How AI Can Help Fix the Balance of Power in Washington — and Make Government Work Better for Everyone. A Policy Note.
Over the last several decades our government has grown a powerful “Fourth Branch” –an “administrative state” comprised of federal agencies. To an increasing extent, these agencies have been operating independently of the Executive and Legislative Branches that our Constitution did create and provide for. This expanding and increasing agency role has been encouraged, until recently, by a doctrine (called the Chevron doctrine) by which courts have deferred, in interpreting the laws empowering agencies, to the interpretations put forward by the agencies themselves.’
As a result, the branch of Government that most significantly affects the lives of citizens sets its own pace (slow) and controls its direction (largely unfettered by oversight from elected officials). Trust in government is at historic lows —as government keeps growing and reaching further into our lives.
Fortunately, the Supreme Court has recently decided that the courts, rather than the agencies, should interpret the laws. Now, there are also practical tools to put citizens and those they elect back in control. Information Technology and Artificial intelligence —with their ability to keep track of what the agencies are doing, assess consistency with law and executive branch policy and empower the legislative branch through quick and thorough determination of the intent and meaning of the laws Congress enacted — offer a path back to the constitutional structure the Founders actually intended: a government run by elected Executive Branch officials, kept in check by the Congress and the courts, and faithfully carrying out what the electors decided.
The way forward runs through the past. To restore balance, the Congress must play a more active role in interpreting and overseeing conformity with the laws it enacts and the Executive must be better able to implement those laws in accordance with Administration policies and priorities as well as the applicable laws. IT/AI makes that possible with scale, speed and precision that, until now, has been unthinkable.
I. What the Constitution Set Up — and How We Drifted Away
The Constitution’s plan for government is simple. Congress makes the laws. The President carries them out and oversees the agencies doing the work. Courts decide disputes about what the laws mean and whether agencies are staying within their bounds.
Agencies are tools — not independent power centers. Congress creates them, defines their jobs, funds them and oversees their use of appropriated funds. The President and his team of appointees direct and manage them. Courts keep them within the scope of the authorizing legislation. The agencies themselves have no independent constitutional status, but they are the machinery that turns laws and policy decisions into real-world action.
For roughly the first 150 years of our history the federal government was small enough that the President could manage it and Congress could watch it.
The New Deal changed this. Faced with a complex modern economy, Congress started handing broad, vague authority to agencies — deliberately vague, to give agencies room to adapt. The Supreme Court initially pushed back but backed down after confrontation with President Roosevelt. Broad delegation and agency autonomy became standard practice.
Agencies filled statutory open space with their own policy judgments. Government theory began to treat agency expertise as a special source of agency authority that didn’t need to be earned through elections. Courts, challenged by lack of clear language in the laws, developed a doctrine called Chevron deference — basically allowing agencies to decide what ambiguous or unclear statutes
mean, and defining the limits of their own power.
By the end of the twentieth century, the administrative state had become a fourth branch of government — accountable on paper to elected officials but in practice running largely on its own.
II. What Presidents Have Tried — and Why It Hasn’t Been Enough
Every President for the past fifty years has tried to assert some level of control over the administrative agencies and make them more efficient and responsive. Their approaches have been very different — and those differences point to a real disagreement about what the solution should be.
Republican administrations, starting with President Reagan, have generally tried to assert more presidential control over the regulatory machine. President Reagan’s executive order created the Office of Information and Regulatory Affairs (OIRA) inside the White House, and required agencies to submit major regulations for presidential review. The idea: that the President, as head of the
executive branch, should be calling the shots.
Though that logic is compelling, in practice, presidential control through OIRA has never been enough. Agencies can delay, paper over, and litigate around directives. The sheer volume of federal rulemaking has been too much for any White House to really oversee.
