Growing up in a working-class family and as a first-generation college student, I did not know that social science was a career path. In fact, I do not think I had ever heard the term social science before I applied to college.

If there was an early sign that I was on the path that eventually led me here, it was the extent to which I felt the pain in the world around me.

As social scientists, we do not often talk about pain and suffering. That is not the language we typically use. Instead, we talk about social “problems.” We talk about poverty and inequality, disparity and marginalization. We discuss these issues as if we can maintain an objective distance from them.

Yet I am not sure that is how I ever experienced these so-called problems. If you grew up in poverty, you know that research, in the name of science, often leaves little room for forms of pain and suffering that are difficult to measure using scientific methods. I place great value on scientific evidence. Yet that does not mean that everything important can be measured, or that everything worth knowing is fully answerable through science alone.

More fundamentally, engaging with these pains reminds us of what it means to be human. Machines have problems too. They produce random and systemic errors. What, then, makes human problems distinct? My answer is simple. I do not feel pain when I see a machine malfunctioning. To illustrate, I really do like my Mac laptop, but if it stops working, I will either repair it or replace it. There is no deep emotional attachment involved. Can we say the same about our loved ones? Or even about strangers?

I feel pain when I see a person suffering because I see their dignity being harmed. Long before I learned that we could name these pains, measure them, and study their patterns, causes, and consequences, I was already drawn to these questions because I saw, heard, and felt suffering. I still do.

At this point, I should tell you something else about my identity. There is one thing that Trevor Noah, John Legend, and I have in common. Unfortunately, it is not comedy or musical talent. What we share is that we all grew up in households where attending multiple church services on a Sunday was normal.

I mention this because the problem of pain not only drew me toward social problems, but also placed me in an uneasy relationship with my religious upbringing. This tension became more pronounced as I became more serious about scholarship.

If God is good, why is pain so pervasive? Why is there so much suffering in our personal lives and communities?

I have wrestled with this question for decades. I have found some partial answers, but I am still searching. To some degree, I am comfortable, and even grateful, to live in that tension. Being troubled, curious, and engaged helps me remain sensitive to the suffering around me and continually reflect on my own moral and ethical values, responsibilities, and actions.

More importantly, having incomplete answers makes me willing to seek, learn, and grow with others. I do not want faithful living to mean closing my eyes and ears to the world. Instead, I hope it means opening our hearts and minds and sharing our lives with one another.

In that context, my favorite verse in the Bible is Hebrews 11:39-40:

These were all commended for their faith, yet none of them received what had been promised, since God had planned something better for us so that only together with us would they be made perfect.

These are discouraging words if you believe that God works like a vending machine. If you provide certain inputs, such as attending church, reading the Bible, and praying, He will deliver the outcomes your heart desires.

Yet the examples in this chapter suggest otherwise. Among the people mentioned in this chapter were those who were “tortured,” who “faced jeers and flogging, and even chains and imprisonment.” Some were “put to death by stoning,” “sawed in two,” or “killed by the sword” (Hebrews 11:35-37).

I did not find this passage discouraging because it reminds us that democracy, education, and religion are all, fundamentally, shared projects.

We live in an age in which our attention is often fixed on individual accomplishments. When that happens, we risk losing sight of the forest for the trees. The promise in this passage is not made to an individual, but to a community. It is not confined to a single generation, but extends across generations. That is why the author writes that “only together with us would they be made perfect.”

But how does this coming together actually work?

I have essentially grown up in churches, so I can say with a clear conscience that churches have good, bad, and ugly sides. In that respect, they are not so different from families. They can be sources of hope, resilience, and belonging. They can also be sites of manipulation and abuse.

Aside from the problem of pain, another question has long troubled me: if God’s presence is universal and His grace extends to everyone, why have churches so often become sources of division rather than unity?

Examples abound. To my knowledge, the first interracial church in the United States was the Church for the Fellowship of All Peoples in San Francisco, founded in 1944 by Black theologian Howard Thurman and white minister Alfred G. Fisk. Thurman, the famed author of Jesus and the Disinherited (1949), which Martin Luther King, Jr. read and re-read, left his tenured position as Dean of the Chapel at Howard University to establish this church. He remained there until he moved to Boston University in 1953. The movement to make churches racially integrated spaces began less than one hundred years ago in this nation’s long history.

Even knowing this, I remain stubbornly hopeful. Political scientists Rogers Smith and Desmond King have argued that American political development contains multiple and competing racial orders (2005). Progress has never been linear because forces of inclusion and exclusion coexist.

There are parallels in theology. We can follow what Walter Brueggemann calls the prophetic imagination (1978), which begins with lamenting the status quo and imagining a transformed world. Similarly, Willie James Jennings (Yale) has written about the intertwined histories of race, empire, and Christianity, and how they continue to shape Christian beliefs and practices today (2010, 2017). Which vision we pursue depends on our collective agency.

My coauthor and colleague Hahrie Han’s book Undivided: The Quest for Racial Solidarity in an American Church (2024) is one of the best books written about American politics, democracy, and religion in recent years.

What makes the book distinctive is that it does not caricature faith communities. Hahrie immersed herself in a megachurch in Cincinnati, Ohio, seeking to understand how members wrestled with racial divisions in the aftermath of George Floyd’s murder. She followed four people, two Black and white men and two Black and white women, through the details of their daily lives because only in those specific times and places can we truly understand lived experience, faith, and human connection. Their beliefs changed only after those connections were formed, not before.

I was fortunate to read and comment on the book manuscript before it was finalized. Hahrie’s book inspired me to find a local church in Oakland, affectionately known by locals as “Oaktown.”

For those unfamiliar with Oakland, it is one of the most racially diverse cities in the United States. No single racial group holds a demographic majority. The church reflected that reality. It was racially and economically diverse. The congregation looked like the community around it.

Like its counterpart in the Midwest, it was trying to reckon with historical injustices while offering alternative visions of what community could look like. The groundwork was not easy, but people kept showing up. Simply sitting there and observing that congregation filled me with a tenacious sense of hope.

This week I was in Park City, Utah, attending the 2026 Veritas Scholars Summit. According to the organizers, it is “an annual event where Christian faculty come together to explore a vision for faithfully serving our university communities.”

I am not entirely sure whether I fit either the Christian or faculty categories, although technically I qualify for both. In some ways, I feel like a recovering academic and a recovering Christian. I have become disillusioned with parts of both my profession and my faith community, yet I am still trying to find purpose and, most of all, community. Given where I am in that journey, I am grateful for the opportunity to reflect on these questions with fellow scholars and learn from their wisdom and experiences.

This morning, I woke up before sunrise and walked through the mountain neighborhood where I was staying. Below is a picture I took during that walk.

As I reflected, I found myself returning to the same old question: Given the pain in the world, what shall I do?

I also began to wonder whether that is the wrong question.

The task for building a mission-driven organization was never meant to belong to a single individual, whether that is me or anyone else. No matter how flawed we are as individuals, or how broken our communities may be, we are not alone in this work.

I am still skeptical. I am still flawed. But I am certain there are signs of hope. I am stubbornly optimistic that we can cultivate them together.

Why do I think this book deserves a wide audience, including scholars, students, and practitioners?

In short, it is one of the best books written on policy implementation in recent years.

After the book was published, I emailed Martin to congratulate him. I wrote:

I see you as one of the most compelling interpreters of Wildavsky in our generation.

Martin may not be a household name, but among scholars of implementation he has become one of the most insightful interpreters of Aaron Wildavsky’s intellectual legacy. Wildavsky, together with his student Jeffrey Pressman, authored the classic Implementation: How Great Expectations in Washington Are Dashed in Oakland; Or, Why It’s Amazing that Federal Programs Work at All. Wildavsky also helped found UC Berkeley’s Goldman School of Public Policy.

The book remains influential more than fifty years after its publication.

What makes it remarkable is not its famously long title, but its enduring relevance. Even practitioners, who are often skeptical of academic research, continue to engage with it. For example, Jennifer Pahlka, founder of Code for America, where I previously worked as a data scientist, and author of Recoding America, recommended the book during her 2023 interview with Ezra Klein. She described it as highly relevant to how she thinks about government capacity and reform.

