The book is a relevant addition to current debates on AI and digital governance. This volume takes a distinct approach. It views AI as an administrative change project and examines the internal changes within public institutions as AI becomes integrated into routine work. The cases illustrate how public organizations reorganize work, manage risk, and protect legitimacy as automation becomes part of routine processes. National libraries provide a valuable perspective as they exist at the crossroads of public value, legal limitations, and public trust, all while facing pressures to enhance the accessibility of extensive collections. The synthesis of cases is based on 90 interviews with specialists in libraries and AI across 12 instances, including 10 national libraries from Europe, as well as the British Library and the Library of Congress. Produced as an output of the EU-funded LibrarIN project, the book uses a digital co-production framework that keeps attention on the multi-actor nature of implementation. This model is illustrated through six stages: co-commissioning, co-design, co-implementation, co-delivery, co-assessment, and public value creation. I found this framework compelling as a disciplined scaffold that makes the cases easier to compare. Its novelty stems from the clear and repeatable application of co-production principles to AI projects. The 90-interview, multi-case design is well suited to the book's implementation claims, although the co-production framework is applied more rigorously in some chapters than others. For readers, the key question is whether this framework bridges classic public administration concerns with the distinctive challenges of algorithmic governance. It does so best when it clarifies who owns risk, who can contest errors, and how accountability is documented across actors. At the same time, the framework could foreground power imbalances more explicitly, since co-production can become co-dependence when core infrastructure, tools, or expertise sit with large technology vendors. Across 12 cases in the European Union and the United States, AI adoption is shaped not only by managerial goals but also by national policies, institutional capacity, and legal regimes. Many European libraries operate under stringent GDPR and are preparing for compliance with the EU AI Act, which pushes organizations toward documentation, risk management, and accountability practices. By contrast, the Library of Congress case highlights a different constraint set, where copyright governance, contracting, and project-level oversight do much of the regulatory work. This makes the American response less “regime-driven” and more project-specific, which helps explain why documentation and accountability routines evolve differently across the Atlantic. In comparative governance terms, this contrast echoes classic accounts of Europe's regulatory-state orientation versus the more adversarial, case-by-case style often associated with U.S. oversight (Kagan 2019; Majone 1994). The cases also point to a policy gap: while libraries are expected to provide AI services that adhere to ethical standards, they are often not explicitly included in national AI strategies, especially when funding is limited. The comparative significance of the book between the EU and the USA becomes most apparent when it illustrates how these variations lead administrative organizations to adopt different governance practices, even when the technical objectives appear to be alike. Three cases illustrate how these dynamics play out in practice. First, the Denmark case (the postcard digitization project) shows co-production in a direct, observable way. The project uses a commercial vendor tool to generate metadata, and the public value is straightforward: improved access, search, and discovery across an extensive collection. Yet the administrative story is the real point. Legal consultation becomes part of the implementation team because copyright and privacy questions shape what data can be processed, how, and at what scale. Users also appear in the co-delivery and co-assessment logic, not as symbolic “participants,” but as real end users who test whether access and search actually work. The case captures an important lesson for the digital state: external capacity can accelerate implementation, but it also increases dependence and legal exposure. Second, the Germany case on automated subject cataloging shows how AI changes internal administrative structure. The project uses open-source tools and heavy librarian involvement; the implicit message is that AI is best treated as a colleague rather than a threat. Human review is not framed as a failure. It is a quality control and legitimacy mechanism. The case also makes the workforce issue concrete. Skills, patience, and willingness to experiment are described as core to success, and the project becomes more than a pilot. It evolves into an organizational unit that collaborates across the library. The broader governance lesson is that implementation is not a sprint, but a marathon. It requires a durable internal home, not only a project plan. Third, the Netherlands “Retrotool” case links AI investment to a broader political context, including concerns about administrative error following the Dutch childcare benefits scandal. One might read the project, beyond the chapter's straightforward account of an automated cataloging initiative, as aligning with legitimacy concerns, where documentation, transparency, and quality controls help demonstrate responsible administration. Across cases, the book clearly discusses the risks related to the digital state. These include algorithmic bias, data privacy, copyright limits, harmful language in heritage collections, and transparency. Several chapters question whether users should be informed when metadata is generated by machines and whether confidence levels should be disclosed. These are governance choices that affect trust and accountability because they shape what institutions are willing to disclose, how they justify automated decisions, and what kinds of errors are considered acceptable. The final section, including the recommendations chapter and the LibrarIN toolkit, is the book's most policy-relevant contribution. The recommendations emphasize technical independence from large international technology vendors while staying engaged with them, keeping humans in the loop, building collaborative governance across departments, and treating AI as an ongoing transformation rather than a one-off pilot. The diagnosis is convincing, but feasibility is the hard part. Public managers face procurement rules and short budgeting cycles that reward pilots more than maintenance, monitoring, and staff development. The recommendation is realistic as a diagnosis; it is harder as a prescription unless funders accept long-run operating costs as part of “innovation.” A missing piece is a clearer political pathway: how public managers can build coalitions with finance offices, procurement units, and oversight actors to shift AI from grant-funded pilots into base budgets, shared services, and durable governance routines. The toolkit moves from lessons to usable infrastructure. It provides benchmarking tools (a policy tracker), comparative cases (a “what works” database), and an evidence base that supports policy learning and advocacy. I found this approach unusually practical. It lowers the cost of learning for managers and policy staff, and it supports diffusion without pretending that one country's solution can be copied whole. It also reinforces the book's core message: AI in public institutions is a governance and organizational problem first, and a technology problem second. “AI Innovations in Public Services” provides more than just an overview of technology. It offers a scholarly perspective on how administration is evolving in the digital state. Its comparative approach highlights how context and law shape AI adoption. Chapters are generally accessible, supported by consistent headings, interview excerpts, and appendix summary tables, although depth varies across an edited volume. Compared with recent work that foregrounds algorithmization and ethical or legal risks at a more conceptual level (Henman 2020; Meijer et al. 2021), this book stands out for its cross-national implementation detail and its practical attention to organizational change. Scholars of public management, digital governance, and state organization will find a clear framework and important evidence on how institutions manage transformation. Practitioners will receive trustworthy advice on how to adopt AI while keeping their legitimacy and public value intact. A main limitation is that outcomes are discussed more than they are measured across cases. Future work would benefit from clearer outcome indicators, such as documented changes in administrative efficiency, error rates and cataloging accuracy, time and cost savings, and longitudinal evidence on user trust and perceived transparency, meaning repeated measures over time, that can show whether disclosure practices and accountability routines actually sustain public confidence. These measures would help ground claims about public value and legitimacy and strengthen the policy implications of this important comparative project. Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.
Salman Bin Habib (Fri,) studied this question.