Leading Through the Unknown
An Equity-Grounded Framework for Navigating AI
Introduction
As I sit down to write about navigating AI, I feel the same pull I imagine you do: the urge to have a position, to know, to be able to say clearly what is right and what is wrong and what everyone should go do about it. I have spent my career working alongside organizations that have always known what they needed and have rarely had the infrastructure to translate that knowledge into the language that power requires. I have built data systems for nonprofits, grassroots organizations, and government agencies navigating that gap. That proximity shapes everything about how I practice, and it is what brought me to this question with more than an academic interest in getting the answer right.
In attempting to gather my own thoughts, I spent time seeking out and reading the positionality of other leaders in this space. What I found, in gratitude, is that the leaders doing the most honest and durable work in this moment are the ones who have found a way to stay in motion without pretending to have arrived somewhere they haven't.
KEY TERMS
ADAPTIVE LEADERSHIP
A framework for navigating problems that require changes in values, beliefs, and behaviors, where no expert holds the solution.
IDEAS, ARRANGEMENTS, EFFECTS
A systems design framework developed by the Design Studio for Social Intervention for understanding how ideas become embedded in social structures and produce outcomes, evenly or not.
ALGORITHM INTERROGATION
The practice of asking whether an automated system was involved in a decision affecting your community, requesting documentation, and supporting community members in contesting outcomes.
CO-CREATION
A governance design choice that positions community voice as the architecture of decision-making, not a feature added afterward.
ETHICAL DATA GOVERNANCE
The foundation that makes responsible AI possible, built when communities define what gets measured, how, and what happens with what is learned.
I am sitting with my own curiosity about what leading through genuine uncertainty actually requires in relation to AI, and why the complexity itself might be exactly the opening we have been waiting for. AI is ever-changing, and the moment we learn one thing, something is marketed to us as even better and even faster to replace it. That uncertainty is an invitation to shape what comes next, together.
AI is not a settled question, and that is the most honest invitation this moment has extended to us.
What the Complexity Is Actually Made Of
The speed of AI's movement meant that my role as a leader in the social data space crossed over into AI territory before I had given myself or my community the honesty that crossing deserved. I am naming that here because I suspect I am not alone in it.
The harms are real and documented. Algorithms are making decisions about who gets housing, who qualifies for healthcare, and who gets screened out of a job. The U.S. Department of Justice settled a discrimination case against Meta for using ad-targeting algorithms that excluded protected groups from seeing housing listings.¹ Courts have heard cases involving algorithmic housing classifications that left people in dangerous conditions.² At least ten lawsuits have been filed against major AI companies for mental health harms, including wrongful death cases involving minors.³
The environmental cost is real. Data centers in Texas are projected to consume 49 billion gallons of water in 2025, rising to nearly 400 billion gallons by 2030.⁴ That water disappears into the atmosphere through evaporative cooling systems and does not return. The communities closest to that infrastructure are rarely the communities whose leaders are making siting decisions.
The labor cost is real. The workers who annotate training data and moderate harmful content so that AI systems can function earn approximately two dollars an hour in Kenya, Colombia, and the Philippines, while their counterparts in the U.S. earn ten times that for the same work.⁵ A 2025 survey documented 60 independent incidents of psychological harm among a group of 76 data workers.⁶ These are the people whose invisible labor makes AI feel seamless.
The policy environment changed dramatically on January 20, 2025, when the Biden administration's Executive Order on AI safety was revoked. A replacement order issued in December 2025 directs federal agencies to identify and preempt state AI laws, including laws specifically designed to protect communities from algorithmic discrimination.⁷ The protections that existed are being dismantled faster than replacements are being built.
And at the same time, the benefits are also real. AI can do work that chronically under-resourced organizations desperately need done: grant writing, data analysis, intake documentation, translation. Researchers have estimated that mission-driven organizations with meaningful AI adoption could see efficiency gains equivalent to a $100,000 unrestricted grant for every million dollars of operating budget.⁸ That is not nothing for organizations running on empty.
Both things are true at the same time. The question is how we hold this, and what we do with that, and I do not think either of those things happens alone. The holding and the doing are community work. That is where co-creation begins.
I want to name one more tension before we go further. A practitioner I follow recently asked something that has stayed with me:
“Do you truly need to complete tasks that quickly, or has conditioning into capitalism internalized urgency and speed, dysregulated your nervous system, tainted work structures, and convinced you that this is truly how it is supposed to be?” (IG @jessicaddickson)
It is a fair question. For organizations that have always had to do more with less, the efficiency argument for AI can feel like relief. It can also be another way of accepting under-resourcing as permanent rather than naming it as a structural problem that deserves a structural response. The efficiency frame deserves interrogation before it becomes the whole argument.
