Chapter 11 — The Executive Branch in Action: Bureaucracy, Agencies, and the Administrative State

"The most important political development of the last hundred years has been the rise of the administrative state — the apparatus of agencies that write the rules, enforce the rules, and adjudicate disputes about the rules. Whether you celebrate it or fear it, you have to understand it." — paraphrased from a graduate-level lecture in administrative law

Where we are in the book

Chapter 9 introduced the presidency as constitutional office and political institution. Chapter 10 covered the vice presidency, the Cabinet, and the Executive Office of the President — the small apparatus directly around the President. This chapter now asks what is, by any honest measure, the larger question: what about the rest of the executive branch?

Roughly 2.1 million civilian federal employees. Roughly 1.4 million active-duty military. 15 Cabinet departments, ~60 independent agencies, ~50 government corporations, ~40 boards and commissions. A federal budget of more than $6 trillion in fiscal year 2025. Rules running to roughly 80,000 pages a year in the Federal Register. A "shadow government" of perhaps 5 million federal contractors and 3 million federal-grant employees working alongside.

This is the bureaucracy, the administrative state, the "fourth branch." It writes the regulations that determine what your employer must pay you, what your car must do in a crash, how your bank must verify your identity, and what your insurance must cover. It administers Social Security, Medicare, food stamps, veterans' benefits, federal student loans, and unemployment insurance. It runs prisons, parks, weather forecasting, the air-traffic-control system, and the nuclear arsenal. Most of what people experience as "the federal government" — outside of an election year — comes from this apparatus.

How should we think about it? That is the question this chapter takes seriously.

11.1 The "Fourth Branch" question

Open the Constitution. Article I creates Congress. Article II creates the President — singular — and "the executive Power" is "vested" in him. Article III creates "one supreme Court, and . . . such inferior Courts as the Congress may from time to time ordain and establish."

Three branches. That is it. There is no fourth article.

And yet by the early twentieth century, and emphatically by the New Deal, an apparatus had grown that did not look like any of the three. It was technically inside the executive branch — the President is, formally, the boss of the Environmental Protection Agency, the Securities and Exchange Commission, and so on. But practically, this apparatus operated with a degree of independence from the President, a depth of technical expertise, a permanence of personnel, and a power to make binding rules that no purely "subordinate" body would have. Critics began to call it the fourth branch of government.

The "fourth branch" label is partly a description and partly a critique. As description: the federal bureaucracy makes binding national policy, on its own initiative, with limited day-to-day political control. As critique: this is a constitutional problem, because the Constitution did not authorize a fourth branch and did not contemplate that the binding rules of national life would come from agencies rather than from the elected Congress.

This chapter will not adjudicate the "fourth branch" question. It cannot — that is a question of constitutional interpretation about which serious scholars genuinely disagree. What it can do is lay out the strongest version of each side of the debate.

The "fourth branch is unconstitutional" position (steel-manned)

The Constitution vests the executive power in one President. That phrasing is deliberate. The Framers had just fought a revolution against a king and were, on the whole, suspicious of executive power — but they were also suspicious of plural executives, having seen the problems with multi-headed leadership during the Confederation. They settled on a single elected President accountable to the people for the entire executive function.

Independent commissions whose members the President cannot fire at will, agencies whose career staff effectively run policy regardless of who wins elections, and rule-making powers exercised by employees who never face an election together amount to a de facto fourth branch — one the Founders did not authorize and that severs the connection between voters and the rules that govern them. When the EPA writes a rule that closes coal-fired power plants, the voters have no direct mechanism to reverse that rule short of electing a President who will stack the agency with different appointees, who will then run a multi-year process to repeal the rule, which will then be litigated for years more. By the time the voters' wishes are translated into law, an entire administration has come and gone.

The proposed remedy: restore presidential control of the executive branch (the unitary-executive theory), restore judicial review of agency statutory interpretation (overrule Chevron, which the Court did in Loper Bright), restore the non-delegation doctrine so that Congress cannot hand off open-ended legislative authority to agencies, and require Congress to make the major policy calls itself rather than punting them to bureaucrats.

This position has serious adherents — Justice Antonin Scalia in his later years, Justice Clarence Thomas, Justice Neil Gorsuch, scholars like Philip Hamburger (Is Administrative Law Unlawful?) and Adrian Vermeule in some of his work, and a generation of conservative legal scholars trained in the Federalist Society tradition.

The "fourth branch is institutionally necessary" position (steel-manned)

The Constitution was written in 1787, when the federal government had a few hundred employees, ran a postal service and a small army, and did very little else. The economy was agricultural. Public-health crises were addressed at the city level. There were no airplanes, no radio, no nuclear weapons, no securities exchanges, no antibiotics, no electricity grid. To imagine that the institutional design adequate to that world is also adequate to twenty-first-century governance is to misread the Founders, who explicitly built a Constitution flexible enough to accommodate change ("we the People . . . do ordain and establish this Constitution").

Modern governance requires technical expertise that elected officials do not have and could not acquire. A senator running for re-election cannot personally evaluate the safety of a new pharmaceutical, the airworthiness of a redesigned 737, the carcinogenicity of a chemical compound at 0.5 parts per billion, or the systemic-risk implications of a new derivatives instrument. Someone has to make those calls. The choice is not "Congress versus agencies." The choice is "agencies, or worse agencies." Even if Congress wrote every regulation itself, it would still need staff to draft them, and that staff would be the bureaucracy under another name.

The proposed virtues of the system: technical expertise, institutional memory, insulation from the worst lobbying pressures (the FDA can say no to a drug company even when the President needs that company's campaign contributions), and continuity across administrations. Career civil servants outlast any one President — which is how it should be, given that the regulations they administer outlast any one President.

This position, too, has serious adherents — most of the New Deal generation of legal scholars, James Landis (whose 1938 The Administrative Process remains the classic defense), much of contemporary administrative-law scholarship, and a substantial body of empirical political science showing that agency expertise produces better policy outcomes than ad-hoc legislative judgment in many domains.

