Governing Advice... from AI?

Is this advice any good?

August 29, 2026

Why Not?

I asked several frontier AI systems how the US governing system could be redesigned for long-term human flourishing. This isn't a rigorous experiment. It's a musing.

If you find the long-term vision of the current US president or Congress, or Supreme Court, this won't interest you.

If you are against AI output on principle, also no need to read further.

Would you take advice from someone who has never been alive?

But if you have no problem accepting ranked links or summaries from your browser search, then why not, out of curiosity, ask an AI how best to govern a country?

Maybe it will simply steal the best human thoughts. If so, why not take a look?

You can ask your favorite AI chat yourself, in your own words. It may complain about vague formulation and ask you to clarify, but it will give detailed advice anyway.

I asked three.

Suggestions, generated by AI

These are points that appeared repeatedly across outputs from Claude (Opus 5), ChatGPT (GPT-5.6 Luna) and Mistral Vibe, summarized with help from Claude:

Electoral system

1.
Replace single-member winner-take-all House districts with multi-member proportional districts (STV or MMP), with a 3–5% threshold to prevent fragmentation.
2.
Abolish the Electoral College. Equal weight per vote.
3.
Remove districting from politicians via independent or algorithmic commissions.
4.
Independent election administration — no elected official controls the machinery determining whether they stay in power.
5.
Easy voting: automatic registration, early and mail voting, auditable paper ballots.

Executive

1.
Emergency powers expire automatically absent legislative renewal (30–60 days).
2.
Reclaim delegated powers — tariffs, war powers, spending, agency restructuring back under legislative control.
3.
Protect enforcement and administration from political removal: civil service, inspectors general, prosecutors, election officials, statistical agencies. All three treat this as central rather than incidental.
4.
No presidential self-pardon.

Courts

1.
Fixed staggered terms for the highest court, 12–18 years, one appointment on a predictable schedule, plus an enforceable ethics regime.

Legislature

1.
Strip the malapportioned upper chamber of its veto over ordinary legislation, restricting it to genuinely federal questions.
2.
Independent analytic capacity protected from political interference — evaluation, statistics, forecasting.

Error correction and time horizon

1.
Sunset clauses on programs and powers, with renewal conditional on independent evaluation.
2.
Prefer reversible policy designs: pilot, measure, scale, reassess.
3.
Independent fiscal council publishing long-horizon debt and demographic projections, plus formal intergenerational accounting.
4.
Long-term planning body producing 20–50 year projections that governments must respond to publicly.
5.
Sortition-based citizens' assemblies for questions where partisan politics performs badly.

Anti-corruption

1.
Radical lobbying and campaign-finance transparency: real-time disclosure, meeting logs, beneficial ownership, cooling-off periods.

The presidency

1.
Do not retain a strong single-person presidency alongside a proportional legislature. Replace by parliamentary or collegial executive.

What's wrong?

Here we can peek at the reasoning for the suggestions. Claude Opus 5 lists the failure modes that actually generate the symptoms:

1.
Winner-take-all districts mechanically produce two parties, and two parties in a diverse country produce coalitions that agree on nothing internally and treat every election as existential.
2.
Primaries select candidates. Roughly 10–20% of voters, skewed toward the ideologically committed, choose who appears on general-election ballots in the ~85% of districts that aren't competitive. The general election is often ceremonial. This is the single largest polarizing engine in the system.
3.
Congress has abdicated. War powers, tariffs, emergency powers, spending discipline — all delegated or surrendered. Power flows to the presidency regardless of occupant, so each election gets more dangerous.
4.
Vetocracy plus unchecked unilateralism at once. The US simultaneously cannot pass ordinary legislation and cannot restrain executive action. This is the worst combination: hard to build a rail line, easy to start a war.
5.
Two-year House terms make the time horizon of the median legislator shorter than the payback period of nearly every worthwhile investment.
6.
The Constitution is effectively unamendable. Article V has the highest bar of any major democracy. Since the system can't self-correct through amendment, it adapts through judicial reinterpretation and executive expansion instead, which is less legitimate and less reversible.