What would it mean to let the questions of AI breathe, to sit with them in community before reaching for an answer?
What Adaptive Leadership Actually Asks of Us Here
Ronald Heifetz's adaptive leadership framework makes a distinction that I find useful here.⁹ Technical problems have known solutions. You can bring in an expert, apply the solution, and the problem gets solved. Adaptive problems are different. They require people to change their values, beliefs, and behaviors. There is no expert who can do that work for you. That work belongs to each of us, and it looks less like a checklist than it does like a practice: staying curious, staying current, resisting the pull toward rigidity, building and rebuilding your values alongside your community, and letting the people closest to you hold you accountable to the things you say you believe.
AI adoption across mission-driven organizations and public agencies is being treated as a technical problem. Find the right tool. Write the right policy. Get the training. Underneath those technical questions are adaptive ones that we are not asking loudly enough. What I find goes hand in hand with where I have landed is the framework of ideas, arrangements, and effects. As opposed to prescribed steps, this framework grounds us in a practice that carries the complexity of the environment we are actually in, instead of reducing us to black-and-white thinking.
Ideas, Arrangements, Effects is a systems design framework developed by the Design Studio for Social Intervention.¹⁰ Their premise is straightforward: ideas are embedded in social arrangements, and those arrangements produce effects. Think about something as simple as chairs in a classroom. Rows of chairs facing forward carry an idea about where knowledge lives and who holds it. A circle carries a different idea entirely. The arrangement itself produces a different experience, a different set of possibilities, a different set of effects, and it is never neutral.
What I find most useful about this framework for this moment is that it resists the human tendency to jump from an unjust effect directly to blaming a person. We tend to ask who is at fault when something goes wrong, rather than asking what arrangements made that outcome possible in the first place. When we apply that to AI, it becomes a genuinely different kind of question. The harm sits less in the intentions of individual developers or the carelessness of individual organizations, and more in the ideas embedded in the systems, the arrangements that give those systems their power, and the effects that land, unevenly, on the people with the least ability to contest them.
This framework gives us somewhere to look that is bigger than the tool and smaller than the abstraction.
Ideas: The Assumptions We Are Carrying
What beliefs are guiding our decisions about AI, and do those beliefs hold up under scrutiny?
The belief that faster is better. The assumption that efficiency is always the goal. The idea that being behind on AI means falling behind on mission. The notion that under-resourcing is a condition to optimize around rather than a structural inequity to name and resist. These are ideas worth examining before they make decisions for us.
There is also a harder one: the assumption that AI adoption is primarily a technical and ethical question for your organization to sort out internally. That assumption undersells the stakes. AI adoption is a political and structural question about who gets to set the terms of a technology that is already shaping the lives of the people your organization serves.
Arrangements: The Structures That Need to Shift
Who in your organization is currently making AI decisions? Is it one person, or a small group with technical comfort? What does that concentration of judgment mean for accountability? Who is not in that room, and more importantly, whose political and structural reality is absent from those conversations?
Research suggests that roughly 41 percent of nonprofits rely on a single staff member for all AI decisions, a single point of failure rather than a governance structure, and it almost certainly means that community voice is not upstream of those decisions.¹¹
What would it take for community stakeholders to have input before your organization adopts a new AI tool, not after? What existing governance structures would need to change, and what new ones would need to be created? These are design questions, and the answers are going to look different for every organization.
Effects: Who Bears the Loss
This is the question this framework invites us to ask before any decision is made: when things go wrong, who pays the cost? Rarely the executive director, the funder, or the AI company. Almost always the community member whose application was screened out, whose benefits were cut, whose face was misidentified, whose data was used to train a system they never consented to, whose neighborhood absorbed the environmental cost of infrastructure sited without their input.
Naming who bears the loss is accountability, and it is the prerequisite to making decisions that are genuinely aligned with mission rather than just with convenience.
When we ask who bears the loss before a decision is made, we change the outcome and the nature of the conversation itself.
Why Opting Out or Holding Out Is Not Neutral
I hear this often: we have decided not to engage with AI right now. We are waiting until it is clearer. We do not want to cause harm.
I understand that instinct. And I want to push on it.