If you find Wildavsky compelling, you should also read Martin Williams. His book belongs in the same conversation as Pam Herd and Don Moynihan’s Administrative Burden (Russell Sage Foundation 2018), Dan Honig’s Mission Driven Bureaucracts (Oxford University Press 2024), and Akshay Mangla’s Making Bureaucracy Work (Cambridge University Press 2022).

All of these books examine how policy actually works on the ground. They draw our attention to the operational details that are often overlooked when we focus exclusively on legislation or formal rules.

One of the key insights of Reform as Process is that there is far more happening behind the scenes than most observers realize.

In public bureaucracies, especially at the state and local levels, much of the work revolves around budgeting, procurement, workforce (including but not limited to performance) management, and interagency coordination. These activities may not attract the same attention as passing a major bill through Congress, but they are often where outcomes are determined.

The insider’s view of government reform resembles a duck gliding across a lake.

From the surface, the movement appears smooth and effortless. Beneath the water, however, the duck is paddling constantly. Traditional stage-based models of policy change often fail to capture this continuous process of adjustment, coordination, and problem-solving.

The metaphor breaks down in one important respect. A duck has two feet. Government has many. Public organizations are less like a duck and more like a many-headed dragon. Different departments, agencies, and units frequently pursue different goals, operate with different information, and face different incentives. Implementation is difficult because coordination is difficult.

What, then, is implementation? Following Pressman and Wildavsky, it is the process through which policy objectives are translated into outcomes.

How do governments improve implementation? Through collective learning-by-doing. To borrow Wildavsky’s language, doing is implementation and learning is evaluation. The two are inseparable. By repeatedly acting, evaluating, adapting, and acting again, governments incrementally improve how they work.

This theme runs throughout Martin’s book. Although the book covers many topics, including performance management, its core contribution is to show how reform emerges through iterative processes of experimentation and adaptation rather than through singular moments of transformation.

Let me highlight two insights from the book that I found especially valuable.

For scholars and students of public policy, consider this observation:

“Nearly every reform can be viewed both as a partial failure and as a partial success.” (p. 60)

Across six African countries studied in the book, Ghana, Zambia, Kenya, Nigeria, Senegal, and South Africa, Williams finds that reforms were more numerous than we typically assume. They often overlapped with one another and frequently targeted similar problems. Many reforms failed to achieve all of their stated objectives. Yet many also generated meaningful improvements.

This should not be surprising. Plans rarely unfold exactly as intended because they must confront complex realities. Ambitious reforms are particularly likely to be labeled failures because they begin with ambitious expectations.

In this respect, Williams’s argument echoes the lessons of Pressman and Wildavsky. Their subtitle, How Great Expectations in Washington Are Dashed in Oakland, captures a case in which implementation fell short of expectations. Yet neither their work nor Martin’s suggests that ambitious programs should simply be judged as failures. The point is not that government intervention inevitably fails. The point is that implementation is difficult because planning has clear limitations and must contend with reality.

Depending on where we look, we often find meaningful successes embedded within apparent failures. Returning to the War on Poverty programs, my own research suggests that one of their successes was the way they encouraged marginalized groups to organize, build civic infrastructure, learn how government works, and find ways to influence public institutions. These dynamics form a major theme of my first book project. One reason I appreciated Martin’s book so much is that it emphasizes how much we can learn from both partial successes and partial failures. These mechanisms are the foundation of collective learning.

For practitioners, I hope readers pay particular attention to Martin’s critique of what he calls the “projectization (one-off intervention)” of reform.

“One common feature across the majority of reforms was that they were designed and implemented as one-off projects. They tended to diagnose a performance problem with the civil service system, propose some kind of intervention into that system, and posit that this intervention would ‘fix’ the problem once and for all. Bureaucratic reform was typically viewed as a discrete and often time-bound action: the passing of a new law, the implementation of a three-year plan, or the creation of some new bureaucratic process. This mental model of reform as a one-off intervention, the projectization of reform, constituted the second main mechanism of reform failure.” (p. 104)

Reading this passage reminded me of William Eggers and Donald Kettl’s Bridgebuilders: How Government Can Transcend Boundaries to Solve Big Problems (Harvard Business Review Press 2023). Although they focus on the United States, they identify a remarkably similar pattern.

As Eggers and Kettl write:

“Historically, many of the management fixes have been based on what Kettl has termed a vending-machine model of government: You have a problem, you build an organization to deal with it, you put money in the slot, and you expect it to yield results.” (p. 9)

Too often, reform becomes synonymous with launching a new initiative, creating a new office, or introducing a new project. Instead of improving existing processes through continuous learning, governments search for dramatic interventions. This can amount to a form of magical thinking, the belief that a bold new program will solve problems without sustained attention to details.

The lesson from Williams’s book is that reform is less about grand gestures and more about maintaining the steady paddling beneath the water.

I want to close with Herbert Simon’s famous parable of the ant on the beach (1962). Simon observed that an ant’s path across a beach appears extraordinarily complex. Yet the ant itself follows simple decision rules. The complexity arises not because the ant possesses a complex mental model of the world, but because it is navigating a complex environment.

The same insight applies to implementation and government reform. Reform appears complex because governments operate in environments filled with competing interests, uncertainty, organizational constraints, institutional variation, and changing political circumstances. Complexity emerges from the interaction between actors and their environments.

This perspective also captures what makes Martin’s book so valuable. Rather than offering a universal playbook for reform, he develops a set of practical, mid-range insights that serve as a toolbox for thinking about organizational reform.

As Martin writes:

“Instead, useful and generalizable knowledge about systemic reforms has to consist of a set of midrange theoretical insights about different parts of the reform process, mental models that reformers can absorb, remix, and apply to the unique demands, constraints, and opportunities of their own particular contexts. In other words, learning from evidence about reform isn’t like following a blueprint that tells you exactly what to do. Instead, it’s like compiling a toolbox and learning what each tool does and doesn’t do so that you can use these tools to solve your own challenges in your own context.” (p.11)

The strength of books like Reform as Process is that they provide simple, but not simplistic, ways of understanding complex problems.

Subscribe now

I’m doing this work because improving policy requires more than producing research. It requires learning from others and engaging with the people who design, implement, and experience policy.

My version of policy entrepreneurship is closer to the role of a facilitator. I aspire to be someone who helps create space for collective learning and problem-solving.

Researchers analyze problems. Analysis originates from the Greek word analyein, which means breaking down complex ideas into manageable parts. Despite the value of analysis, I have begun to wonder whether this is really how things work on the ground, or whether it is the best way to improve practice.

1. Problem definition and problem-solving are not opposite ends of a process. They are integral parts of the same process. Sometimes, you move in the other direction. You realize that you defined the problem inaccurately while trying to solve it. If you cannot solve a problem based on the premise that your diagnosis is correct, perhaps the diagnosis itself is wrong. Then you need to start over. Research, and policymaking, is inevitably an iterative search process.

2. Yet iteration alone is not enough. Solving real-world problems also requires connecting insights across different parts of a system. We study policy by breaking it into manageable pieces, but improving policy requires putting those pieces back together. In short, fixing is not the same as improving.

3. Donald Stokes was a leading figure in the Michigan school of political science, which helped establish the scientific study of mass politics. He later served as dean of Princeton’s School of Public and International Affairs. Near the end of his career, he published Pasteur’s Quadrant through the Brookings Institution in 1997. In the book, Stokes challenged the dichotomy between basic and applied research that had been articulated roughly fifty years earlier by Vannevar Bush, who led the U.S. Office of Scientific Research and Development . Bush’s report, Science, the Endless Frontier (1945), became the intellectual foundation for the National Science Foundation and the postwar American research enterprise more broadly. Stokes argued instead that there is a middle ground between basic and applied research, which he called use-inspired research. Discovery and problem-solving, he suggested, often go hand in hand .