Democratic administrations, particularly President Clinton’s, had a different focus — less about control and more about making the agencies run better. Clinton’s National Performance Review deliberately empowered agencies rather than reining them in, focusing on efficiency and customer service. His OIRA order weakened presidential oversight compared to Reagan’s, affirming what it
called “the primacy of Federal agencies in the regulatory decision-making process.”
Both approaches missed the same deeper problem: neither asked whether agencies are faithfully doing what Congress authorized. A President can grab the wheel of an agency that has been running off-road for thirty years, but that doesn’t put it back on the right road. And deliberately loosening oversight just makes the accountability problem worse.
III. What the Courts Have Done — and What’s Still Missing
The Supreme Court has recently started correcting some of the legal doctrine that gave agencies so much room to roam.
Most significantly, in 2024’s Loper Bright Enterprises v. Raimondo, the Court overruled Chevron deference. Courts — not agencies — are now responsible for saying what statutes mean. This is a real restoration of constitutional structure: the judiciary interprets the law, the executive carries it out, and neither branch should give agencies the power to define their own authority.
Decisions like this matter. But they also expose a gap. Even without Chevron, when statutory text is genuinely unclear, courts still must decide what Congress meant — and without better tools for doing that, the answer can easily become whatever the judge’s policy instincts suggest rather than a genuine recovery of what Congress actually decided.
The fix to that problem is a robust, reliable method for determining what Congress authorized. That is exactly what AI-assisted legislative history analysis can provide.
IV. The Legislative History Problem — and What AI Can Do About It
Legislative history is the full record of everything that happened while a bill was being considered, written and passed — hearings, floor debates, committee reports, amendments that were proposed and rejected, conference committee proceedings, and statements by the lawmakers involved. That is the most complete and direct evidence of what Congress was actually trying to do.
Courts have long used legislative history, but in a messy and controversial way. The record is huge, full of contradictions, and easy to manipulate. Lawyers learned to plant favorable statements in the record to win future cases. Judges tended to pick out the parts that supported where they already wanted to go. And doing a truly thorough review of legislative history has always been so expensive
that it only happened in the most heavily litigated cases — and even then, incompletely.
These are real problems. But they are criticisms of how legislative history has been used, not of the idea itself. The issue was never that what Congress intended is irrelevant — in a democracy, it’s the most relevant thing there is. The problem was that we lacked the tools to recover that intent reliably. AI changes that. For the first time, it’s possible to analyze the complete legislative record of a statute — not cherry-picked excerpts, but the whole thing — with consistency and a resistance to manipulation that human researchers simply can’t match.
What comprehensive AI-assisted analysis delivers:
It shows the full pattern of congressional understanding, not isolated statements. A floor statement planted to influence future litigation looks very different when the surrounding record shows no other member mentioned it, it may contradict dozens of other statements, and the complete history points the opposite way.
It gives proper weight to failed amendments — arguably the most compelling form of legislative history. When Congress explicitly considers and rejects a specific provision, that’s not ambiguous. That’s Congress saying directly what it chose not to authorize. When an agency later claims exactly the authority Congress voted down, AI analysis of the amendment record makes that history unmistakable.
It brings meaningful legislative history to the thousands of regulatory disputes that never get it today, because thorough research is too expensive and time consuming. The result is more consistent, more predictable, and more democratically honest interpretation across the full range of federal law.
Most importantly, it shifts the question from “what does this text mean?” to “what did Congress
actually decide?” — which is the constitutionally right question. The text is the expression of that decision. When it’s genuinely unclear, the legislative record is the best evidence of what the decision
was.
V. A Framework That Actually Works
This points toward is a three-part framework that is both constitutionally sound and practically viable.
Presidential direction, clearly grounded in law. The President heads the executive branch and is responsible for making sure the laws are faithfully carried out. Within the authority Congress has granted, the President should direct agency policy. Agencies do what elected officials authorized —
and the President is responsible for making sure that happens.