The honest middle

Most working scholars of the bureaucracy do not adopt either pure position. They observe, instead, that:

• Some agencies have been captured by the industries they regulate (the FAA's relationship with Boeing leading to the 737 MAX certification failures; the SEC's failure to detect Bernie Madoff for years; the EPA's slowness on PFAS; the Mineral Management Service before Deepwater Horizon). The "expertise" defense of agency power has limits.
• Some agencies have demonstrated genuine technical competence that no plausible alternative could match (the NIH's grant-review process, NASA's mission engineering, NOAA's hurricane forecasting, the BLS's economic statistics).
• Some areas of agency rule-making have drifted far from the statutory authority Congress actually granted (a complaint with merit on both sides — the EPA's Clean Power Plan and the CDC's eviction moratorium were both struck down on these grounds, the first under a Democratic and the second under a Republican administration).
• Some areas of agency rule-making have been blocked by hostile political control even when the underlying policy was technically correct (a complaint also with merit on both sides — Trump-era environmental rule rescissions struck down for inadequate analysis; Biden-era student-loan forgiveness struck down for exceeding statutory authority).

The chapter, accordingly, will not tell you whether the administrative state is good or bad. It will tell you what it is, how it works, what the legitimate critiques are from both directions, and what the major doctrinal fights of the 2020s — Schedule F, Chevron's overruling in Loper Bright, the major-questions doctrine, the unitary-executive theory's revival — are actually about.

11.2 Structure and size

The departments

Fifteen Cabinet departments, in rough order of founding:

1. Department of State (1789, originally Department of Foreign Affairs). About 75,000 employees including Foreign Service. Diplomacy, treaty negotiation, passport and visa administration.
2. Department of the Treasury (1789). About 110,000 employees, of which roughly 87,000 are at the IRS. Revenue collection, debt management, currency, financial-system regulation, sanctions enforcement (OFAC).
3. Department of Defense (1947, consolidated from War and Navy). About 770,000 civilian employees plus 1.4 million active-duty military and 800,000 reservists. The largest single employer in the United States.
4. Department of Justice (1870). About 120,000 employees, including the FBI (~38,000), DEA (~10,000), ATF (~5,000), Bureau of Prisons (~35,000), U.S. Marshals (~5,000), and the U.S. Attorneys' offices.
5. Department of the Interior (1849). About 70,000 employees. Manages public lands (Bureau of Land Management, National Park Service), Native American policy (Bureau of Indian Affairs), and natural resources (Geological Survey, Fish and Wildlife Service).
6. Department of Agriculture (1862). About 100,000 employees. Farm subsidies, food-safety inspection (FSIS), nutrition programs (SNAP, WIC), the Forest Service, and rural development.
7. Department of Commerce (1903). About 47,000 employees. Census Bureau, NOAA (weather forecasting and ocean policy), NIST (standards), Patent and Trademark Office.
8. Department of Labor (1913). About 15,000 employees. Bureau of Labor Statistics, OSHA, Wage and Hour Division, and unemployment-insurance oversight.
9. Department of Health and Human Services (1953, originally HEW). About 80,000 employees. Includes the FDA (~18,000), CDC (~13,000), NIH (~20,000), CMS (administering Medicare and Medicaid, about 6,500 staff but managing a trillion dollars), and the IHS.
10. Department of Housing and Urban Development (1965). About 8,000 employees. Public-housing programs, mortgage insurance (FHA), community development.
11. Department of Transportation (1966). About 55,000 employees. FAA, FHWA, FRA, NHTSA, FMCSA.
12. Department of Energy (1977). About 14,000 federal employees plus a much larger contractor workforce running the national laboratories. Nuclear-weapons stockpile, energy research, nuclear-waste management.
13. Department of Education (1980). About 4,000 employees, the smallest Cabinet department. Federal student-aid administration, civil-rights enforcement in education, K-12 grant programs.
14. Department of Veterans Affairs (1989, elevated from the VA). About 440,000 employees — the second-largest department by headcount, almost entirely because the VA runs the nation's largest integrated health-care system and an enormous benefits-administration apparatus.
15. Department of Homeland Security (2002). About 250,000 employees. Customs and Border Protection (~63,000), ICE (~20,000), TSA (~60,000), Coast Guard (~50,000 active including military), FEMA, USCIS, Secret Service.

These fifteen departments together employ the substantial majority of the civilian federal workforce. Notice how far the headcount mix is from intuition. The State Department is small. Education is tiny. The VA is enormous. Defense civilian alone is larger than half of state-government workforces combined.

Independent agencies

Beyond the departments, the executive branch contains roughly sixty independent agencies. The category includes:

• Independent regulatory commissions — the FCC (telecommunications), FTC (consumer protection and antitrust), FEC (campaign finance), NLRB (labor relations), SEC (securities), CPSC (consumer product safety), NRC (nuclear regulation), CFTC (commodity futures), and others. Multimember boards. For-cause removal protections. Substantial independence from the President.
• Single-administrator independent agencies — the EPA (environmental protection), SSA (Social Security Administration, run by a single Commissioner; about 60,000 employees managing $1.5 trillion in benefits), the GSA (federal real estate and procurement), the SBA (small-business loans).
• Other agencies — the CIA, NSA, ODNI (intelligence community); NASA (space); the Federal Reserve System (monetary policy, with a unique structure that makes it "independent within the government"); USAID (foreign assistance, in 2025 partially merged into State); and so on.

Government corporations

About fifty federally chartered government corporations — entities that operate more like businesses than like agencies. The U.S. Postal Service (about 640,000 employees, the largest civilian employer in the federal universe, technically an "independent establishment of the executive branch" with corporation-like operations), Amtrak, the Tennessee Valley Authority, the Federal Deposit Insurance Corporation, the Pension Benefit Guaranty Corporation, Fannie Mae and Freddie Mac (since 2008 in conservatorship), and others.

The shadow workforce

Here is a fact most political science textbooks underplay. The civilian federal workforce has been roughly stable at about 2.1 million for decades — slightly down from its 1960 ratio relative to U.S. population. But federal spending, federal programs, and federal regulatory output have grown dramatically.

How? Because the work is done by contractors and grant recipients. Paul C. Light's research at NYU has tracked this growth for thirty years. By his estimate, the federal government's "true size," counting contractors, grant employees, and consultants, exceeds 9 million people. The visible 2.1-million headcount is roughly a quarter of the actual workforce executing federal policy.