You can compare the diagnosis and proposals above with the alternative, the human-generated one: government documents, the president's vision as voiced by himself, and the system of government we have now.

The comparison is oddly asymmetric. The AI proposals come from systems that have never lived in a society. The alternative comes from a system produced entirely by humans: elections, institutions, advisors, think tanks, courts, legislatures, and a few centuries of trial and error.

Which approach you'd trust with the next fifty years is the actual question.

That's all. No rigor intended, just enough for musing. A couple of appendices are there, only if you're curious.

Appendices

What to ask the AI?

Ask in your own words, the main points will be about the same. For example:

Make a list of suggestions on how US governing system could be improved, if the goal is to ensure florishing of US population for the next 50–100 years.
Consider this as an optimization problem. Assume that any changes you think of can be implemented.

A few versions suggested by AI:

Diagnose the structural failure modes of the US constitutional system, then rank reforms by expected leverage. Assume any change is implementable.
Optimize for robustness against bad occupants, not peak performance.

A longer version:

The US constitutional system is 240 years old and shows signs of strain: deep partisan polarization, a Congress that has ceded power to the presidency, low confidence in government, and near-zero capacity for formal amendment. Treat this as an institutional design problem.
Diagnose the structural failure modes first, then give a list of reforms ranked by expected leverage. Assume any change is implementable, including constitutional amendment. Optimize for robustness under a bad-faith or incompetent officeholder rather than peak performance under a good one, since the design has to survive the full distribution of who ends up in power.
Include: (1) reforms that sound appealing but are net-negative, with reasons; (2) the transition problem, meaning how a reform gets adopted when its partisan valence is visible in advance; (3) an honest account of what institutional design cannot fix.
Be partisan-neutral. Where a reform advantages one side, say so.

Another angle:

Treat the US government as an institutional design problem. Design the system maximizing the probability of long-term, broadly shared flourishing over the next 50–100 years.
Design for the actual United States: 340 million people, continental scale, deep geographic and partisan sorting, low and declining institutional trust, high internal diversity. Assume it remains a single polity; everything else is open. Do not assume the presidential system, two-party system, electoral arrangements, Congress, Supreme Court, or federal structure should be preserved.
Optimize for competent evidence-based government, prosperity and innovation, political equality, individual liberty, social stability, error correction, resistance to corruption and authoritarianism, fiscal and intergenerational sustainability, and institutional legitimacy. These conflict. Where they do, name the conflict, state which you are trading off, and justify it.
Assume every rule will be probed for exploits by intelligent, well-resourced actors over decades. Design for the 10th-percentile officeholder.
Deliver: (1) the current system's structural failure modes; (2) which current features are load-bearing and should survive, with reasons; (3) the highest-impact reforms, ranked, going deep on the top five to seven rather than listing thirty; (4) for each, its adoption path, reversibility, and whether it is self-entrenching; (5) reforms that appear attractive but are net-negative; (6) how your proposed system itself fails, and what its collapse looks like in year 80.
Use international experience, but flag confidence levels and note where comparison cases differ from the US in ways limiting transfer. Optimize for the system's long-run performance, not for any party, ideology, or group. Where a reform has partisan valence, say so.

A long version of suggestions, by Claude Opus 5

Tier 1: Change how representatives are selected

This is where the returns are highest, because everything downstream is a consequence of selection.