4. There are multiple pathways to use-inspired research. One is to define and solve problems in close partnership with the people who experience them. Whether we call this approach applied research, engaged scholarship, co-production, or human-centered design, the core idea is straightforward. Researchers cannot remain confined to the ivory tower. They need to engage with the real world, what political scientist Richard Fenno called “soaking and poking.” Put differently, this requires us to reflect on the fact that we are not above the world; we are in it. There is no knowledge from nowhere, as sociologist Michael Burawoy reminds us.

Expertise stems from defining narrow contexts and immersing ourselves in them. That is how PhD programs are designed. That is how most academics, including myself, are trained.

But once you build expertise, it can also become a ceiling. The professional world that our younger selves once tried to enter can become a comfort zone that limits both growth and impact.

Deepening expertise and making it widely useful requires breaking that mold intentionally, consciously, and collectively. Hadley Wickham, Chief Scientist at Posit (previously RStudio), argued that one way to create impact is by “being useful.”

If you want to be useful, you need to understand who your users are and what they need. This distinction matters because the essence of creating impact is not fame or influence. It is not about “me.” It is about the “mission” and serving the public.

If I want my research to be useful to civil servants and policymakers, I need to understand their needs and constraints. If I want my research to be useful to members of the public and advocacy groups, I need to do the same homework.

To do so, I cannot simply ask people to read my papers. Most will not. They are too busy. If I want their attention, I need to meet them where they are. My task is to be a partner in their efforts to define and solve problems. That means immersing myself in messy, multi-stage, multi-actor implementation processes. It means managing stakeholders, working under tight deadlines, and recognizing that opportunities to translate ideas into action are often time-sensitive.

Most important of all, it means letting go of ego. Listen carefully. Listen respectfully. Listen critically. To that end, I spend a great deal of time meeting and talking with practitioners. Even during this summer break, outside this DC event, I have met with think tank staff and public servants and had multiple coffee chats with people working in civic technology. I also co-founded the Data for Good Roundtables to make these conversations more structured, public, and open.

My mantra is that no one knows everything, and everyone is bound to make mistakes. We all have limitations. Therefore, we need not only analysis but also synthesis. Understanding the parts matters, but so does seeing how those parts fit together into a larger whole.

In software engineering, there is a role called a full-stack engineer. It describes someone who can build both the front end and the back end of a tech stack. Or, to use a restaurant analogy, someone who can run both the dining room and the kitchen.

I believe there is something analogous in public policy, which I call the full-stack policy scientist. These individuals understand not only research, but also implementation, communication, stakeholder management, and policy translation. They know how the different pieces of policymaking fit together.

I learned how to make myself useful to people implementing policy while working as a data scientist at Code for America, where I partnered with federal, state, and local governments to improve access to safety net programs. I learned how to make myself useful to people designing policy through my collaboration with the Federation of American Scientists, where I helped translate research on AI procurement into a policy memo.

Through these experiences, I came to see policy work as a process of collective learning-by-doing.1 If you prefer Popper’s framework of falsification, you might think of engagement as exposing ideas to broader criticism and therefore increasing the likelihood that they become testable. Whether the issue is the strength of evidence, the usefulness of a conceptual framework, or communication across audiences, wider engagement creates more opportunities for learning.

There is a long list of people who helped me develop this capacity and engage with practitioners and the public. I will not try to name them here because the list is too long, and I would inevitably leave someone out.

The point is that there is now a growing community of people across disciplines who care about the full cycle of policymaking, from design to implementation. We need these people because both upstream (design) and downstream (implementation) policy processes shape delivery outcomes.

Professionalization is necessary and useful. Analysis is necessary and useful. Division of labor is necessary and useful. But policy scholars should not become so professionalized that we stop listening to people outside the ivory tower because we are too invested in our models and methods. Breaking down problems and isolating causes and effects help make complex issues tractable. But important problems rarely produce clean analyses. They are messy, entangled with human interests, organizational constraints, and competing values.

We need full-stack policy scientists because understanding policy requires analysis, but improving policy requires synthesis. It requires connecting the entire process, from design to implementation.

Leverage the power of analysis, but do not forget that, as education scholar David Perkins (2010) argued, in the end, we need to “play the whole game.”

A few weeks ago, we started reading J.R.R. Tolkien’s The Hobbit and The Lord of the Rings. That led to Harry Potter. That, in turn, led to a book on owls by Leigh Calvez because my daughter happened to find Hedwig, Harry Potter’s owl, interesting. Since then, each night, we’ve learned about one type of owl after another. Sometimes it’s tiring because it comes at the end of a long day. But it has always been a joy to read and discuss books, and the worlds of words and ideas they contain, with the apple of my eye and the object of my deepest affection.

I was born with birth defects and went through many surgeries and long hospitalizations. I was a strange kid who did not talk much. My focus was always on my inner world. If you were in deep pain, and if you believed that other people did not understand you, you dug the well, built the wall, and lived inside yourself.

Years later, when I got married, I was hesitant to have a child. I knew that my condition was not entirely genetic. As a matter of fact, it was relatively minor. Still, I didn’t want to take the chance.

When my daughter was born without the conditions I had, I focused only on what she had been spared.

What I didn’t realize was how many blessings that she would bring into my life.

The Beginner’s Mind

The greatest of these blessings is the beginner’s mind (chosim in Korean). From the moment she first walked, spoke a word, and read a book, everything was new, and therefore, new to me as well. Watching her take on challenges and overcome them has been a true joy. It helped me realize how much I missed during my childhood because I was so absorbed in my internal world. Through her eyes, I had the chance to experience childhood, its wonder and beauty, again. Now I’m filled with her memories, not mine. For that, I’m eternally grateful.

The Continuation of Life

Another blessing is the continuation of life. My father passed away one year before I came to graduate school. He missed so much of my life, even though he had worked so hard to make that life possible. I carried that absence with me for years.

My daughter’s presence somehow filled that hole in my heart. One life is gone, but another life arrives and continues. She bears a strong resemblance to both me and her grandfather. Perhaps each of our lives is incomplete, but together we become something greater than the sum of our individual parts.

I believe parenting is a gift. Not everything associated with parenting is ideal. It is far from it. For starters, sleep deprivation has been a real issue. Parenting is a source of happiness, but also anxiety. It expands the range of emotions that we experience.

Parenting is a gift because it pushes us to understand ourselves in ways we never would have otherwise.

Parenting helped me learn that life is broken and fragile. You may be born with birth defects. You may lose your father far earlier than you expected. Still, you have a second chance. Still, you can find meaning and recover joy.

I love spending time with my daughter because every moment is an opportunity to appreciate that gift. When I reflect on this blessing, I am reminded of the words in Hebrews 11:39–40.

³⁹ These were all commended for their faith, yet none of them received what had been promised, ⁴⁰ since God had planned something better for us so that only together with us would they be made perfect.

Simply put, procurement is the door through which technology enters government, including but not limited to AI. If we want responsible AI in the public sector, procurement is the first line of defense.

In the past few years, I’ve been working on how state and local governments buy AI technologies by analyzing thousands of AI contracts with my colleague Aniket Kesari (associate professor at Fordham Law School).

Although many scholars have argued that we should use procurement contracts as leverage for embedding AI responsibly in governance (see Coglianese and Crum 2025), we don’t see strong evidence that responsible AI provisions are being incorporated into procurement contracts in practice. While governments are rapidly purchasing AI systems, very few contracts include meaningful requirements for transparency, fairness, accountability, or bias testing (see our FAS policy memo).

State-level AI Procurement contracts are not only expensive, but also long-term commitments. On average approximately 7 years. The terms negotiated (or not) today can shape public services for years. Some AI contracts last even a decade or longer.

This year, Aniket and I worked with the Federation of American Scientists and the Kapor Foundation to translate our ongoing research into a policy memo. In addition to highlighting key findings from our research, we sought to draw attention to important grassroots efforts such as the GovAI Coalition, as well as promising practices already implemented by local and state governments, including the City of San José and the State of Colorado.

The memo was released today as part of Source Code, a collection of policy memos focused on building trustworthy, accountable, and people-centered AI governance.

I will also be in DC this week from Wednesday through Saturday for the launch event from June 11th 3:00 PM - 5:00 PM EDT.

Please feel free to reach out if you'd like to connect, chat, and/or collaborate!