This matters institutionally:

• Congressional oversight is harder when the work is done by employees Congress did not authorize, of companies Congress did not directly fund.
• Personnel rules (civil-service protections, conflict-of-interest rules, FOIA disclosure obligations, pay caps) often do not apply to contractors.
• The career civil servants who do exist often manage contractors rather than do the work themselves — the "hollowing out" critique.
• Cost transparency is poor. Contracting saves money in some cases and wastes it in others, but figuring out which is which requires the kind of agency-by-agency analysis that contractor-driven structures discourage.

Both parties have grown the contractor workforce. Republican administrations because of an ideological preference for private-sector delivery; Democratic administrations because hiring civil servants requires congressional appropriations and political capital that contracting does not.

Your district

Pick the federal employees in your congressional district. The Office of Personnel Management publishes federal-employee counts by metropolitan area. The Postal Service alone has employees in essentially every zip code. The VA has medical centers in most major metros. Defense contractors and military bases are concentrated regionally — visible in places like Hampton Roads, San Diego, and Huntsville; nearly invisible elsewhere. SSA field offices, IRS service centers, federal courthouses, NPS sites — every district has a federal footprint.

The political point: your representative's votes on the federal workforce affect employment in your district. The cuts and reorganizations debated as abstract questions in Washington are job changes in specific places. This is why federal-workforce reductions are politically harder than they appear in budget projections.

11.3 The civil service

From spoils to merit: a 140-year history

For most of the nineteenth century, federal jobs were the property of the winning party. The "spoils system" — most associated with President Andrew Jackson, though the practice predated him — meant that when a new President took office, postmasters, customs collectors, land-office clerks, and most other federal employees were replaced with the new President's supporters. The argument for spoils, in its honest form, was democratic: the people elected a President, the President's program would be carried out by the President's appointees, and rotation in office prevented an entrenched government class.

The argument against spoils, in its honest form, was institutional: the resulting government was incompetent (postmaster appointees couldn't run post offices), corrupt (jobs sold for kickbacks to party machines), and unstable (every four years the entire government changed personnel). The breaking point came in 1881 when President James Garfield was assassinated by Charles Guiteau, a disappointed office-seeker.

The Pendleton Civil Service Reform Act of 1883 created the merit-based civil service. Initially it covered only about 10% of federal jobs. Over the next several decades, the classified service expanded by Presidential order (each outgoing President had an incentive to "blanket in" his appointees by making their positions civil-service positions before leaving). By the mid-twentieth century, the merit-based service covered most federal positions.

The Hatch Act of 1939 restricted the political activities of federal employees — they could vote and hold opinions, but they could not run for partisan office, solicit campaign contributions, or use their official position for political purposes. The Act has been amended several times (1993 to permit some off-duty political activity; subsequent guidance on social media). Its core remains: civil servants are not allowed to use their jobs for partisan campaigning.

The Civil Service Reform Act of 1978 (Carter) restructured the system. It created the Office of Personnel Management (OPM, replacing the Civil Service Commission), the Merit Systems Protection Board (MSPB, an independent agency to hear employee appeals of adverse actions), and the Federal Labor Relations Authority (FLRA, governing public-sector unions). It also created the Senior Executive Service — a corps of about 7,000 senior career executives (and a smaller number of political appointees) intended to provide management continuity across administrations.

How the system actually works

Federal hiring runs through usajobs.gov (the centralized listing) and individual agencies. Most positions are competitive service — open to all qualified applicants, with hiring based on objective criteria. Excepted service positions (foreign-service officers, intelligence analysts, attorneys, and others) follow agency-specific rules outside the standard competitive process. Senior Executive Service positions are filled through a separate qualification process emphasizing executive competencies.

Federal pay is set by the General Schedule (GS) for most white-collar civilian employees: GS-1 through GS-15, with ten "steps" within each grade and locality adjustments. A GS-13 mid-career analyst in Washington earns roughly $115,000 to $150,000 in 2025; the same grade in a low-cost-of-living locality earns less. Senior Executive Service pay starts above the GS-15 maximum and tops out around the Vice President's salary.

Career federal employees have substantial due process protections against adverse personnel actions. To fire a career employee for cause, a manager must document performance problems, provide an opportunity to improve (typically through a Performance Improvement Plan), and follow procedural steps that can be reviewed by the MSPB. Reductions in force (layoffs) follow seniority and veterans-preference rules. These protections exist for the same reason most due-process protections exist: to prevent arbitrary action and to insulate decisions from political retaliation. They have the same downsides most procedural protections have: they make it hard to fire poor performers and slow to reorganize.

Public-sector unions

Federal employees can join unions — the American Federation of Government Employees (AFGE) and the National Treasury Employees Union (NTEU) are the largest — but federal unions are sharply limited compared to private-sector unions. Federal employees cannot strike (the Postal Strike Act and related statutes; air-traffic controllers learned this in 1981 when President Reagan fired 11,000 striking PATCO members). Federal unions cannot bargain over pay or benefits (Congress sets those by statute). They can bargain only over working conditions, grievance procedures, and certain personnel practices — what's called "limited collective bargaining."

State and local public-sector unionization varies dramatically by jurisdiction. Some states (California, New York, Illinois) have robust public-sector collective bargaining; others (Mississippi, North Carolina, Virginia) prohibit it. Public-sector unionization rates (roughly 33% of public-sector workers nationally, compared to 6% in the private sector) drive much of the partisan politics of teachers' unions, police unions, and the federal-workforce reform debate.

Schedule F

The most contested civil-service question of the 2020s concerns Schedule F.

In October 2020, in the closing weeks of his first administration, President Trump issued Executive Order 13957, creating a new excepted-service category called Schedule F. The order authorized agencies to reclassify federal employees in "positions of a confidential, policy-determining, policymaking, or policy-advocating character" out of the competitive service and into Schedule F. Employees in Schedule F could be fired without the procedural protections that apply to competitive-service employees.

President Biden rescinded the order in January 2021 before significant reclassifications occurred. President Trump, returning to office in January 2025, revived a modified version through additional executive action.