1.
Multi-member House districts with proportional or single-transferable-vote allocation. Three to five seats per district. This is the highest-value single change available. It structurally eliminates gerrymandering (you cannot pack and crack a five-seat district), makes 85% of voters relevant again, gives urban Republicans and rural Democrats actual representation, and permits four to six viable parties. Coalition governance forces bargaining as a routine practice rather than a betrayal. A version of this already exists as legislation — the Fair Representation Act would establish ranked-choice voting for House elections, require multi-member districts in states with more than one representative, and mandate independent redistricting commissions. Wikipedia
2.
Ranked-choice or approval voting for all single-winner offices. President, Senate, governors. Removes the spoiler dynamic, so voting your preference is never self-defeating, and rewards candidates with broad acceptability over candidates with intense factional support. Alaska and Maine already use it statewide, though five states have banned it. Substack
3.
Replace partisan primaries. Either nonpartisan top-four primaries feeding a ranked general election, or restore genuine party gatekeeping through conventions and endorsements. The current system stripped party elites of vetting power in the 1970s without replacing it with anything. Parties that cannot deny their own nomination to someone cannot discipline anyone.
4.
Eliminate the Electoral College, or achieve the same result through the National Popular Vote Interstate Compact. It currently gives a systematic 3–5 point advantage to one party's coalition, distorts campaigning toward six states, and makes national elections hinge on tens of thousands of votes in a handful of counties. Note the honest partisan valence: this currently disadvantages Republicans, which is why it will not pass on its own merits absent the delayed-implementation trick discussed below.

Tier 2: Restore legislative capacity

1.
Abolish the 60-vote filibuster, replacing it with a mandatory deliberation period. A body that cannot legislate by majority cedes its function to executive orders and judicial rulings.
2.
Rebuild congressional expertise. Congress has fewer committee and personal staff than in 1979 while the complexity of what it regulates has exploded. The expertise gap is currently filled by lobbyists, for free, in exchange for influence. Restore the Office of Technology Assessment, triple committee staff, raise legislator pay substantially to reduce revolving-door capture. This is cheap and unglamorous and probably has one of the best cost-benefit ratios on the list.
3.
Four-year House terms, staggered or synced with the presidency. Cost: reduced responsiveness. Benefit: a legislator whose next election is four years away can vote on something that pays off in three.
4.
Automatic continuing resolutions. If appropriations lapse, spending continues at prior levels. This removes shutdowns as a hostage-taking device permanently.
5.
Reclaim delegated powers with sunsets. Tariffs above a threshold require congressional approval. Emergency declarations expire in 30 days absent affirmative vote. Military deployments face automatic funding cutoff at 90 days without authorization. The point isn't that presidents have abused these; it's that a power delegated indefinitely will eventually be used by someone you didn't anticipate.

Tier 3: Constrain the executive

1.
Statutory and constitutional limits on the pardon power: no self-pardon, no pardons for offenses connected to the president's own conduct or those of family and appointees. This is the clearest unpatched hole in the original design.
2.
Insulate law enforcement. Fixed terms for the FBI Director and Inspectors General, removal only for cause with congressional notification, and a prosecutorial mechanism for senior officials that doesn't run through an appointee of the person being investigated.
3.
Replace impeachment. A two-thirds Senate threshold is unreachable in a polarized party system, which means the removal mechanism effectively doesn't exist. Options: a constructive vote of no confidence, a supermajority of both chambers combined, a mixed tribunal including state officials, or a national recall referendum with a high signature bar. Any of these is better than a rule that has never once been used successfully.
4.
Binding conflict-of-interest and financial disclosure law covering the presidency, enforced by a body the president cannot fire.

Tier 4: Fix selection quality and the courts

1.
Eighteen-year staggered Supreme Court terms, one appointment every two years. This ends the actuarial lottery, ends strategic retirement, ends the incentive to appoint the youngest plausible nominee, and lowers the stakes of any single confirmation. Add a binding, externally enforced ethics code.
2.
On age: I'd steer away from hard age caps. They're blunt, they exclude capable people, and the correlation between age and capacity is loose. The better mechanism is mandatory independent cognitive and medical certification, publicly disclosed, for all federal officeholders and candidates above 70. Let voters decide with information rather than having the rule decide for them. The real reason gerontocracy persists isn't the absence of an age limit — it's that safe seats and incumbency advantage make retirement optional. Fix the districts and much of this fixes itself.

