse: AI is crossing the threshold that changes everything.
Engineers and researchers are building and deploying AI agents, better described as AI frameworks or “harnesses” that don’t just respond to prompts, but have demonstrated that they can operate autonomously for extended periods with complex tasks. These agents can be given a goal, access to tools, and boundaries to work within, and they’ll execute multi-step processes, self-evaluate and correct when they hit obstacles, and even coordinate with other AI agents to accomplish complex work.
For example, a day after I had Claude Code running on my machine, I connected it to my Google Drive Desktop and as a test I asked it to review and create a plan for reorganizing thousands of backup files from one of my old laptops that I had no hope of getting around to on my own. It created a plan for subfolder organization including identifying personal documents to put into a separate drive folder, renaming vague/cryptic file names, and deleting duplicate files. I had known deep down that I would never have time to do this manually, but that someday an AI agent would be able to do it for me. That day is here— the entire process took about 35 minutes of background run time while I periodically checked terminal to read through outputs and approve next steps.
However, the most interesting use case I’ve read recently was when Dan Shipper shared that he “[downloaded] meeting recordings, put them in a folder, and asked Claude Code to tell [him] all of the times [he] subtly avoided conflict”. This is like being able to bring your executive coach into meetings with you to quietly observe and provide feedback— true AI agent stuff. The shift isn’t from “AI as tool” to “better AI as tool.” It’s from “AI as a tool” to “AI as autonomous system performing multiple congruous, complex tasks at a professional level.”
It matters for nonprofits and government agencies because the gap between what insiders know is possible and what everyone else believes is growing exponentially. The technical community is already planning for a world where AI systems handle significant portions of knowledge work autonomously. They’re debating governance frameworks, safety protocols, and economic implications. These conversations will shape how AI gets deployed in social services, benefits administration, healthcare, education… literally every domain that mission-driven organizations care about.
I am grateful that Claude is attempting to ease entry into high-powered AI agents with the recent introduction of its “Claude Cowork” tool, which is basically Claude Code lite without some of key features or flexibility of Claude Code that make it so powerful. But the problem persists that it’s not that nonprofits are using AI incorrectly, but that they’re avoiding it entirely. A lower wall is still a wall, and it feels insurmountable if you’re afraid of seeing what’s on the other side.
The Concerns Keeping Organizations Out
I don’t want to belabor the issue because experts like Andy Masley have written great pieces debunking these concerns, but the hesitation for nonprofits to adopt AI usually clusters around four concerns: environmental impact of “boiling the ocean” for generative AI, negative bias in outputs, intellectual property theft, and a general paralysis about not knowing where to start. All have merit, but there is context and nuance nonprofits should consider:
Environmental Impact: Estimates show that daily AI use is probably less harmful than a workday’s worth of video conferencing, or watching a few hours of streaming entertainment at home. I tell nonprofits that their daily AI environmental impact is going to be less than grabbing their favorite drink at a coffee shop, and it’s not even close. This is not to say AI has no environmental impact, but it’s not the catastrophic drain that justifies complete disengagement. There’s a cost to disengagement too: ceding development of these tools to actors who may care even less about sustainability than you do. Ultimately, I believe AI will be a crucial tool in helping humanity solve the climate crisis. Abandoning it entirely for fear of environmental damage is penny wise, pound foolish.
Bias in Outputs: There is valid concern that AI is inherently racist, sexist, ableist etc. and that bias is baked into the technology itself. The more accurate framing is that AI reflects the data it’s trained on, which reflects broader human sentiments over the course of history including our missteps and shortcomings around inclusivity. AI doesn’t create bias; it surfaces and sometimes amplifies biases that already exist in the systems we’ve built. It’s an active area of focus in AI ethics, with researchers, advocates, and some tech companies working on detection and mitigation. This is exactly why nonprofits and the government need to be a part of these conversations. The more voices shaping what “fair” looks like, the better the outcomes.
Plagiarism and Intellectual Property: Others are concerned that generative AI is built on creative work without consent, and that using it makes one complicit in that extraction. There are real questions about how training data are sourced, and tech companies haven’t helped their own narrative here. The legal landscape remains unsettled, and the lines around intellectual property are being drawn right now, which is another argument for engagement, not avoidance. If you want those lines drawn in ways that protect creators and respect ethics, you need to be part of the conversation.
Paralysis (Not Knowing Where to Start): The technology feels vast and technical. Headlines swing between utopian promises and apocalyptic warnings. Leaders look at the landscape and freeze. The truth is, AI is more accessible than it has ever been, and you don’t need a data science team or an understanding of how large language models work under the hood. All you really need is a basic machine, and curiosity & willingness to try. Try using AI to help draft a tricky email to a difficult funder. Ask it to explain a confusing policy update in plain language. Have it pressure-test your strategic plan or identify blind spots in your theory of change. Upload a grant report and ask it to analyze where the gaps are. Prep for a hard conversation with your board chair. Plan a staff retreat when you’re out of ideas. These aren’t transformational projects. They’re Tuesday afternoon tasks, and they’re how you familiarize yourself for the more complicated stuff to come.
What Engagement Actually Looks Like
I’m not suggesting every nonprofit or state agency needs to become an AI-native entity this month, but I do believe treating AI literacy and adoption should be regarded as capacity building the same way you’d approach financial management or communications training.
When I recently spoke at the OpenAI Nonprofit Jam, my specific example was language justice, or using AI to break down barriers for communities navigating services in languages other than English. It’s easy, practical, mission-aligned, and immediately useful.
Beyond internal adoption, organizations can sit in policy conversations, bringing community perspective to governance discussions. They can partner with researchers working on AI ethics. They can share what they’re learning with peer organizations. None of this requires massive budgets or technical staff.
The Wave Is Coming Either Way
There are real risks, real ethical questions, and real reasons for caution. You don’t have to like all of AI’s offerings or implications. Dogmatic reverence for the technology is off-putting and ignores its genuine shortcomings.
But here’s what I need the nonprofit sector to understand: we are in the infant AGI stage right now. It is not hyperbole to say that the timeline for when this transforms every sector of the economy will henceforth be measured in years, not decades.
This is that watershed moment. Organizations that engage now will be the ones ensuring that when AI transforms how we deliver services and allocate resources, it does so in ways that reflect our values as a human society. The timing is too important for organizations that care about communities to sit idly by and hope for the best.
Written in collaboration with Claude Opus 4.5