The genuine debate:

The case for Schedule F (steel-manned). A democratically elected President is constitutionally responsible for the executive branch. When the President is unable to remove career officials who actively obstruct his policy agenda — leaking to the press, slow-walking implementation, formally complying while substantively undermining — democratic accountability has broken down. The Senate confirms several thousand political appointees per President, but the President-appointed leadership of an agency like EPA or the Department of Education sits atop a workforce of tens of thousands of career employees whose preferences typically lean left of the median voter (federal workforce campaign contributions skew Democratic by a substantial margin, especially in agencies like EPA and the Departments of Education and Justice). Schedule F, in this view, is not about firing people for being Democrats. It is about restoring the President's ability to choose the policy-implementing layer of his own administration, in the same way that any incoming CEO of a private company chooses her own management team. The 50,000-employee estimate often cited as the maximum Schedule F reach is roughly 2% of the federal workforce — leaving the substantial majority of federal employees with the same protections they have always had.

The case against Schedule F (steel-manned). The civil-service system was created precisely because the spoils system produced incompetent and corrupt government. Restoring presidential control over a "policy-determining" tier — where "policy-determining" can be defined broadly — risks reverting to the pre-Pendleton world where federal jobs depend on partisan loyalty and where the technical work of government suffers accordingly. The category "policy-determining or policy-advocating" is highly elastic; in practice, agencies have considered classifying scientists, attorneys, economists, and other technical specialists as policy-related, because their analyses inform policy. An IRS examiner whose audits affect tax-policy outcomes; an EPA scientist whose risk assessments inform regulatory rule-making; a State Department analyst whose intelligence shapes diplomatic posture — under broad Schedule F readings, all could become subject to political dismissal. The result, critics argue, would be politicized science, politicized law enforcement, and a workforce that tells political superiors what they want to hear rather than what the evidence shows.

The debate is ongoing and is one of the most consequential disputes about the structure of American government active in the 2020s.

11.4 Independent agencies and the unitary executive

Multimember commissions and for-cause removal

The independent regulatory commissions — FCC, FTC, FEC, NLRB, SEC, CPSC, NRC, CFTC, and others — share an unusual structure. Each is governed by a multimember board, typically five commissioners, with staggered terms. The President appoints commissioners, with Senate confirmation. Critically, the President cannot fire commissioners at will; he can remove them only "for cause" — typically defined as inefficiency, neglect of duty, or malfeasance.

This for-cause-removal structure was designed to insulate the commissions from short-term political pressure. The reasoning: certain regulatory decisions (the FCC granting a broadcast license, the SEC sanctioning an investment firm, the NLRB resolving a labor dispute) should be made on the merits of the case under the relevant statute, not based on what serves the current President's political coalition. By making commissioners difficult to remove, the law gave them functional independence.

Humphrey's Executor and the constitutional foundation

The constitutional foundation for this design is the 1935 Supreme Court decision in Humphrey's Executor v. United States. President Franklin Roosevelt had tried to fire FTC Commissioner William Humphrey for political reasons (Humphrey was a Hoover appointee unsympathetic to the New Deal). The Court held that Roosevelt could not do so, because the FTC was a "quasi-legislative" and "quasi-judicial" body that Congress had legitimately insulated from at-will removal.

Humphrey's set the framework for a generation of regulatory design. Congress, when it wanted an agency to be insulated, used a multimember commission structure with for-cause removal. Wiener v. United States (1958) reinforced the doctrine in the context of the War Claims Commission. For most of the twentieth century, the constitutionality of for-cause-removal protection for multimember commissions was settled.

Seila Law and the recent erosion

The Supreme Court's 2020 decision in Seila Law LLC v. CFPB signaled a major shift. The Consumer Financial Protection Bureau, created by the 2010 Dodd-Frank Act, was led by a single Director with for-cause removal protection — a structural innovation borrowed from the Federal Housing Finance Agency. Chief Justice Roberts, writing for a 5-4 majority, held that the for-cause removal protection for the CFPB Director was unconstitutional. The "executive Power" must be vested in a President accountable to the people, and a single agency head insulated from the President by for-cause removal threatens the constitutional structure.

Seila did not overrule Humphrey's Executor, but it limited it. The Court suggested that Humphrey's applies to multimember commissions exercising "quasi-legislative" or "quasi-judicial" power (the FTC, FCC, etc.), but does not extend to single-administrator agencies exercising "significant executive power."

In Collins v. Yellen (2021), the Court extended Seila's logic to the FHFA. The boundary of Humphrey's Executor* — and therefore the constitutional viability of the entire independent-agency structure — is now uncertain. Future cases may further erode for-cause-removal protections, potentially even for the multimember commissions.

The unitary-executive theory

Behind these decisions sits the unitary-executive theory, a constitutional doctrine that has gone from academic position in the 1980s to functional majority on the Supreme Court in the 2020s.

The theory's core claim: Article II's Vesting Clause ("The executive Power shall be vested in a President of the United States of America") gives the President plenary control over the executive branch. Officers exercising executive power must be removable by the President. Independent agencies whose officers the President cannot remove violate this constitutional structure, regardless of what statutes Congress may pass.

The originalist case for the unitary executive (steel-manned). The Founders, fresh from a revolution, were keenly aware of the risks of executive power — but also of the risks of plural executives. They considered and rejected proposals for a council of revision and for a multi-headed executive. They settled on a single President precisely because they wanted accountability to be clear: the voters elect one person responsible for executing the laws, and that person controls the apparatus of execution. Independent agencies whose heads cannot be removed sever this accountability link. They place officials with substantial governmental power outside the chain of democratic responsibility. Humphrey's Executor, in this view, was a New Deal-era pragmatic compromise that the modern Court is correctly rebuilding into doctrinally coherent shape.

The institutional case for independence (steel-manned). Some governmental functions should be insulated from electoral cycles. Bank regulation needs to be tough on banks even when the bank lobby is contributing to the President's campaign. Election regulation needs to be neutral between parties even when one party controls the White House. Monetary policy needs to focus on inflation and employment over multi-year horizons rather than on the President's re-election needs. Securities regulation needs to be technically rigorous regardless of who owns the firms being regulated. The Founders did not anticipate any of these institutions because they did not exist in 1787 — but the Founders' broader commitment to constraining the abuse of governmental power supports, rather than undercuts, structures that prevent executive overreach in technical regulatory domains. Humphrey's Executor, in this view, was a sensible application of constitutional principles to a regulatory state the Founders did not foresee, and the modern Court's erosion of it threatens the institutional independence on which much of American regulatory policy depends.

The doctrinal direction of the 2020s has favored the unitary-executive position, but the multimember-commission structure of the FTC, FCC, FEC, SEC, and similar agencies remains, formally, intact. Whether it remains intact in 2030 is one of the more important open questions in American constitutional law.