I am aware of this trend and have tried to engage with it. This is similar to the pandemic period, when everyone did some form of public health research. I have worked on projects involving AI, from analyzing state-level AI procurement contracts to evaluating AI chatbots for safety net delivery. About half a year ago, when I was invited by University of British Columbia to join a panel on AI and social science research, I told the audience that my reaction to this technology is similar to the attitude I hold toward my seven-year-old daughter. I am very excited and also very worried. My reaction is bimodal because there is enormous potential, but also substantial risk.

Experience does not equal insight. I do not have any fundamentally new insight into this emerging technology. I am nowhere close to the point where I can confidently recommend a definitive set of strategies that public organizations should follow.

I nonetheless have begun to develop a set of heuristics, to use Herbert Simon’s term, that may be useful for navigating potential AI-enabled solutions to societal problems, in safety net programs and beyond.

I share them here as a way to expand the dialogue and learn from others. I know some things, but only some things, and I am often wrong.

My basic premise is that, following Landau (1969) and Bendor and Kumar (2005), all public sector organizations face the FDA problem. The FDA’s mission is to review food and drugs to ensure that they are safe and effective. There’s a similar relationship between governments and emerging technologies. Governments want to benefit from emerging technologies, including but not limited to AI. If they miss valuable opportunities, that is a type II error: rejecting good innovations. But if they embrace these technologies without a healthy dose of skepticism, that becomes a type I error: accepting bad ones.

This is why Aaron Wildavsky, one of the pioneers of implementation research, emphasized that implementation and evaluation are two sides of the same coin (1984).

Civic tech in general is based on a simple theory of change: “show what’s possible, and inspire others.” The aim here is to avoid type II error. That approach is not wrong, but it is insufficient.

I argue that pushing in this direction should be balanced by strengthening evaluation capacity. Otherwise, we risk increasing type I error.

Given the scale and speed of AI adoption in government, balancing these two approaches to reform, implementation and evaluation, is more important than ever.

Greater power through innovation must come with greater self-examination through criticism. Ironically, strengthening state capacity requires both enabling innovation and imposing certain constraints1.

What, then, might these heuristics look like? I cannot offer a comprehensive list. Given the evolving nature of AI, I do not believe there is a single optimal way to deploy these tools. Instead, governments should rely on a set of reasonable heuristics, learn from data and implementation experience, and continuously improve over time.

Below are six heuristics for AI adoption in government that can be used selectively or in combination to balance type I and type II errors. My focus here is the use of AI tools to reduce administrative burden, one of my main research areas.

Can AI reduce administrative burdens and make government services easier to use?

Yes, but it depends on how AI adoption is designed, which is fundamentally not a technological problem, but an organizational, institutional, and political one.

1. Start with the burden, not the (AI) technology. Identify the pain point and understand its form and origin. Using the administrative burden framework (Herd and Moynihan 2018), ask whether the problem is primarily a learning cost, compliance cost, or psychological cost. Focus not on “human in the loop,” but on human-centered design principles. Problems may stem from applicants, caseworkers, paperwork systems, policy interpretation, or workflow coordination. Start with problems, not solutions.

2. Map the process before adding AI. Some steps require simplification, policy change, staffing, or better service design, not AI. Not every problem needs an AI solution. Governments should be explicit about this boundary.

3. Use AI where the task is bounded. The more open-ended the task, the higher the risk. AI is attractive because it is flexible, but that flexibility is also a source of limitation. Returning to Herbert Simon’s The Sciences of the Artificial (1970), perhaps the better approach is not understanding or improving the model itself, but making the external task environment more targeted, constrained, and simplified.

4. Treat every AI tool as fallible. AI systems produce errors, and those errors emerge from multiple sources: models, systems, users, and their interactions. The key question is not whether errors exist, but who bears the burden of correcting them.

5. Evaluate effectiveness, usability, and safety together. A tool that is accurate but unusable will fail. A tool that is usable but unsafe can cause harm. Again, all organizations face the FDA problem. We must balance type I and type II errors.

6. Design not for applicants and caseworkers, but with them. AI should reduce burdens on both sides of state-citizen interactions. It should not shift work from agencies to applicants or from systems to frontline staff. The challenge is that much of this information, especially the experiences of applicants and caseworkers, is private or difficult to observe. Value expertise, but not too much. Expertise in one area can create blind spots in another. Familiarity with one part of government service, whether law, budgeting, or technology, can create what psychologists call the “curse of knowledge.” This can make it harder to recognize usability problems from the perspective of ordinary users or frontline staff who actually manage with the system.

Subscribe now

We are both academics, but previously practitioners1, and currently working at the intersection of research and practice.

Field experiments have been one of our main tools for evaluating what works on the ground and help agencies to continuously improve their practice.(It also used to be my full-time job at Code for America. I worked with federal, state, and local governments to improve safety net delivery through design thinking and field experimentation.) Using field experiments to build reliable evidence has been a trend in the U.S. since at least the 1960s. This history is well documented in Judith Gueron and Howard Rolston’s book Fighting for Reliable Evidence (2013). Before A/B testing became a buzzword, field experimentation helped governments do a better job.

The method is powerful first because it help us casual questions in real-world environments. But more importantly, they are great communication tools. They produce evidence that is simple, credible and often easier to communicate and defend to decision-makers (Ashenfelter 2014).

But field experimentation is still seen and felt as hard. Our argument is that the main issue might be not methodological but relational. Field experimentation requires a partnership between researchers and practitioners. Unlike Disney movies, the rest of the story is not always living happily ever after.

The tension between these two cultures, scientific rigor and discovery on the one hand, and operational feasibility and problem-solving on the other, is what we call the alignment problem.

The paper is practical and pedagogical.

We summarize and explain the four failure points as scoping, implementation, measurement, and communication (SIMC, intentionally named to sound similar to “sync”). Depending on how researchers and practitioners find alignment around these points, the collaboration can succeed or fail.

If you’re interested in field experimentation, or working with partners more broadly, I argue that the most important thing is to be humble. Respect anyone and everyone regardless of their background (read Carolyn Heinrich’s recent interview).

To me, humility means caring for and respecting everyone and striving to understand and respond to the needs of others, regardless of their status, wealth, background, or social position. It means recognizing that no one is inherently “better” or more deserving of compassion, attention, and kindness than anyone else.

WHY GOOD POLICIES FAIL, Carolyn J. Heinrich (2026)

Researchers can help design evidence and support decision-making, but frontline administrators, implementers, and public servants are often the real gatekeepers of experimentation. They absorb the operational and political risks when experiments succeed or fail. They deserve much more recognition in conversations about evidence-based policymaking. This is why we quoted the following passage from Gueron and Rolston at the beginning of the paper.

“In our view, the most important group—the main heroes in this book—were the state and local administrators who displayed the courage, determination, and political skill needed to put their programs on the line, and in some cases, win the battles that erupted when we sought to open up real-world programs to a new type of scrutiny. ”

Fighting for Reliable Evidence, Gueron and Rolston (2013: 428)

This project began because Donald Moynihan2 invited me to speak at a civic tech pre-PMRC conference at the UW Evans School when I was still working as a data scientist at Code for America. That’s actually how I met Florian!

I also gave a talk on an earlier version of this paper at the Toronto Data Workshop, where I was invited by Rohan Alexander. You can find the recorded talk here.

Winston Lin from UPenn has been very supportive and provided many important ideas and references throughout this project. I’m grateful to all of these people for their generosity and thoughtfulness.

Lastly, I’m interested in turning this paper into workshop material.

I’m hoping to do that through the field experimentation workshop that Vince Hopkins from UBC and I are organizing next summer (May 2027) in Vancouver. The slide deck version of this paper was already used as a reference in a two-day workshop on field experiments in governments at Georgetown’s McCourt School of Public Policy, co-hosted by LBJ School of Public Affairs, a few years ago.3

If you have any suggestions, please do not hesitate to reach out to me. I would love to hear your thoughts and learn from them.

I took it twice: once early on, and again later in the dissertation writing stage. I wasn’t alone. Many of the best dissertations, award-winning papers, and eventual books passed through that seminar.