The communities closest to your organization are already subject to algorithmic systems. Resume screening software decides who gets an interview. Algorithmic tools determine benefits eligibility. Predictive systems shape how law enforcement is deployed in neighborhoods. Credit scoring, insurance pricing, healthcare resource allocation. These systems are operating whether your organization uses AI or not. Opting out removes your voice from the room where these systems are being built, contested, and sometimes dismantled, without protecting the people you serve from them.
Opting out is also a racialized decision in ways that are worth naming directly. The communities most harmed by biased AI are overwhelmingly Black, Indigenous, and Brown communities. They are the communities least represented in the rooms where AI policy gets made. And they are the communities that practitioners close to this work have the most ability to bring into those rooms, if we choose to show up.
This is the structural argument that grounds my practice: AI is not neutral, and neither is opting out. These systems already shape the lives of the communities you serve. The question is whether the organizations closest to those communities will have practitioners in their corner who understand both the harm and the tools.
THE NONPROFIT DECIDES
"We are not ready to engage with AI right now."
MEANWHILE, THE CITY ADOPTS AN ALGORITHM
Their clients start receiving benefit denials at higher rates.
NO STANDING TO CONTEST
They were not in the room. Opting out removed their voice while their exposure remained.
AI is already shaping decisions in your community, with or without your consent. The open question is what integrity looks like in that reality, and who gets to define it.
Co-Creation as the Leadership Move
Here is what I actually find hopeful about this moment: the field is genuinely unsettled. The frameworks are still being written. The governance structures are still being designed. The policy fights are still being fought.
That is the reason our collective voices will always be consequential.
For organizations that center community, the constant change of AI becomes the condition under which co-creation turns from a value into a practical necessity. The communities you serve have as much standing to shape how AI develops as any vendor, any foundation, or any tech company. And right now, most of that standing is going unused.
Co-creation in this context is a governance design choice. Who is at the table when your organization decides what AI to use and for what? Whose experience shapes how you evaluate whether it is working? When something goes wrong, who has the standing to say so, and what happens when they do?
Community voice has to be the architecture, present from the start of the decision process. This is especially true for organizations that are simultaneously adopting AI tools and building or refining programs designed to serve community. Those two tasks run concurrently. The program design shapes what data gets collected. The data shapes what AI can do. The AI shapes what the program can become. If community voice is not present at every stage of that cycle, the resulting system will reflect the priorities of whoever was in the room. Ethical AI governance grows out of ethical data governance, the foundation that makes responsible AI possible in the first place, rather than sitting as a layer added on top of AI adoption. When communities are part of defining what gets measured, how it gets measured, and what happens with what is learned, the entire system changes.
What becomes possible when the people closest to the problem are the ones designing the response?
What This Looks Like in Practice
These are the practices I am implementing in my own work, continuing to learn from my community about how and why they need to show up, and remaining curious about what belongs here that I have not yet named.
ASK BEFORE YOU ADOPT
Picture this: it is late, the deadline is tomorrow, and you are already stretched thin. You open an AI tool and ask it to help you draft a one-pager for a program your community has been building for months. It feels like a lifeline, and in that moment, it might be. But that one-pager will carry assumptions your community never agreed to. The model was trained on data that skews toward dominant narratives about who needs help and why. It may describe your community's circumstances in language that centers deficiency rather than resilience, that frames poverty as personal failure rather than structural outcome, that uses words like “at-risk” or “underserved” in ways your community has explicitly rejected. The document goes out. A funder reads it. The framing sticks. And the people your program was built to serve had no say in how they were described to the people with power over their resources. The harm is rarely visible in the moment. It accumulates in the language that gets used, the priorities that get centered, and the voices that do not make it into the room. Asking who had input before you adopt means building the habit of noticing when a tool is making decisions that belong to your community.
INTERROGATE ALGORITHMS YOUR COMMUNITY ENCOUNTERS
You do not have to have built a system to scrutinize it. If the people you serve are subject to automated decisions in benefits systems, housing applications, or criminal justice processes, it is worth asking whether an automated system was involved, requesting documentation, and supporting community members in contesting outcomes they believe are wrong. Algorithm interrogation in practice is the habit of asking questions. It does not require a data science degree.
REDIRECT THE CAPACITY AI CREATES
Human work is the anchor. The trust built over years of showing up in the same neighborhood. The conversation that shifts someone's understanding of what they need. The presence in a room that cannot be replicated by a summary or a generated report. When AI creates capacity, the question worth sitting with is whether that capacity flows back toward the work that only humans can do, or whether it gets absorbed into organizational throughput and disappears. Redirecting what AI gives back is a practice of remembering what this work is actually for.