11.5 Rulemaking and the Administrative Procedure Act

When an agency makes a binding rule — a regulation with the force of law — it follows a process set by the Administrative Procedure Act of 1946 (APA). Understanding this process is the price of admission to understanding how the modern federal government produces policy.

The APA process in brief

1. Statutory authority. Congress, in some prior statute, has delegated rule-making authority to the agency. Sometimes the delegation is narrow ("the Secretary shall set the maximum permissible level of lead in drinking water"); sometimes it is broad ("the Administrator shall promulgate such regulations as are necessary to protect the public health"). The agency cannot make rules where Congress has not delegated authority.
2. Notice of proposed rulemaking (NPRM). The agency publishes a proposed rule in the Federal Register. The notice describes the rule, the statutory basis, the empirical justification, and the agency's analysis. It opens a public comment period — typically 30 to 90 days, sometimes longer for major rules.
3. Comment period. Anyone — corporations, trade associations, advocacy groups, individuals, members of Congress, foreign governments — can submit written comments. Major rules attract tens of thousands or even millions of comments. Comments must be considered, though "considered" does not mean "accommodated."
4. OIRA review. Under Executive Order 12866 (Clinton, 1993), economically significant rules go through the Office of Information and Regulatory Affairs in OMB. OIRA reviews the cost-benefit analysis and may require revisions.
5. Final rule. The agency publishes the final rule in the Federal Register, accompanied by a "preamble" explaining the agency's response to substantive comments. The final rule typically takes effect 30 days after publication, sometimes longer for complex rules.
6. Judicial review. Within a statute-of-limitations window (often 60 days), affected parties can challenge the rule in federal court. The court reviews under the APA standard: "arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law" (5 U.S.C. § 706(2)(A)).

What "arbitrary and capricious" means

The arbitrary-and-capricious standard, despite its forbidding name, is reasonably workable. The Supreme Court's 1983 Motor Vehicle Manufacturers Association v. State Farm decision established the modern formulation, often called the hard-look doctrine. Reviewing courts ask:

• Did the agency rely on factors Congress did not intend it to consider?
• Did the agency fail to consider an important aspect of the problem?
• Did the agency offer an explanation that runs counter to the evidence before it?
• Is the agency's explanation so implausible that it cannot be ascribed to a difference in view or the product of agency expertise?

In practice, hard-look review forces agencies to do their homework. The administrative record — the documents the agency considered when making the rule — must show that the agency examined the relevant data and reached a reasoned conclusion. Agencies often lose challenges because the record is inadequate, not because the rule is substantively wrong.

Cost-benefit analysis

Since Executive Order 12866 (Clinton, 1993, revising Reagan-era EO 12291), agencies have been required to perform cost-benefit analysis for economically significant rules — defined as rules with annual economic impact of $100 million or more. The analysis estimates the rule's costs (compliance costs to industry, administrative costs to government, indirect economic effects) and benefits (lives saved, illnesses prevented, environmental harm avoided, time saved).

The methodology is contested. Monetizing a statistical life — the value of a statistical life (VSL), used to estimate the benefits of safety regulations — has been set by various agencies between roughly $9 million and $13 million. Discount rates for benefits accruing in future years (climate-policy benefits in particular) have been the subject of major political fights: the Trump administration used higher discount rates that produced lower estimated benefits; the Biden administration used lower rates that produced higher estimated benefits.

President Trump's first administration issued EO 13771 (January 2017), the "two-out-one-in" rule: agencies had to identify two existing regulations to repeal for every new significant regulation issued, and the net regulatory cost had to be zero or negative. Critics argued this was an arbitrary across-the-board cap that ignored substantive analysis; defenders argued it was necessary to combat regulatory accumulation. Biden rescinded EO 13771 in January 2021 and revised the executive order on regulatory review (EO 14094, April 2023) to update OIRA procedures. The Trump 2.0 administration has issued new directives on regulatory review structure, the details of which are still being implemented as of 2026.

The volume

Understanding the Federal Register helps to internalize the scale. A typical year produces roughly 80,000 pages. About 3,000 to 4,000 distinct rules are issued annually, of which perhaps 50 to 100 qualify as "economically significant." The Code of Federal Regulations — the consolidated body of all current federal regulations — runs to roughly 200,000 pages across more than 200 volumes. For comparison, the United States Code (federal statutes) runs to about 40,000 pages.

The growth of the Code of Federal Regulations is one of the central facts of modern American government. In 1960, the CFR was about 23,000 pages. By 1980, about 100,000. By 2000, about 140,000. By 2020, around 185,000. Whether this growth represents necessary specification of complex statutes, agency overreach, or a mix depends on which regulations one examines. But the trend is clear.

11.6 Chevron and its overruling

For four decades, the central doctrine of administrative law was Chevron deference.

What Chevron held

In Chevron U.S.A., Inc. v. Natural Resources Defense Council (1984), the Supreme Court reviewed an EPA rule on Clean Air Act permitting. The case turned on whether the EPA's interpretation of an ambiguous statutory term should bind the courts. The Court, in an opinion by Justice John Paul Stevens, established a two-step framework:

• Step One. Has Congress directly addressed the precise question? If yes, the statutory text controls.
• Step Two. If the statute is ambiguous, is the agency's interpretation reasonable? If yes, the court defers to the agency, even if the court would have reached a different reading.

The doctrinal premise: when Congress writes an ambiguous statute and delegates implementation to an agency, the ambiguity should be resolved by the agency (which has technical expertise and political accountability through its presidentially appointed leadership) rather than by judges.

For forty years, this framework structured agency-court relations. Tens of thousands of judicial opinions cited Chevron. Generations of administrative-law students learned the two-step. The doctrine's stability was one of the foundations of agency rule-making — a Trump-era EPA rule, a Biden-era OSHA rule, a Reagan-era SEC rule, all received the same analytical treatment.

Loper Bright

In June 2024, the Supreme Court overruled Chevron. Loper Bright Enterprises v. Raimondo involved a National Marine Fisheries Service rule requiring Atlantic herring fishing vessels to pay for the cost of federal observers stationed onboard. The vessel operators challenged the rule, arguing that the statute did not authorize the cost-shifting. Lower courts had upheld the rule under Chevron deference.

