Blog · Review Essay · Last reviewed June 25, 2026

The Technological Republic and the State as Software Customer

Alexander C. Karp and Nicholas W. Zamiska's The Technological Republic is a manifesto from inside Palantir's political imagination. It argues that Silicon Valley has spent too much talent on consumer convenience and advertising markets while the democratic state has lost the technical ambition needed for an age of AI, drones, cyber conflict, and strategic rivalry.

For this review, the technological republic is not a slogan for patriotic software. It is a public-capacity test: whether democratic institutions can understand, build, buy, audit, constrain, and exit the technical systems that now mediate security, administration, logistics, intelligence, and force without letting vendors define the mission, the records, and the limits of oversight.

The recurring concern is concrete. When an interface becomes an institution's operational picture, belief and authority begin to follow the records that interface can produce. The question is whether public judgment remains strong enough to govern the picture, or whether the state has rented the means by which it knows what is happening.

The Book

The Technological Republic: Hard Power, Soft Belief, and the Future of the West was published by Crown Currency on February 18, 2025. Penguin Random House lists the hardcover at 320 pages with ISBN 9780593798690, and identifies the authors as Palantir co-founder and CEO Alexander C. Karp and Nicholas W. Zamiska, head of corporate affairs and legal counsel to Palantir's CEO office. Palantir's investor materials separately identify Karp as co-founder, CEO, and director.

The book is not a neutral survey of technology policy. It is an argument from a company whose business depends on public institutions, security agencies, defense work, data integration, and operational software. That origin matters. The book's strength is that it refuses the fantasy that software floats above geopolitics. Its risk is that it can make one company's preferred public-private settlement sound like civic philosophy.

That does not make the book disposable. It makes it evidence of the moment it describes: elite technical builders now want to be understood as civic actors, not only as vendors. The analytical question is whether their civic language is accompanied by institutional arrangements that let the public inspect, refuse, replace, and remember what those builders create.

It helps to be specific without turning the review into a company dossier. Palantir's 2025 Form 10-K says the company has four principal software platforms: Gotham, Foundry, Apollo, and AIP. It describes Foundry as a data-operations platform, AIP as a generative-AI platform for agents, automations, applications, and production evaluations, Gotham as software for allied defense and intelligence operations, and Apollo as continuous delivery infrastructure. The same filing says 54 percent of 2025 revenue came from government customers. When Karp and Zamiska argue that elite software builders should work on state power, they are not writing from outside the market they praise. They are arguing from inside it.

It belongs beside The Tech Coup, Surveillance Valley, The Whale and the Reactor, Power and Progress, and The Coming Wave. Each asks whether technical capacity is being governed by public purpose, private incentive, emergency, or myth.

Current Context

As of June 25, 2026, the book lands inside active procurement and defense-governance change, not an abstract culture debate. DoD Directive 3000.09, updated January 25, 2023, requires autonomous and semi-autonomous weapon systems to be designed for appropriate human judgment, rigorous verification, validation, testing, transparent and auditable data sources and methodologies, understandable interfaces, and activation or deactivation controls. That matters because the book's call for software to serve hard power now meets a live rulebook for what force-related autonomy must preserve.

The civilian side is equally concrete. OMB memoranda M-25-21 and M-25-22, issued April 3, 2025, put agency AI governance and acquisition discipline around federal AI use: Chief AI Officers, governance boards, high-impact AI review, cross-functional acquisition teams, privacy, civil-rights and civil-liberties risk, IP rights, data rights, vendor lock-in, ongoing testing, and monitoring. OMB M-26-04, issued December 11, 2025, adds a content-governance layer for federal LLM procurement by requiring OMB-defined "Unbiased AI Principles," including truth-seeking and ideological neutrality, in covered contracts. That memo should be read as procurement policy, not as proof that neutrality has a settled technical definition.

GSA's Buy AI guidance, last updated May 11, 2026, tells agencies to start from mission needs, test before large purchases, manage data, involve key officials, and monitor costs. GAO's April 13, 2026 report found that selected agencies were not yet systematically collecting lessons learned from AI acquisitions. The republic at issue is being written in solicitations, contract clauses, inventories, test plans, audit trails, and exit rights.

The hard-power question is therefore not simply whether the state should use advanced software. It is where the public control plane sits. If operational awareness, model access, task routing, identity permissions, logs, updates, and audit evidence all live inside a private platform, then sovereignty depends on contract terms, records architecture, technical staff, and the authority to stop a workflow during crisis.

The Diagnosis

Karp and Zamiska's central complaint is that the culture of technology has become too small for the historical moment. The publisher frames the book as an indictment of Western complacency and a call for software companies to work on urgent public problems, including the AI arms race. Kirkus summarizes the thesis similarly: Silicon Valley, in this telling, has drifted toward consumer toys, advertising systems, and libertarian self-regard while government security institutions remain bureaucratic, slow, and difficult for software builders to work with.