Among Berkeley political science alumni, there’s a quiet fan club of Ruth Collier. A few years ago, I co-organized a happy hour at APSA just so we could swap Ruth stories.

Ruth cared deeply about her students. She was kind, generous, and attentive in ways that mattered. Yet she did not let you off the hook. Each student was expected to present at least twice per semester. The presenter circulated a 5–6 page draft in advance, and everyone else came prepared to read it closely and provide comments. Ruth read everything, literally everything. She read every footnote and every endnote. She also told you, clearly and directly, when something did not make sense. You received the most honest and demanding feedback, along with clear-eyed and generous encouragement. She saw the potential in drafts you were ready to abandon and pushed you toward it, without letting you take shortcuts.

A good scholar, she modeled, must be curious, dedicated, and deeply caring for the craft. And as Rainer Maria Rilke once wrote in Letters to a Young Poet, she helped us go deeper within ourselves and find strength there we didn’t know we had. That’s what her push was for.

One of the most distinctive parts of her method was what happened when you couldn’t explain your idea clearly. If you were struggling -mumbling through a mechanism, circling around a concept - Ruth would stop you. “Go to the board,” she’d say. At least once per semester, each of us had to stand up and diagram our core argument. We then needed to draw a diagram and show how it worked. In private conversations in her office, she told me that she loved “grappling with ideas,” and it was her way of helping us do the same. It also served as a useful heuristic or criterion. If you could not diagram it, you did not yet understand it. The chalkboard exercise was uncomfortable, sometimes even humbling.1 It was also clarifying. It forced us to confront the fuzziness in our thinking. It sharpened our ideas and made our arguments more honest and more our own. It narrowed the distance between us and our ideas. The funny thing is that many of us, Ruth’s students, now use this method to mentor and teach our own students (including myself).

Over time, many of us developed what we jokingly called our “inner Ruth.” It was the voice that told us when a sentence was not clear enough, when a claim was not grounded, and when a paragraph was faking it. I do not know exactly how it happens, but the gift of her teaching is that even after the course ended, that voice did not go away. It became part of how we wrote and how we thought.

Ruth’s brilliance was not just in her intellectual depth, though she was one of the best comparativists, or political scientists, of her generation.

She studied at least three world regions, Latin America, Africa, and Europe, and could move fluently between them. For many political scientists outside these regions, they may know Ruth through her work on historical legacies with her husband, David Collier, but her expertise is vast.

Her gift as a teacher was in how she trained us to think with precision and write with integrity. She asked us to be true to ourselves and tell the truth, in our own voices, to the audiences that mattered. She did not just teach us how to write better. She taught us to know what we were trying to say, and why. I remember telling her that I did not want to offend gatekeepers in the field while working with her in the workshop on the paper that eventually became my first academic publication. In response, she told me that if I believed what I wrote, I should go for it and own it.

In doing all this, Ruth created a space that was both demanding and caring.

It was a place where it was safe to take risks, and expected that you would. Her workshop trained us to write dissertations, yes. But it was more than that. It taught us how to do scholarship with clarity, honesty, and courage. Ruth changed our habits of mind and the way we tackle new problems. She also shaped how we educate and train the next generation of scholars.

Looking back, I have often wondered why her class was so influential. This is a frequent subject among Ruth students because taking the course with Ruth was such a transformative experience. In graduate school, you learn many things from many people, including advisors, instructors, peers, and others. Even so, it is not easy to find an intellectual role model. Ruth was that for me, and for many of us. Perhaps that is why I will not, and cannot, forget her lessons.

Ruth did not just tell us what scholarship should be. She showed us.

Since his debut in 1979 with Kaze no uta o kike (Hear the Wind Sing), Haruki Murakami has been a prolific writer. The figure below traces the novels, short story collections, and non-fiction essays that he published over four decades.

This year, Haruki Murakami’s new fiction is coming. Its title is The Tale of Kaho and it’s first time he features solo female lead. This person, who is now 77 years old, is still experimenting and writing a major novel.

Although much of Haruki’s life is private (he does not use social media or give interviews often, so do not expect to find him on podcasts), he wrote a couple of books in which he explains how he works as a professional. Two that come to mind are Novelist as a Vocation and What I Talk About When I Talk About Running. I personally like the latter more than the former.

Academic researchers are not much like novelists when it comes to career goals. We also want to be original and prolific. Like Haruki, we want to play a long game and do meaningful work for as long as possible. You can like or dislike Haruki, as everyone’s taste is different. But no one can deny that he is original and prolific. The record speaks for itself.

Here are a couple of things I wrote down years ago while thinking about what I learned from Haruki.

In What I Talk About When I Talk About Running, Murakami reflects on two central aspects of his life: writing and running. For him, the two are deeply connected. He explains how becoming a professional writer meant completely reinventing his lifestyle, including when he wakes up, how he works, what hobbies he enjoys, where he lives, and who he spends time with.

After returning to this book again and again, I’ve come to hold onto three lessons that continue to shape how I think about talent, focus, and everyday discipline.

1. “I stop every day right at the point where I feel I can write more.” (Haruki Murakami)

This principle runs counter to how many of us are tempted to work. When things are going well, it’s easy to keep going and chase the short-term satisfaction of crossing something off the list. But Murakami urges the opposite: stop while you’re still in flow. That way, you preserve energy and motivation for the next day.

Trying to do too much, too fast, is a surefire path to burnout. As Murakami puts it, “To keep on going, you have to keep up the rhythm. That is the important thing for long-term projects.” And in academia, every project is a long-term project. Early in my career, I used to believe some ideas could be wrapped up in a few months. I no longer believe that. What I’ve learned instead is that finishing well requires knowing when to pause, how to conserve energy, and how to stay steady over time.

2. “I’m struck by how, except when you’re young, you really need to prioritize in life, figuring out in what order you should divide up your time and energy. If you don’t get that sort of system set by a certain age, you’ll lack focus and your life will be out of balance.” (Haruki Murakami)

Murakami writes that talent is the most important quality for a novelist, but it’s also unreliable: “Talent has a mind of its own and wells up when it wants to, and once it dries up, that’s it.” Not only is talent unevenly distributed; it can also disappear.

That’s why, he says, the second most important quality is “focus, the ability to concentrate all your limited talents on whatever’s critical at the moment.” For Murakami, that means waking up early and writing for three or four hours each morning. “I don’t see anything else. I don’t think about anything else.” Without that kind of focus, nothing meaningful gets done.

But focus, as Murakami describes it, isn’t just about maintaining distraction-free work hours. It’s also about knowing what not to pursue.

He reminds us that you can’t please everyone. When he ran a bar, where he wrote his first novel at the kitchen table, he realized that if one out of ten customers liked the place enough to come back, that was enough. “It didn’t matter if nine out of ten didn’t like my bar. This realization lifted a weight off my shoulders.” What mattered was being clear about his stance and sticking to it.

The same applies to creative pursuits, including academic ones. What matters is knowing what you want to focus on, working on it consistently, and reaching the people who matter to you. You can’t do everything or reach everyone. Focus doesn’t come from grinding harder; it comes from letting go of what’s not essential. You need to know where to invest your time, how to channel your abilities, and what goals are truly worth pursuing.

3. “If concentration is the process of just holding your breath, endurance is the art of slowly, quietly breathing at the same time you’re storing air in your lungs.” (Haruki Murakami)

This third lesson is about rhythm, about knowing when to push and when to pause. Murakami continues, “Unless you can find a balance between both, it’s difficult to write novels professionally over a long time. Continuing to breathe while you hold your breath.” To endure as a writer, or any kind of creative, means learning to resist extremes. That’s what makes you a good contrarian.

Sometimes, you need to push through discomfort. Other times, you need to slow down, listen to your body, and adjust. The goal isn’t balance in the abstract; it’s rhythm, a pace that isn’t too fast or too slow, one that adapts to your needs over time.

And finding that rhythm often requires building routines or rituals, ways to remind yourself how to return to it.

For me, that means running, hiking, and cooking. Running helps me sync my mind and body. Hiking reminds me I’m part of nature. Cooking connects me with the seasons I observed and experienced when I was young, and brings back childhood memories of a home that is, quite literally, an ocean away. There’s no single right way to recover, but we all need one.