Congress as the real source of agency authority. Agencies do what Congress authorized them to do — no more, no less. The end of Chevron deference points this direction. AI-assisted legislative history analysis makes it enforceable in practice, by giving courts and the executive branch reliable evidence of what Congress actually authorized in any specific case.
Courts that interpret, not defer. Courts interpret the law — not deferring to agencies and not substituting their own policy preferences for Congress’s. AI gives courts a tool for genuine, comprehensive, manipulation-resistant access to what Congress actually said and decided.
This framework has a direct payoff for government efficiency. Much of the waste and confusion in federal regulation doesn’t come from too much or too little presidential control. It comes from agencies operating under unclear or contested understandings of their own authority — forcing them to fight major rules in court, reverse course when judges disagree and burn enormous resources
defending positions that a clearer reading of congressional intent would have settled years earlier.
Better legislative history analysis cuts through that. When the record clearly shows what Congress authorized, agencies can act with confidence. When it clearly shows what Congress didn’t authorize, agencies are saved from years of litigation over claims they were never entitled to make. Both outcomes serve efficiency and the public interest at the same time.
VI. What to Do Next
Putting this into practice doesn’t require new legislation, though congressional buy-in would help. Several steps are achievable right now.
The executive branch should direct agencies to conduct and document comprehensive AI-assisted legislative history analysis for significant rulemakings. “What did Congress actually authorize here?” should be answered before “what policy should we adopt within that authority?” OIRA’s review function could include checking whether the statutory record actually supports the agency’s claimed authority.
Courts should embrace comprehensive legislative history as a tool for resolving genuine statutory ambiguity. Now that Chevron is gone, courts are responsible for their own statutory interpretation. Better tools for doing that job honestly should be welcomed.
Congress should continue to invest in building and maintaining comprehensive, AI-accessible legislative record databases covering significant statutes. It’s a relatively modest investment that would pay off across the entire federal regulatory system.
Academic and legal communities should develop standards for AI-assisted legislative history analysis — methodologies that ensure thoroughness, guard against new forms of manipulation, and produce results that courts and agencies can actually rely on.
VII. What’s Really at Stake
This note focused on constitutional structure and how we interpret laws. But the deeper stakes are about democratic governance–whether Americans can trust their government and whether our government can be sufficiently efficient to meet global competition.
Americans have watched government efficiency and resulting trust in government fall steadily for decades, even as government has grown. A big reason for the distrust is the perception — largely accurate — that government runs according to its own internal logic rather than in response to democratic direction or the demands of global competition. Agencies have pursued agendas that elected officials never clearly authorized. Courts have deferred to those agencies’ views of their own authority. Presidents of both parties have pursued reforms that produce modest improvements but leave the fundamental accountability problem unresolved.
An AI assisted framework can address these issues directly. When agencies operate within clearly established statutory authority and in conformity with the policies of the elected Administration, what they do carries legitimacy that autonomous agency policymaking never can. When courts recover what Congress actually meant rather than deferring to agencies or substituting judicial preferences, they respect rather than displace the democratic process. When the President directs an executive branch that is faithfully implementing what Congress authorized, we get accountable governance rather than drift.
AI doesn’t resolve the substantive disagreements that divide Americans about what government should do. Those disagreements are healthy and properly settled through political dialog. What AI can do is help make sure that when political decisions get made — when Congress passes a law and the President signs it — the law actually means what its authors decided it should mean, and the
agencies charged with carrying it out do so faithfully.
That would be a modest but profound contribution to democratic governance. It is, in a deep sense, back to the future — using the most advanced technology available to recover what the Constitution’s designers intended: a government that does what the people’s elected representatives authorized, no more and no less, checked by courts that interpret rather than make the law, and directed by a President accountable to the voters who chose him and in compliance with applicable
laws.
The tools are there to restore the Founders’ vision in dealing with the complexities of the 21st century. We must be smart enough, brave enough and motivated enough to use them properly in order to continue the American story.
— End —
Recent Comments