That diagnosis lands because there is an obvious mismatch between the scale of AI-era problems and the incentives of much product culture. Recommender feeds, engagement markets, growth loops, and enterprise dashboards have trained elite engineering talent to optimize measurable behavior inside privately owned systems. The authors are right that the most consequential technical questions now involve war, infrastructure, energy, public administration, education, health, democratic resilience, and state capacity.

The stronger form of the argument is not "build weapons instead of apps." It is that a society's highest-status technical institutions reveal what the society believes is worth building. If every capable system is pulled toward consumer capture, ad targeting, speculative finance, or platform lock-in, then public purpose has already lost before policy debate begins.

But public purpose is not restored by moving the same incentives into national-security or public-administration markets. A society can reject triviality and still create systems that are opaque, coercive, or dependent on private operational doctrine. The missing term is public capacity: the ability to set the question, inspect the evidence, keep the records, defend rights, and refuse a system that performs well for the vendor but badly for the polity.

Soft Belief, Hard Power

The subtitle is the most interesting part of the book. "Hard power" names defense, deterrence, cyber capability, industrial capacity, and the material systems that let a state act. "Soft belief" names the cultural confidence, shared purpose, institutional trust, and willingness to defend a political order. The book is really about their feedback loop.

This is where it becomes useful for thinking about belief formation. Technical systems do not only execute decisions. They give institutions a picture of what is happening, what can be acted on, and which futures are plausible. A defense platform, intelligence system, procurement dashboard, emergency-response model, or border database can become an epistemic machine. It tells the state what it can see and therefore what it thinks it can responsibly do.

The authors want democratic societies to recover ambition. But ambition is not self-validating. A technical culture can believe in public purpose and still build systems that compress people into targets, risks, anomalies, suspects, customers, scores, or optimization surfaces. Civic seriousness requires more than mission language. It requires forms of inspection, refusal, appeal, and memory strong enough to discipline the mission.

That is the book's useful pressure point. Soft belief cannot merely mean morale among builders. It has to include the public's ability to know how power is being exercised in its name. Hard power without inspectable memory becomes command by interface.

The State as Customer

The book's most practical demand is a revived partnership between government and the software industry. The Washington Post review captures the blunt version: Karp and Zamiska want Silicon Valley engineers to work more closely with the Pentagon and other state institutions, especially as AI expands the range of military and security applications.

That demand should be taken seriously and skeptically at the same time. Democracies do need technical competence inside the state. Public agencies cannot govern AI, cyber operations, procurement, surveillance, benefits administration, health infrastructure, or battlefield software if they are dependent on vendors for basic comprehension. A technically hollow state becomes either slow and performative or captured by whoever can sell it an interface.

But a state that revives capacity mainly by becoming a better customer for private platforms has not solved the sovereignty problem. It has outsourced cognition with better branding. The crucial question is whether public institutions can build durable internal expertise, own critical records and audit trails, maintain exit options, and enforce rights against the very systems they buy.

The phrase "state as customer" hides the constitutional stakes. A customer evaluates features, price, speed, and support. A democratic sovereign also needs lawful authority, public records, appeal paths, legislative and judicial access, and an emergency brake. In high-consequence systems, contract terms decide whether the state can inspect evidence, preserve logs, change vendors, tell affected people what happened, or reconstruct a failure after classification, trade-secret, and security claims arrive.

A useful procurement question is brutal: can the agency still explain, audit, pause, migrate, and defend the workflow if the vendor changes terms, updates the model, refuses disclosure, loses staff, suffers a breach, or leaves the market? If not, the system is not state capacity. It is leased dependency with public authority attached.

This is the tension between The Technological Republic and The Tech Coup. Karp and Zamiska fear a soft, underpowered state that cannot act. Marietje Schaake fears democratic authority being transferred to firms that act without democratic constraint. Both fears can be true. A capable public sector that cannot say no is not democratic capacity. A rights-conscious public sector that cannot operate is not capacity either.

The AI-Age Reading

Read in 2026, the book is less a general technology argument than a document of the AI-security turn. Frontier models, autonomous systems, cyber operations, military targeting support, intelligence analysis, synthetic media, drone warfare, data centers, and compute supply chains have made it harder to keep "technology policy" separate from national power.

The book's warning is useful against a complacent consumer model of AI. General-purpose models are not only writing tools and customer-service engines. They are also planning aids, code generators, persuasion surfaces, data-analysis layers, weapons-system inputs, bureaucratic accelerants, and new ways for organizations to see the world through software.