Tier 5: Build long time horizons into the structure

1.
An independent fiscal council with a statutory debt anchor, modeled on the UK's Office for Budget Responsibility or Chile's structural balance rule. Not a balanced-budget amendment, which is procyclical and dangerous in recessions.
2.
BRAC-style commissions for decisions with diffuse benefits and concentrated costs: infrastructure siting, tariff schedules, entitlement parameter adjustments. An expert body produces a package; Congress votes up or down without amendment. This is a commitment device against the politics of individually rational, collectively disastrous carve-outs.
3.
Expand automatic stabilizers so recession response triggers on unemployment data instead of on a discretionary fight that takes nine months.

Tier 6: Repair the epistemic layer

1.
Citizens' assemblies by sortition for issues where elected politicians are trapped by primary incentives. Ireland used randomly selected citizens deliberating with expert input to break decades-long deadlocks on abortion and same-sex marriage, with referendum ratification. It works because randomly selected people have no reelection to worry about.
2.
Election administration floors: automatic registration, multi-day voting or a national holiday, uniform paper-trail requirements for federal races, and a firm separation between election administration and partisan officials.
3.
Small-donor matching funds on the New York City model, plus full disclosure of political spending. Requires overturning parts of Buckley and Citizens United, so it's a constitutional-amendment item.
4.
Local news. The collapse of local journalism correlates with rising polarization, lower split-ticket voting, and higher municipal borrowing costs from reduced oversight. Fund it through tax credits or public-media endowments with structural independence, not content regulation.

Tier 7: The meta-reform, which may matter more than everything above

1.
Make the Constitution amendable again. Lower the Article V bar to something like three-fifths of Congress plus three-fifths of states, or add a national-referendum path, or mandate an automatic convention-call referendum every 20 years (several state constitutions do this). This is the master variable. Every item on this list might turn out to be wrong. A system that can be amended can correct its errors; a system that cannot must route adaptation through judicial reinterpretation and executive expansion, which is exactly the pathology you're complaining about. Converting one-shot design into iterative design with feedback is the highest-value change in expectation, even though it's the least exciting.

Reforms that sound good and probably backfire

Worth listing, because an optimization that ignores negative-value moves is incomplete.

Legislative term limits. State-level evidence is fairly consistent: they shift power to lobbyists and unelected staff, who become the only people who understand the process. You get amateurs governed by professionals nobody elected.
Balanced budget amendments. Force austerity during recessions, when deficits are the correct policy.
Direct national referenda on ordinary policy. Vulnerable to manipulation, poor at handling package tradeoffs, and prone to majoritarian overreach. Sortition-based deliberation is the better version of the same impulse.
Simply removing veto points. Careful here. Hungary and Poland demonstrate that parliamentary systems with proportional representation and amendable constitutions can be captured faster than the US system, precisely because there's less friction. The correct move is asymmetric: reduce friction on ordinary policy, increase it sharply on changes to the rules themselves. Constitutional amendments, court composition, election law, and emergency powers should be harder to change than tax rates. The current US system has this backwards in places.

The transition problem

This is the part most reform lists skip, and it may be the binding constraint.

Any reform adopted in a moment when one party sees it as advantageous will be regarded as illegitimate by the other and reversed at the next opportunity. Rule changes need buy-in from people who don't yet know whether they'll benefit.

The mechanism for this is delayed implementation behind a veil of ignorance. Pass reforms that take effect in eight to twelve years, past the point where anyone can confidently predict the partisan landscape. The 27th Amendment's congressional-pay rule works partly because of this logic. It's the only realistic route to adopting something like the National Popular Vote or House proportional representation, both of which currently have obvious partisan valence and are therefore dead on arrival as immediate changes.