Chief Justice Roberts, writing for a 6-3 majority, overruled Chevron. The opinion's reasoning, in its strongest form:

The Administrative Procedure Act, in 5 U.S.C. § 706, instructs courts to "decide all relevant questions of law" and "interpret constitutional and statutory provisions." That instruction is incompatible with a doctrine that requires courts to defer to an agency's interpretation when the statute is ambiguous. Chevron shifted the interpretive responsibility from courts to agencies — a shift that, in retrospect, distorted the constitutional allocation of power. Courts, applying ordinary tools of statutory interpretation (text, structure, history, purpose, longstanding agency practice as evidence of meaning), can determine the best reading of a statute. Where ambiguity remains after those tools are exhausted, the answer is not automatic deference but careful judicial reasoning.

What changes, what does not

After Loper Bright:

• Questions of statutory interpretation go to courts. When a regulation depends on the agency's reading of an ambiguous statutory term, courts will determine the best reading rather than deferring to the agency.
• Agency expertise still matters on factual findings. Whether a chemical causes cancer at low doses, whether a financial product creates systemic risk, whether a labor practice harms workers — these are factual questions on which courts continue to defer to agency expertise (under different doctrines, including Skidmore deference and the substantial-evidence standard).
• The agency's interpretive role is not eliminated. Agencies still propose interpretations, defend them in court, and prevail when their interpretations are persuasive. They simply don't get the automatic deference Chevron provided.
• Existing rules are not automatically invalid. Loper Bright did not retroactively void all Chevron-era rules. Existing regulations remain in force; they can now be challenged on different grounds going forward.

The strongest case for Loper Bright (steel-manned)

The Constitution vests "the judicial Power" in courts (Article III) and tasks courts with "say[ing] what the law is" (Marbury v. Madison). The APA reinforces this by directing courts to "decide all relevant questions of law." Chevron was an aberration — a doctrine under which judges, the very officials charged by the Constitution with interpreting law, abdicated that responsibility to administrative agencies that are part of the executive branch. Worse, the doctrine effectively made law-interpretation political, because the agency's interpretation could shift each time the White House changed parties. Under Chevron, the same statute meant one thing under President Bush, another under President Obama, another under President Trump, another under President Biden — each interpretation receiving deference, each "reasonable," each immune from meaningful judicial scrutiny. Restoring judicial responsibility for statutory interpretation produces more stable, more predictable, and more constitutionally legitimate law.

The strongest case against Loper Bright (steel-manned)

Federal courts are generalist institutions. A district judge handles drug prosecutions on Monday, immigration appeals on Tuesday, civil-rights cases on Wednesday, and technical environmental rules on Thursday. The expectation that this generalist judge will reach better statutory readings on the technical content of, say, the Clean Water Act's "navigable waters" provision — or the SEC's authority over cryptocurrency tokens, or the NLRB's classification of gig workers — than the agency's career staff with decades of subject-matter expertise is not credible. Chevron recognized this institutional reality: where Congress writes an ambiguous statute about a technical subject and delegates implementation to an expert agency, the agency's reasonable reading is more likely to track Congress's underlying intent than a judge's unaided reading. Overruling Chevron shifts policy authority from politically accountable agency leadership (whose heads voters can affect through presidential elections) to life-tenured judges (whose appointments depend on contingent Senate composition decades earlier). The result, critics argue, is a judiciary now empowered to second-guess every technical regulatory choice, with no comparable expertise to bring to the analysis.

The doctrinal aftermath of Loper Bright is in its early years as of 2026. Lower courts are working out how to apply ordinary tools of interpretation when statutes are genuinely ambiguous. Agencies are being more cautious in pushing aggressive interpretations. Litigation rates against significant rules have risen. Whether the long-run effect is more textually faithful regulation, less effective regulation, or just more litigation will not be clear for years.

11.7 The major-questions doctrine

Parallel to Loper Bright — and arguably more consequential, depending on how it develops — runs the major-questions doctrine.

What the doctrine says

When an agency claims power to decide a question of "vast economic and political significance," the doctrine requires clear congressional authorization for that specific power. Agency action under a vague delegation, or under a delegation that does not specifically address the major question, fails — regardless of whether the agency's reading is "reasonable."

The doctrine gestated for years in scattered cases (FDA v. Brown & Williamson (2000); Utility Air Regulatory Group v. EPA (2014); King v. Burwell (2015) referencing the canon at the margins). It crystallized in two 2022 cases:

• West Virginia v. EPA (June 2022). The Court invalidated the Obama-era Clean Power Plan, the EPA's effort to require existing power plants to shift generation toward lower-carbon sources, on the ground that the EPA lacked clear statutory authority to undertake "generation shifting" of that scope and economic significance.
• National Federation of Independent Business v. OSHA (January 2022). The Court stayed OSHA's emergency vaccinate-or-test mandate for large employers on the ground that OSHA lacked clear authority to issue an economy-wide public-health mandate as opposed to a workplace-specific safety standard.

Subsequent cases extended the doctrine. Biden v. Nebraska (June 2023) struck down the Biden student-loan forgiveness program, holding that the HEROES Act's authority to "waive or modify" student-loan provisions in connection with national emergencies did not authorize a half-trillion-dollar debt cancellation. The CDC's 2020-21 nationwide eviction moratorium was similarly struck down on related grounds in Alabama Association of Realtors v. HHS (2021).

The strongest case for the major-questions doctrine (steel-manned)

Article I gives Congress the legislative power. Major policy choices — should the federal government cancel half a trillion dollars in student debt? Mandate vaccination (or testing) for tens of millions of workers? Restructure the electricity grid? Impose a national eviction moratorium? — should be made by Congress, not by agency officials operating under generic statutory delegations. The major-questions doctrine restores accountability: voters elect a Congress that can legislate the major policy moves, and agencies stay within the boundaries Congress has clearly specified. Without the doctrine, every administration is tempted to read maximalist powers into ambiguous statutes, producing whiplash policy reversals each time the White House changes hands, and depriving citizens of the deliberation and consensus-building that real legislation requires. The doctrine forces hard policy decisions back into the elected branch where they constitutionally belong.