There’s a quiet kind of pain in how we face uncertainty and rejection in academia. Sometimes, even doing well can be harmful if we don’t know how to stop and take care of ourselves.

Pain is inevitable. Suffering is optional. To make that suffering optional, we need to know when to push, and when to pause, reflect, and recharge.

If you think about it, running races are a strange activity. Runners come in all forms, but they share something. These people are weird. You wake up very early, sometimes in freezing weather, to stand in line with strangers. If the race is not local, you travel the day before. You train for months. For what? To pay to suffer, on purpose.

That morning, I noticed something unusual. The people in my pace group looked very fit, suspiciously fit. I am not fast. Runners like me are not like that. They are typically parents who look exhausted but run because it is the only time they have for themselves. This is our version of a break.

A man, apparently much younger than me, turned to me and asked how many races I had run in the past few months. I did not know what to say because I had not run any. He told me he runs a race every two months and aims to break his personal record. Good for him. Honestly, give me a break. My only goal is just to finish.

Then I realized I had likely entered the wrong pace when I signed up for this race. That is why I ended up in a faster group. As expected, I quickly fell behind. But I at least finished the race without getting injured.

I am not a good runner. I am terrible in all dimensions: speed, distance, and form. I am not proud of it, but it is okay. I enjoy running because I am not good at it. You will never find me on Strava because I do not care about my performance. The only thing I track is whether I made it home on time.

In graduate school, I realized something early. I entered a graduate program because I liked research. But I also had to become very good at it if I wanted to do it professionally. That was stressful.

So I looked for an area where I did not need to be good but could still enjoy working hard. In the early years of the program, that was martial arts. Berkeley had a great martial arts program, probably one of the best in the country, so I tried several of them for about three years. The training was tough. Most members were undergraduates, and they were full of energy and very competitive. I was neither. We worked out a few days a week. I was terrible. But I did not need to be good, and that was a great feeling. I just needed to keep showing up and trying to improve.

Then I became a parent, and the pandemic hit. That combination made it hard to continue martial arts. I could no longer go to the studio regularly. So I became a runner because, logistically, it was the best choice.

I have been running consistently since then. It has been seven years. Wherever I go, including conferences, I pack my running shoes. I have even made friends through running.

I like running because I am terrible and slow. But I can still enjoy it. I think it is nice to have one thing you really love but are also bad at. It is a good reminder that you do not need to be good at everything.

In some areas of life, showing up is enough.

I am a slow runner, and I’m grateful for it.

Here is their distinction of these two objectives (67):

Where efficiency and its correlate optimality rest on the foundation of optimism, reliability is based on the realistic assumption of fallibility. The difference is not simply semantic. It is that which distinguishes a success-oriented from a failure-avoidance management.

Why does this distinction matter? A short answer is that there’s a trade-off between efficiency and reliability. As Landau (the same author) articulated beautifully in his 1969 Public Administration Review article titled “Redundancy, Rationality, and the Problem of Duplication and Overlap,” removing all redundancies in an organization is not only impossible, but also dangerous.

Imagine you go to an airport and take a flight to a faraway city. How can you be sure that the flight is safe? One reason is that the airport is designed as a redundant system. For instance, a passenger’s ID is checked multiple times. If it were checked only once, it might save time. It might also save resources by allowing the organization to downsize. We can do all of this in the name of streamlining, or more broadly, efficiency. But does that make the system and the process safer?

Jonathan Bendor (1950-2025) was a student of both Aaron Wildavsky, the founder of the Goldman School, and Martin Landau. He taught at Stanford’s Graduate School of Business and passed away last year. Bendor was one of the best interpreters of Landau’s work, and he argued that all organizations face the FDA’s problem (Bendor and Kumar 2005). Like the FDA, they want their products to be both effective and safe. Therefore, designing an effective organization involves a trade-off between efficiency and reliability, or creativity and criticism. They are the yin and yang of the organizational universe. It is not either-or. We need both and must balance them.

Beyond system design, there is also a fundamental reason to take this lesson seriously. An organization can only grow to the extent that it learns from its own experience. Otherwise, what it achieves may simply be luck. Achieving that consistently requires learning.

The question is where that learning comes from. Again, Landau offers an insight. He addressed this in his article titled “To Manage Is Not To Control: Or the Folly of Type II Errors,” co-authored with Russell Stout, Jr. and published in Public Administration Review in 1979. Learning comes from acknowledging that some failures are not only inevitable but also valuable. Denying them leads to the trap of Type II errors. Getting it right is not the only form of knowledge. Proving yourself wrong is also knowledge.

This discussion about human limitations and the value of recognizing them leads us to Karl Popper. Popper argued in Conjectures and Refutations (1963) that the main distinction between science and pseudoscience, which includes religion, is that science is fallible. Science works because it can be proven wrong.

What is a good scientific theory? A theory that is more falsifiable, or more testable. In practice, this implies that an organization that knows how to learn is one that knows how to be wrong. It should not only tolerate being wrong. It should actively seek out disconfirming evidence and potential failure.

This is also a key insight of Aaron Wildavsky, Bendor’s advisor and Landau’s colleague at Berkeley. Evaluation is the other side of implementation. It reflects the gap between intended and actual outcomes. We understand how implementation works by learning from its failures.

Now from Popper to AI, from the mid-20th century to the present.

We live in the age of AI, or at least that is what people say and what many organizations believe. But how can we build the future that we want using this powerful technology? That is the real question if we take human agency seriously, rather than technological determinism.

Most arguments on AI focus on streamlining or efficiency optimization. But there is another side. Faster and more efficient are not necessarily better. These systems must also be reliable. If we cannot overcome this challenge, AI’s societal impact will be limited. A system may work under specific conditions. It may look impressive as a demonstration. But that does not mean it will scale, or that we can manage its potential risks. A scalable solution must work across variation and under stress (List 2021).

Here are my three takeaways:

• Organizations exist because we are fallible. AI may reshape organizations, including those in the public sector, but this fact has not changed.

• Organizations that refuse to be wrong and learn from it cannot grow.

• Proving yourself wrong is powerful. Scalable solutions are reliable, not just efficient.

In the past few weeks, I have been thinking about how to end the course I taught this semester: The Politics of Public Policy. There is no single best way to do it. But I have come to believe there must be a kind of ritual that helps students remember what they learned and carry it forward with intention.

My ultimate goal is for them to take that knowledge into the world and do something impactful. But how can I do that?

Below is the ritual I created using Figma, a design collaboration tool widely used in the tech industry. I used it extensively for facilitation when I was a data scientist at Code for America. When I returned to academia, I decided to bring this into the classroom.

The Setup

The setup is simple (see figure below). There are two whiteboards. The left one is about what students learned (learning), and the right one is about what they will do next (acting).

Students do three things in order:

• (1) they post sticky notes on what they want to share,

• (2) they vote on what they want to highlight and discuss, and

• (3) each person shares what they wrote or what they voted for.

Figma is a useful tool to organize and run these activities smoothly. We first covered what we learned from class and then moved on to what we will do next.

Learning and Doing

As the saying goes, learning happens through doing. These two are deeply connected and are placed side by side intentionally.

Students spend the semester building knowledge and skills. They practice, struggle, and improve. I broke down their policy analysis assignment into two components: implementation analysis and political analysis. From the topic proposal, they went through prototyping and iteration. They received feedback from me several times and also reviewed and commented on each other’s work. The final assignment includes their best polished implementation and political analysis, along with their response letter to the feedback.

Why Grades Are Not Enough

At the end of the semester, they receive a final grade. But I told them, and explained why, that a grade (a proxy) is a weak signal. It does not convey what they can actually do. It is generally imprecise, and this problem is compounded by grade inflation in higher education.

What evaluators, those who screen applications and conduct interviews, want is concrete evidence. They want to see work and the person behind it. That is why I ask students to build a portfolio. AI tools now make this easier.

Collective Reflection

Before they build a portfolio, they need to map what they learned and articulate it clearly. Then they need to plan how to carry it forward into their next steps. That could be a summer internship, next semester’s courses, or job applications.