Human work is the anchor. Everything else is in service of it.
Boards, funders, and staff deserve honest conversations about what AI decisions actually involve: who benefits, who might be harmed, what remains uncertain, and what accountability looks like if something goes wrong. That honesty is harder than a clean adoption narrative. It is also the only kind that holds up over time.
The Question and the Action
What would it look like if the organizations closest to the community were the ones shaping how AI gets used in the community? That is a design challenge worth taking seriously right now.
KNOW THE POLICY LANDSCAPE
The Mapping AI project (mapping-ai.org) is building a public map of who is shaping U.S. AI policy: who the actors are, what they believe, and where the gaps are. This is a resource worth knowing, contributing to, and sharing. The 2026 midterms and 2028 presidential elections are when the intellectual groundwork being laid right now becomes platform, and that timeline is close enough to matter while still being open enough to shape.
State-level legislation is currently the primary accountability mechanism available to communities harmed by algorithmic systems. Over 1,000 AI bills were introduced across U.S. states in 2025.¹² Colorado, California, Texas, and Illinois all have active or pending legislation. The federal government is actively working to preempt those protections, and that fight is happening now in spaces where the advocacy capacity of community-based organizations genuinely matters.
KNOW WHO IS DOING THE ACCOUNTABILITY WORK
Dr. Joy Buolamwini's Algorithmic Justice League is documenting algorithmic harm and building the legal and policy infrastructure to contest it. Dr. Timnit Gebru's DAIR Institute is producing research that holds AI companies accountable to their own claims. These organizations are doing work directly connected to the communities mission-driven leaders serve. Follow them. Amplify their work. Fund them if you have the capacity.
INSIDE YOUR ORGANIZATION, START WITH INTERROGATION
Before adoption: ask who built this tool, on what data, tested with which communities, and with what transparency about errors. After adoption: document your community's experience of the system, name what is not working, and create real channels for that feedback to change decisions. When AI gives your team time back, make the redirection toward community visible. Name it. Let it be evidence of values in practice rather than efficiency absorbed back into overhead.
The field is still being written, and that is the most honest invitation this moment has extended to us: to come to the table without all the answers, to build something together that none of us could have built alone, and to let the communities closest to this work lead the way toward what comes next.
What becomes possible when we stop waiting for certainty and start building with each other?
If this resonated with you, I want to hear from you. The questions this moment is asking of us are too important and too complex to navigate in isolation, and the organizations closest to community cannot afford to be the last ones at the table. Whether you are in the middle of an AI adoption decision, trying to build a governance structure that actually reflects your values, working to understand what algorithmic systems are already affecting the people you serve, or simply trying to find your footing in a field that will not hold still, this conversation does not have to happen alone. At Intentional Data, this is the work we are committed to: building the conditions for mission-driven leaders to engage with AI on their own terms, with integrity, and in community. We would love to think through this with y
End Notes
1. U.S. Department of Justice. (2022). Justice Department Reaches Settlement with Meta Platforms, Formerly Known as Facebook, to Resolve Allegations of Discriminatory Advertising. justice.gov
2. American Bar Association. (2025, August). Recent Developments in AI Cases and Legislation. americanbar.org
3. Psychiatric Times. (2026, April). Legal Cases Against Big AI. psychiatrictimes.com
4. Lincoln Institute of Land Policy. (2026, February). Data Drain: Land and Water Impacts of the AI Boom. lincolninst.edu
5. Brookings Institution. (2025, October). Reimagining the Future of Data and AI Labor in the Global South. brookings.edu
6. Equidem Research, cited in Brookings Institution. (2025, October). Reimagining the Future of Data and AI Labor in the Global South. brookings.edu
7. Baker Botts. (2026, January). U.S. AI Law and Policy Update. bakerbotts.com
8. Nonprofit Quarterly. (2025). AI Efficiency Gains and Nonprofit Capacity. nonprofitquarterly.org
9. Heifetz, R., Grashow, A., & Linsky, M. (2009). The Practice of Adaptive Leadership. Harvard Business Press.
10. Design Studio for Social Intervention. (2020). Ideas Arrangements Effects: Systems Design and Social Justice. Nonprofit Quarterly, July 13, 2020. nonprofitquarterly.org
11. Social Current. (2026, January). AI Adoption in the Nonprofit Sector. socialcurrent.org
12. Baker Botts. (2026, January). U.S. AI Law and Policy Update. bakerbotts.com