The strongest case against the major-questions doctrine (steel-manned)

The doctrine, critics argue, is itself a judge-made canon — invented by the Roberts Court without textual basis in the Constitution or APA, and applied with uneven rigor that suspiciously tracks the policy outcomes the Court favors. (Pre-2022 EPA rules under the same Clean Air Act provisions did not face major-questions scrutiny; the Clean Power Plan did. Trump-era ICE deportation expansions, which had vast practical significance, did not face major-questions scrutiny; Biden-era student-loan forgiveness did.) The doctrine effectively transfers policy power from an executive branch politically accountable to the President to a judiciary insulated from elections. Worse, the "clear authorization" requirement is impossible to satisfy in advance, because Congress cannot anticipate every consequential application of every statute. The result, critics argue, is the freezing of regulatory authority at whatever level Congress codified at the moment of statutory enactment — an outcome the constitutional structure does not require and the modern administrative state cannot accommodate.

What's at stake

The major-questions doctrine, if applied broadly, will reshape much of the modern regulatory state. Many existing regulations rest on statutes that were never updated to address contemporary applications. Climate regulations under the 1970 Clean Air Act. Cryptocurrency regulation under the 1934 Securities Exchange Act. Internet regulation under the 1934 Communications Act. Health-care regulations under the 1965 Medicare statute as amended. If each of these requires fresh, specific congressional authorization to address new applications, and Congress is too polarized to enact such authorization, the doctrine functionally freezes regulation.

If applied narrowly — to the rare cases of agency action genuinely beyond plausible statutory grounding, like the eviction moratorium based on a public-health-quarantine statute — the doctrine becomes a guardrail against egregious overreach without unsettling the basic structure. Which version of the doctrine prevails will depend on litigation patterns over the next several years.

11.8 The accountability problem: technocracy versus democracy

Step back from the doctrinal weeds. The fundamental question the administrative state poses is older than any of these cases.

Some governmental tasks require expertise: technical, sustained, increasingly specialized. The medical research that approves drugs. The actuarial work that prices insurance for the federal flood-insurance program. The engineering that approves aircraft designs. The toxicology that sets exposure limits for industrial chemicals. The cryptography that secures classified communications. These tasks cannot plausibly be done by elected officials choosing day-to-day decisions on the campaign trail.

Some governmental tasks require democratic accountability: choices about values, priorities, trade-offs. Should we tighten emission standards at a cost in jobs and consumer prices for a benefit in cleaner air? Should we expand or contract immigration? Should the federal minimum wage be higher? How should we balance free speech against harassment online? These are not technical questions. They are political questions, and their answers should track what voters prefer.

The administrative state attempts to do both — and the recurring complaint, from both sides of the political spectrum, is that it gets the boundary wrong.

The progressive critique. Agencies have been captured by the industries they regulate. The Federal Aviation Administration's relationship with Boeing — particularly the delegation of safety certification to Boeing engineers — produced the 737 MAX disaster: two crashes, 346 dead. The Environmental Protection Agency's slow response to PFAS ("forever chemicals") in drinking water, where industry-funded studies dominated the science for two decades. The Securities and Exchange Commission's failure to detect Bernie Madoff despite repeated warnings. The Mineral Management Service's regulatory failures before Deepwater Horizon. The "revolving door" between agencies and the industries they regulate, where regulators move into regulated firms and back again. The progressive case is that more agency capacity, less industry-coziness, and better public-interest funding are needed.

The conservative critique. Agencies have expanded their power beyond statutory grounding and beyond democratic accountability. The Clean Power Plan attempted to restructure the U.S. electricity system based on a Clean Air Act provision that did not contemplate generation-shifting. The CDC eviction moratorium rested on a quarantine statute never used for housing-policy purposes. OSHA's vaccinate-or-test rule applied a workplace-safety statute to a public-health emergency. Student-loan forgiveness rested on a national-emergency waiver provision. In each case, an agency reached for a sweeping policy result that its statute did not clearly authorize. The conservative case is that less agency overreach, more faithful statutory interpretation, and more policy-making by the elected Congress are needed.

Both critiques have merit. They are not contradictory, in fact: the same agency can be too powerful in some domains (when it interprets statutes broadly to reach favored policy results) and too weak in others (when it fails to enforce existing statutes against well-connected industries). Honest evaluation of any specific agency requires examining its specific behavior across both axes.

11.9 Reform proposals

Reform proposals span a wide spectrum. They differ less by partisan affiliation than by which dimension of the problem they treat as primary.

Rebuild civil-service expertise. Increase career staff numbers (especially at agencies hollowed out by hiring freezes). Improve federal compensation to compete with private-sector and state-government alternatives. Reform the security-clearance process. Reduce contracting-out where in-house capacity has been lost. (Generally favored by good-government advocates, public-administration scholars, and federal-employee unions.)

Strengthen presidential control. Schedule F or analogous expansion of at-will dismissal to policy-related positions. Stronger OIRA review. More aggressive use of the President's removal power on agency heads. Reduction of for-cause-removal protections via continued Seila-line litigation. (Generally favored by unitary-executive proponents and some conservative reformers.)

Strengthen congressional control. Require congressional approval for major rules (the REINS Act approach, repeatedly proposed but not enacted). More robust oversight hearings. Strengthen the Government Accountability Office's audit role. Sunset provisions in regulations. (Favored by some conservative and some reformist liberal members of Congress.)

Strengthen judicial review. Loper Bright and the major-questions doctrine, applied vigorously. Restore the non-delegation doctrine to limit how much legislative-style authority Congress can hand to agencies. (Favored by the legal-restoration wing of the conservative legal movement.)

Performance budgeting and management reform. The Government Performance and Results Act (1993) and its 2010 modernization required agencies to set performance targets and report against them. The Program Assessment Rating Tool (PART) under the second Bush administration scored agency programs by effectiveness. Both produced incremental improvements and substantial paperwork; neither transformed agency performance. (Generally favored by good-government think tanks like the Government Accountability Office, the National Academy of Public Administration, and the Partnership for Public Service.)

Government Efficiency proposals (DOGE). The Trump 2.0 administration has launched a "Department of Government Efficiency" effort — technically not a Cabinet department, but a White House initiative — focused on aggressive workforce reduction, contract review, and rule rescission. Some proposals have been blocked by courts (workforce reductions challenged on procedural grounds; rule rescissions challenged on inadequate-analysis grounds). Others have proceeded. The full scope and durability of these reforms will become clearer over the late 2020s.