This is a collective problem. Students are in different years, but they face similar transitions. They can learn from each other. They can share strategies and ideas. Figma makes that collective reflection visible and interactive. What matters is not using Figma or any other tool, but going through this experience, a ritual, together.

Play the Long Game

The goal is not to do everything at once. Even small improvements compound. Learning by doing, over and over, is what makes the difference. To explain this idea, I ask students to remember three numbers: 70, 6, and 1. A 1% improvement does not look like much. But if it compounds, it doubles your output in about 70 years. This is why growth rates matter in macroeconomics and more broadly. If that same 1% improvement happens each month, it takes about six years. If it happens each week, it takes just over a year.

Two Goals of University Education

In my view, university education has two complementary goals. One is intellectual. We want students to stay curious and cultivate new habits of mind. We want them to have the capacity to tackle new challenges. The other is practical. They need to know how to apply what they learn.

For the latter, students can figure some of this out on their own, and they should. No one understands their own situation better than they do. But guidance matters. It reduces search costs and makes the process more effective.

Why I Do This

Again, this way of ending a course is not the only way. Also, I’m unsure whether this is the best way. Moreover, it’s not strictly necessary. I already covered the material. It is not required for their final assignment.

But I do it for a different reason.

I care for my students deeply. Not in the way parents or friends care. As an educator, I care about whether students flourish intellectually. I want them to remain curious, stay humble, and keep learning. That is why I assigned readings ranging from medieval Europe to artificial intelligence, and from the early American republic to the slums of India. At the same time, I want them to compete at their highest level. I want them to succeed in whatever field they choose. I pushed them to do better, with close attention to detail, while not losing the big picture.

That is how I designed the course and how I chose to end it.

Rising to the Challenge

This course is not easy. The readings, four papers a week from political science, sociology, economics, public administration, and public law, are close to a graduate-level workload, especially if we add one more paper each week. The same is true for the assignments, which are practical and set at a high bar, though students learn how to approach them incrementally. Most students are new to this. It is difficult. But they rose to the challenge. From the whiteboard exercise above, what I found very interesting and moving is that many students mentioned they enjoyed and learned from reading this wide range of materials. They also appreciated being challenged and rising to the bar. It was an honor to teach these students.

Looking Ahead

These students are young. Their journeys are just beginning. Not all of them will pursue public service. But in a world where government is blended with markets and civil society, many will still shape public policy in some way. When that happens, I want them to remember what they learned in this course, or at least carry its lessons forward. We discussed these career plans as part of the final class activity as well. I was very impressed and look forward to following and supporting their futures.

Teaching Is An Act of Care

Learning sets the upper bound of growth. What matters in the end is doing. You have to keep doing both. Teaching is an act of care. It sets that cycle in motion.

One point we emphasized, which is common wisdom in the data science field, is that breaking into the field requires experience and demonstrable skills. To show that effectively, you need a portfolio, much like designers do. It helps reduce uncertainty about whether a candidate can actually do the job and communicate their work. Both hard and soft skills matter.

That was three years ago, when I was working as a data scientist. Now I am back in academia. I teach a course on the politics of policy implementation in a policy program for students at all levels. In courses like this, the standard assignment is a policy memo or related analysis. I do something similar, although I try to be more systematic. I break the work into implementation analysis and political analysis, then combine them. I also incorporate in-class activities such as process mapping and power mapping of stakeholders on the chalkboard.

Even so, this still operates within familiar parameters. We have been teaching these courses in similar ways for years, perhaps since the 1970s when policy programs emerged across the United States and the world.

Recently, I have been thinking about how this process should and can change given the accessibility of AI tools. Agentic tools are now widespread and easy to use, whether through coding agents like Claude, Gemini, or Codex, or design platforms like Figma.12

These tools change what students can do. Instead of just writing and submitting assignments for evaluation, they can transform those assignments into something more usable and public-facing. A policy memo can become a website. It can be combined with data, extended with simple apps, or layered with interactive elements. Over time, these outputs can accumulate into a portfolio.

A portfolio approach has practical benefits while requiring much lower costs (efforts). By the time students apply for internships or jobs, they can show concrete work. Rather than listing skills on a resume, they can demonstrate what they can actually build and analyze. This allows others to evaluate their capabilities before the interview stage.

I am teaching my students this approach this semester. If I were a college student again, this is exactly what I would do. More broadly, I think policy programs need to reconsider how they design and teach their curricula. No one knows how the job market will change by the time students graduate, so there is a lot of uncertainty about how they can capture the value of their education. But there are ways to increase and amplify the value of their training and potential contributions.

None of this point denies that the foundation still matters. Students need to read widely and write clearly and critically. These are essential skills. In policy settings, decisions are often made by committees, so the ability to write effective memos, from one-pagers to longer briefs, remains crucial.

At the same time, students need to think more intentionally about how they present and share their work. A design-oriented approach to building a portfolio can help them gain exposure, connect with others, and expand their opportunities.

Fellow political scientists and policy scholars attending the upcoming APSA Annual Meeting in Boston, I have a favor to ask.

If your work focuses on how administrative rules and procedures, as parts of broader policy governance, function as hidden distributive or political tools, or how groups and individuals influence (or cannot influence) these policy processes, we would like to learn from you. If you do not work in these areas but know someone who does, please share this post with them. We hope to grow a community of scholars working on administrative politics, from administrative law to interest group politics to policy feedback to administrative burden.

I am Jae Yeon Kim, Assistant Professor of Public Policy at UNC–Chapel Hill. I am co-organizing this with Maraam A. Dwidar, Assistant Professor of Government at Georgetown. Our gathering is supported by Joe Soss, the inaugural Cowles Chair for the Study of Public Service at the University of Minnesota.

If you are interested in attending, please fill out this form. It takes less than one minute. Knowing how many people are interested will be extremely useful for organizing and planning logistics. Because this is an in-person event, there may be limits on attendance. In any case, we will keep you informed as plans develop based on the survey and the mailing list.

Best,

Jae

I think that one common misconception I’ve often seen among students is that policy change only comes from legislation, that is, passing a new law. But what we want, in the end, is not a law but an outcome, no matter how you define it. The gap between the law and the outcome is policy implementation.

When we focus on this problem in the U.S., I think there are two underappreciated levers: state and local governments and tech.

(1) State and local governments: My home field, political science, studies bureaucracy. But the focus is largely on the federal bureaucracy and its political control because, in addition to theoretical reasons, it is relatively difficult to obtain fine-grained, high-quality data on state and local governments. If you want to be productive, you save time and publish more. Nonetheless, if you think about who is doing implementation, in our fragmented federal system, where we have not only 50 states but 3,000+ counties and 8,000–9,000 local governments, we cannot ignore them. They are at the frontier. They can be either bottlenecks of innovation or vanguards of experimentation and change. In either case, we need to understand them (see Anzia 2022). This is particularly true for safety net programs, but also for criminal justice, education, housing, and many other domains, including artificial intelligence.

(2) Tech: Government data (e.g., Census data) are the original big data. Ergo, the technologies used to process these data are the ultimate legacy systems (e.g., COBOL, assembly) (see Wiggins and Jones 2024). Modernization has been a consistent trend in the public sector over the past few decades (see Pahlka 2023). AI is a new addition to this trend, but this underlying dynamic has not changed. So far, we have left this problem to IT departments. In practice, many government employees’ jobs involve hiring and managing vendors, so these systems become vendors’ jobs.

We do not really understand how these systems work, including both front-end interfaces such as websites and applications and back-end systems such as case management. But they have large consequences. As Lawrence Lessig mentioned decades ago (see Lessig 1999), “Code is law.” In a similar sense, tech is policy. Like law and budget, this layer is foundational to the modern state. If I may elaborate, this is how people experience policy. This is how government understands people. If we do not get tech in policy, we do not understand how implementation works. If we do not understand implementation, we cannot improve outcomes. Depending on how these systems are designed and deployed, we can either shift burdens to people, spending one hour on a benefit application and still not getting it, or to the state, simplifying the system so that it works within ten minutes but is still reliable.