The danger is that AI militarization can turn every institutional problem into an emergency competition. Once a system is framed as part of a civilizational race, ordinary democratic friction starts to look irresponsible. Procurement speed can override public deliberation. Oversight can be dismissed as naivete. Worker objection can be recoded as decadence. Technical ambition can become a permission structure for secrecy.

Agentic systems intensify the problem because they reduce the distance between perception and action. A model that summarizes reports is one kind of risk; a model-connected workflow that drafts plans, calls tools, updates records, routes tasks, or recommends operational action is another. In both civilian and military settings, the human can be formally present while the real decision architecture has moved into prompts, permissions, defaults, rankings, and logs that only a vendor or small technical office can inspect.

The responsible reading is therefore double. The state needs technical competence because AI changes what power can do. The state also needs stronger democratic constraint because AI changes how quietly power can do it.

Dual-use language should not blur those boundaries. The same model class might summarize benefits calls, triage cyber alerts, draft acquisition language, analyze drone imagery, or support targeting review, but the governance file cannot be generic. It has to attach to the specific mission, data, tool permissions, operator role, affected population, and failure mode.

Governance and Safety

The governance question is not whether democracies should have capable software. They should. The question is whether public institutions keep enough independent capacity to govern the systems that give them speed, visibility, and force.

Start with an authority map. A serious public-sector AI deployment should name the mission owner, data steward, model or platform vendor, system integrator, cloud host, evaluator, legal reviewer, operator, affected public, pause authority, record owner, and incident owner. If those roles are unnamed, accountability disappears into the stack.

The minimum standard is a public-capacity test. Before deployment, the agency should know what record the system may consult, what model or platform version is in use, what tools it can call, what the vendor can change without notice, where logs and prompts are retained, who can inspect them, how affected people or oversight bodies can contest an outcome, and how the agency exits without losing case history or operational knowledge.

A responsible national-security or public-sector AI stack needs controls at several layers: mission limits, lawful authority, data provenance, model and system evaluation, cybersecurity, human command responsibility, records for after-action review, procurement terms that preserve audit rights and logs, portability, exit, incident reporting, and independent oversight strong enough to pause a tool whose outputs are becoming policy.

Procurement should convert mission language into enforceable duties: model and system documentation, component inventories, evaluation artifacts, data-provenance records, logging, audit access, independent testing, version rollback, change notification, incident reporting, subcontractor disclosure, data-use limits, deletion and export rights, portability, and termination paths. Human oversight should mean authority to slow, refuse, override, escalate, and document dissent, not merely a person looking at a dashboard.

For defense systems, that evidence should look like a safety case tied to an operational concept, legal review, training, rules of engagement, testing under realistic conditions, and after-action records. For civilian systems, it should look like a public administration file: intended purpose, affected population, lawful basis, appeal path, records schedule, accessibility review, civil-rights review, privacy review, labor impact, complaint channel, and decommissioning plan. The stakes differ, but the governance pattern is the same: no authority without records, no automation without responsibility, no vendor dependency without exit.

A useful safety case has two ledgers. The capability ledger says what the system can do, under what conditions, with what evaluation evidence, and how performance changes after updates. The constraint ledger says what the system is forbidden to do, who can override or halt it, what uncertainty must be shown to operators, what evidence survives for review, and what remedy exists when the system misroutes power.

Defense sources already point in this direction. DoD Directive 3000.09 requires autonomous and semi-autonomous weapon systems to be designed so commanders and operators can exercise appropriate levels of human judgment over the use of force, and it ties deployment to verification, validation, testing, transparent interfaces, monitoring, and legal review. The Department's Responsible AI Strategy and Implementation Pathway frames military AI around lawful, ethical, responsible, accountable development and use; its public CDAO summary emphasizes testing standards, accountability checks, acquisition lifecycle, requirements validation, human systems integration, and safety. The U.S.-led political declaration on responsible military AI and autonomy, summarized by the Defense Department in 2023, adds auditable systems, explicit and well-defined uses, lifecycle testing, unintended-behavior controls, and senior review for high-consequence applications.

For civilian agencies, procurement is the control point. OMB M-25-21 and M-25-22 put governance and acquisition discipline around federal AI use and procurement, including Chief AI Officers, cross-functional acquisition teams, privacy, civil-rights and civil-liberties review, data rights, IP rights, and vendor lock-in. M-26-04 shows that even model content and political neutrality claims now enter acquisition terms for covered federal LLMs. GSA's Buy AI guidance tells agencies to start from mission needs, test before large-scale purchases, manage data, involve security, privacy, and AI officials, and monitor costs. GAO's 2026 AI-acquisitions report makes the same operational point: acquisition rules matter because performance, risk, data rights, portability, and contract exit often decide whether an AI system remains governable after award.