Open government and FOIA. The Freedom of Information Act (1966, substantially amended 1974, 1986, 1996, 2007, 2016) gives the public a right to request federal records, with statutory exemptions. FOIA productivity is one of the recurring transparency complaints — agencies often take years to respond, and exemptions are sometimes interpreted broadly. Open-government reformers have proposed strengthening FOIA enforcement, expanding proactive disclosure, and improving recordkeeping. The tension is real: full transparency conflicts with deliberation (officials are more candid when their drafts are not subject to FOIA) and with security (some records genuinely require protection). FOIA reform has been a perennial low-stakes political issue, less contentious than the structural debates above.

11.10 Implementation: where policy meets the bureaucracy

A statute is words on paper. A program is what the bureaucracy actually does. The gap between the two is the subject of implementation studies, a field of political science most associated with Jeffrey Pressman and Aaron Wildavsky's 1973 book Implementation: How Great Expectations in Washington Are Dashed in Oakland, with the original subtitle continuing for an additional six lines. (The book is genuinely useful and the subtitle is genuinely funny.)

The Pressman-Wildavsky lesson: even a simple program — they studied an Economic Development Administration grant for a job-training warehouse facility in Oakland, California — requires dozens of consecutive bureaucratic approvals, each with some probability of failure or delay. Multiply enough approvals together, and the probability that the whole chain succeeds approaches zero. Implementation is hard. It is often where well-intentioned policies fail.

The ACA: a bureaucracy stress test

The Affordable Care Act, signed into law in March 2010, provides the canonical modern case study. The statute was 906 pages. It required:

• HHS (specifically the Center for Consumer Information and Insurance Oversight, CCIIO) to issue regulations defining "essential health benefits," "minimum loss ratios," "guaranteed issue" rules, "community rating" requirements, and dozens of other technical concepts.
• CMS (the Centers for Medicare & Medicaid Services) to design and operate Medicaid expansion oversight and the Medicare-payment changes embedded in the law.
• IRS to administer the individual-mandate penalty (later zeroed out by the 2017 tax law), the premium tax credits, and the employer shared-responsibility payment.
• The Department of Labor to issue regulations on group-health-plan changes.
• HHS and CMS together to design Healthcare.gov, the federal exchange website, on a 42-month build schedule with shifting requirements and contracted IT vendors.

The result, as Chapter 8 covered, was substantial regulatory output (thousands of rules, guidance documents, technical bulletins) and a famously botched website launch in October 2013. Healthcare.gov essentially did not function for the first two months. A "tech surge" rescue team — borrowed from Silicon Valley, U.S. Digital Service prototype — rebuilt key systems through late 2013 and into 2014. Enrollment recovered. By 2015, the federal exchange was operational, the state-run exchanges were a mixed bag (some, like California's Covered California, worked well; others, like Oregon's, were so dysfunctional that the state defaulted back to the federal exchange), and the law's substantive coverage expansions were proceeding.

The ACA implementation illustrates several institutional points:

• A statute does not implement itself. Massive regulatory drafting was required to translate broad statutory language ("essential health benefits") into operational rules. That drafting is bureaucratic work.
• Capacity matters. The Healthcare.gov launch failed in part because no government agency had the IT-procurement capacity to manage an integration project of that size. The U.S. Digital Service was created in 2014 partly in response.
• Federalism complicates everything. The ACA gave states the option to expand Medicaid (after NFIB v. Sebelius in 2012 made it optional). About 40 states have expanded as of 2026; about 10 have not. The same federal statute produces dramatically different access to coverage depending on which state you live in.
• Implementation outcomes drive political legitimacy. The ACA's near-death experience in late 2013 nearly broke it politically. Its eventual operational stability — and the rising enrollment numbers as the website worked, premiums stabilized, and the program demonstrated function — moved it from "failed Obama experiment" to "settled element of American health care" over about five years.

Recent implementations: IRA and infrastructure

The Inflation Reduction Act (2022) and the Infrastructure Investment and Jobs Act (2021) — passed under the Biden administration — together appropriated more than $1.5 trillion for clean-energy investment, bridge and road repair, broadband expansion, electric-vehicle infrastructure, and a wide range of related programs. Implementation has been administered across more than a dozen agencies — the Department of Energy issuing tax-credit guidance, Treasury implementing the production and investment tax credits, EPA administering grant programs, DOT distributing infrastructure funds, the Department of Commerce running the broadband program.

Pace of implementation has been a chronic political fight. By mid-2024, several flagship programs (the EV-charger network, the broadband-expansion program, the residential-electrification programs) had distributed only a fraction of the funds Congress appropriated. Defenders argued that doing the work right (real environmental review, real local consultation, real workforce-training requirements) takes years; critics argued that the bureaucracy had overcomplicated rules in ways that delayed delivery beyond political timelines. By 2026, with the Trump 2.0 administration partially rescinding or rerouting some of these programs, the question of how much had reached operational deployment before the political reversal had become its own contested empirical claim.

The lesson, as old as Pressman and Wildavsky: implementation is where policy lives or dies. Pay attention to it.

11.11 What this chapter argued

The federal bureaucracy is not a "fourth branch" in the formal constitutional sense — but it is a structurally distinct institution whose scale, operations, and policy authority no eighteenth-century framework anticipated. Whether to celebrate, fear, or try to reshape this institution is a serious political question on which serious people disagree.

What this chapter has tried to provide is the institutional baseline: how big the bureaucracy is, how it is organized, who works there under what rules, how it makes binding regulations, what doctrines have governed its relationship to the courts and the President, and where those doctrines are heading.

The 2020s are a period of substantial doctrinal change. Chevron is gone, replaced with judicial primacy on statutory interpretation. The major-questions doctrine has carved out a class of agency action that requires clear congressional authorization. Seila Law has begun unwinding the Humphrey's Executor protection of for-cause removal. Schedule F is being implemented in a modified form. Government Efficiency initiatives are challenging the established structure of the workforce.

Whether the resulting institutional landscape is better or worse than the one it replaced depends on values the chapter does not adjudicate. What is undeniable is that the administrative state of 2030 will look meaningfully different from the administrative state of 2020. Citizens who want to evaluate that change for themselves need the institutional baseline this chapter has tried to provide. The next chapter turns to a related set of institutions whose role in this period of transformation has been at least as important: the federal courts.