Civic tech is a buzzword.

But this is why and how I think it matters.

It is not just a trend; it matters because it touches on this age-old, fundamental problem: how we can make policies work.

Panelists:
Beth Simone Noveck, Professor at Northeastern University, the first Chief AI Strategist for the State of New Jersey, and Founder of InnovateUS, expert in AI, public innovation, and state capacity

Ric Edinberg, Executive Director at Conversational City Lab, expert in civic innovation and human-centered design

Ted Enamorado, Associate Professor of Political Science at Washington University in St. Louis, expert in applied data science and AI methods

Emily Binet Royall, Assistant Director at Partners for Public Good, expert in AI procurement

The syllabus focuses on policy implementation and process, and how they interact with political institutions, organizations, groups, and their strategic decisions. I cover both classic and cutting-edge research from multiple disciplines (political science, economics, sociology, public administration, and public law).

For the graduate-level course, I also place strong emphasis on professionalization. The course includes opportunities to write and receive feedback on research memos, a research proposal, and a full paper. I am old-fashioned, so you will complete two chalkboard exercises and one mock conference presentation. You will also write a review report and a corresponding letter.

My goal is to help students produce a publishable paper by the end of the semester. Any comments or suggestions are welcome.

If you are a UNC Chapel Hill faculty member and have students who would benefit from this course, please consider recommending it. It is not an easy course, but it offers strong theoretical and practical training.

Good research questions are not always obvious. Here are three criteria I tend to use, based on my own taste and experience.

1. Pick a topic that most people find not “sexy,” but that is fundamentally important. These topics are often boring, so there is less competition because they draw little attention, but they remain important even if trends change. They are, however, hard to identify unless you engage deeply with the field. My recent obsession with procurement is one example. One of the biggest lessons I have learned from my applied work is how much procurement matters for policy implementation, including in the context of public sector AI. Many questions about the distributional and political consequences of administrative procedures (my main research agenda) share this feature.

2. Find a way to study this understudied but important topic in a more empirically rigorous way. I am always curious about new empirical methods for this reason. Perhaps the empirical problem I cannot solve in this area has already been addressed in another field. This does not mean theory is unimportant. It is quite the opposite. I believe almost everything is theory-laden. Nonetheless, the discovery process is iterative, and we need to acknowledge that. As we get deeper into the data, and more broadly immersed in the field, we often understand the subject better and find a better way to conceptualize the problem. I believe that good theory and good empirics complement each other.

3. Find a way to communicate the problem, method, and evidence in a way that is accessible and convincing. This requires honest and frequent feedback. A recurring (career) lesson is that the sequence of discovery should not be the sequence of presentation. You do not need to write a paper in the order in which you reached the findings. Many people think reframing a paper is a waste of time. I agree to some extent, especially when reviewers make what I see as unreasonable requests, particularly on data analysis. But on framing and organization, I am almost always inclined to agree with reviewers (and readers), as they are often better judges of this problem. An author, including me, suffers from the curse of knowledge. We often work on a problem for so long that some assumptions are taken for granted, or some new approaches are not obvious.

The way I think about this is to focus on your competitive edge. It is important to stay up to date with new tools and understand how to use them properly. The only way to do that is through trial and error. I encourage students to learn by actually using AI and figuring out how to use it effectively. In my course, I do not prohibit the use of AI. However, students are responsible for whatever they submit. If their work includes hallucinated information, they do not receive credit. If they misattribute citations, for example by confusing two papers by the same author, that is a serious mistake and they receive only partial credit. These standards are not about discouraging AI use, but about maintaining accountability. If your name is on it, you’re responsible.

That said, using AI effectively is not what makes someone truly competitive, especially as it becomes a commodity. You need to think about what becomes scarce.

First, think about focus. Even before AI, many people are attached to screens, and I feel guilty too. We spend so much time scrolling and lose the capacity to do deep, focused work. As Cal Newport argues, developing the ability to concentrate and do sustained, careful work (what he calls “deep work”) is becoming more important. In practice, this means learning how to block your time and do one thing in a highly focused way.

For example, I block off a couple of days each week for focused work. I might go out for lunch, then spend time at a cafe with only a book, and read one or two chapters (or two to three papers). You can also set aside a couple of blocks each week for writing (or programming or designing). When you are drafting, it helps to work in an environment without distractions. You can use AI later for proofreading or editing, but the initial thinking and writing should come from focused effort. (I still handwrite an outline.)

Second, think about connectivity. It is also important to build skills that are harder to automate, such as working in teams, collaborating, and building partnerships. Much of what I do now involves talking to people, doing interviews, and understanding how others, especially practitioners and ordinary people, see problems. This helps me identify where my own knowledge is limited and where my skills can be useful. That kind of connection is fundamentally human and remains important.

Creativity, the essence of human capacity, at least as I understand it, comes from being deeply immersed in a problem. It involves paying close attention to details and working through them carefully. It is also about figuring out how to test, build, and scale ideas into real projects by working with others.

Finally, think about standards. AI can help you produce something acceptable very quickly. But if you want to be a professional or an expert, “good enough” is not the goal. That level of output is useful for exploration or prototyping, but not for final work. High-quality outcomes require iteration, feedback, and persistence. They require time and care. That is not something you can achieve by solely relying on AI. It requires your expertise, which comes from consistent deep work and experience working with others to achieve common goals.

In one interview, Caro explains why he writes these biographies. He also discusses this in his book Working, which is one of my all-time-favorites. His goal is to show how power actually works in the real world, not in a textbook. Robert Moses is a good case because, based on his formal title and position, he was not supposed to have much power. But he did. That is how he helped build one of the greatest cities in the world, while at the same time destroying and sacrificing many lives. This knowledge is crucial. If people understand how power operates, they can better understand how policy is made and implemented. That understanding helps citizens become more informed and active participants in a democracy.

I found this deeply inspiring because it reflects the motivation behind my own research and teaching (as a political scientist who studies public policy). In my courses on the politics of public policy, I try to help students understand how power works in practice. I ask them to read widely and deeply in the liberal arts tradition, and to produce both implementation analysis and political analysis. By the end of the semester, they combine these perspectives and analyze the full policy process.

So far, my pedagogical approach has been influenced by two sources. One is David Perkins’s book Making Learning Whole. The idea is that students learn the components of a field, but eventually must learn how to “play the whole game.” Think about basketball practice. You need to learn how to throw, catch, and hit the ball. But that is not how the game is played. You must learn how to play the whole game. The other influence comes from Francis Su’s book Mathematics for Human Flourishing. He argues that the core of education is cultivating habits of mind that allow people to tackle new problems. Ironically, the most practical education is often the most foundational one.

After watching the film, I realized that I have begun to develop a third view of pedagogy. I have always liked Susan Athey’s explanation of how she approaches applied problems. She focuses on solving not just a particular instance, but “a class of problems.” I have been thinking for a long time about how to teach this mindset to my students. This semester has been one of those experiments.

While listening to how Caro described writing his biographies, I realized that I had been trying to do something similar, though not consciously. Caro wanted to analyze power, which is a class of problems. To do that, he selected very specific cases, such as Robert Moses and Lyndon B. Johnson, and studied them with extraordinary depth, turning every page and not assuming anything, a skill and mindset he developed during his investigative journalism career. This is where the title of the documentary comes from. The goal is not simply to explain those cases. The goal is to reveal the underlying mechanism of how power works. The case thereby becomes a window into a larger set of problems.

My students are going through a similar intellectual journey with their assignments this semester. They need to go “really” deep into their cases. I encourage them to become obsessed with the details of policy implementation. At the same time, I push them to identify how the specific case they study reflects broader problems that share similar challenges and possible solutions. I hope they develop this mindset about public problem solving and carry it forward as their careers advance.

I want them to understand how power works, challenge and negotiate it when necessary, and amass and exercise it when needed. This is also what Hahrie Han and I argued in our piece published in The ANNALS of the American Academy of Political and Social Science a few years ago.

In these challenging times, we need all the talent (policy entrepreneurs) we can get to address a wide range of public policy problems. Teaching at a university is a privilege because it is an opportunity to help develop the next generation of leaders.