NIST's AI Risk Management Framework supplies a useful non-statutory grammar: govern, map, measure, and manage risk throughout the lifecycle. For this book's argument, the hard part is not naming those verbs. It is assigning leverage: who owns the records, who can audit the system, who can pause deployment, who can compel a vendor to explain a material change, and who can help an affected person challenge an outcome.

Where the Book Needs Friction

The book's biggest weakness is that it treats public purpose as if it can be stabilized by seriousness, patriotism, and elite technical ambition. Boston Review's critique presses the obvious political problem: a call for Silicon Valley to join the national-security state can look less like democratic renewal than a more centralized machinery of repression and violence. The Independent Review raises a different objection, warning that the book underplays criticisms of top-down control and central planning.

Those critiques matter because the book is not only about tools. It is about who gets to define the common good when the tools are powerful. A public-private AI stack can defend democracy, surveil dissent, accelerate warfare, improve disaster response, ration benefits, classify migrants, monitor workers, or make bureaucracies more humane. The same vocabulary of urgency and capability can travel across all of those uses.

Hard-power realism can also become a shield for opacity. Classification, procurement urgency, cybersecurity sensitivity, and proprietary software can each be legitimate in context. Together, they can make the most consequential systems the least inspectable. A technological republic worth defending has to build controlled forms of public memory even where full disclosure is impossible.

The book also underdefines democratic disagreement. It is easy to praise seriousness when the mission is already chosen. It is harder to say who can challenge the mission, slow the deployment, demand evidence, protect dissenting workers, or represent people turned into data objects by a security system. Builder seriousness is not a substitute for political legitimacy.

The authors are most persuasive when they attack triviality and institutional drift. They are less persuasive when the remedy depends on trust in the builders' seriousness. Democratic societies should not have to choose between shallow consumer platforms and security-state solutionism. They need procurement competence, public-interest engineering, labor voice, civil-liberties enforcement, independent auditing, adversarial journalism, congressional and judicial expertise, and technical institutions that can survive vendor pressure.

What This Changes

The lasting value of The Technological Republic is that it makes software governance inseparable from political theology: what a society worships, fears, funds, protects, and treats as worth building.

If a state cannot understand its own systems, it becomes dependent on private interpreters. If a company sells the state its operational picture, the company's categories can become public reality. If AI systems mediate intelligence, security, logistics, public services, and warfare, then the question is not simply whether the models work. It is whether the society has retained enough independent judgment to decide what work should be done.

The book should be read as a challenge, not a destination. It usefully asks technical culture to stop pretending that apolitical convenience is enough. The counter-question is just as important: when technology returns to the state, does it return as democratic capacity or as privately operated command infrastructure?

A serious technological republic would not merely recruit engineers into national projects. It would make power legible upward and downward. It would keep public expertise outside vendor capture, protect refusal and dissent, preserve audit trails, put rights in the path of deployment, and remember that the common good is not whatever the most capable system can optimize.

The practical test is whether a public institution can still think after the interface is gone. If the records, categories, workflows, permissions, and institutional memory can only be reconstructed through a vendor's platform, the state has not recovered civic ambition. It has purchased a stronger dependency.

Source Discipline

This review separates book metadata, company self-description, public filings, defense-policy sources, procurement guidance, standards guidance, and critical reception. Penguin Random House supports book facts and author descriptions. Palantir investor pages and SEC filings support company roles, platform descriptions, revenue segmentation, and customer categories; they do not independently prove that any particular deployment is effective, lawful, or democratically governed. DoD, OMB, GSA, GAO, and NIST sources support the governance context; they do not establish that any specific vendor or program satisfies those controls.

Reviews from Kirkus, The Washington Post, Boston Review, and The Independent Review are used for reception and critique, not for legal or procurement facts. The policy claims come from primary public records wherever possible because the subject of the article is itself the danger of letting vendors, reviewers, or manifestos substitute for inspectable evidence.

OMB M-26-04 is cited here as federal procurement policy for covered LLMs, not as a neutral scientific settlement of truth, bias, or ideology. Its importance for this review is that content behavior, documentation, evaluation, and contract remedies are now part of the public software bargain.

Claims about defense AI and government software should name the mission, legal authority, contract vehicle, data sources, audit trail, human role, failure mode, affected population, and review date. A vendor demo, patriotic mission statement, agency press release, or book manifesto is not enough to establish public value.

This review makes no claim that any AI system is conscious, divine, or AGI. It treats models, platforms, command systems, and agents as institutional software that routes perception, authorization, and accountability through human-made organizations.

Sources

Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.

Buy on Amazon Browse Books

Return to Blog · Return to Books