Jennifer Pahlka on Reforming Government

Jennifer Pahlka, founder of Code for America and former US Deputy Chief Technology Officer, talks with Tyler Cowen about why Congress has gone passive, whether DOGE-style shock is the only route to reform, the imbalance between ‘go energy’ and ‘stop energy’ in bureaucracy, accountability’s collapse into process, preparing the state for AGI, public-sector unions, and what fifty different state civil services reveal about the myth that government can only be one way.

Key ideas

1. Congress has lost the feedback loop, not just the will to act. Pahlka reads congressional passivity as members sensing that the laws they pass do not produce the effects they intend — ‘turning the steering wheel, but it’s not connected to the wheels of the car’. The fix is not more oversight but a shift from a one-way ‘waterfall’ legislative process to an iterative one that checks back against intent as implementation proceeds.
2. ‘Go energy’ and ‘stop energy’ are badly out of balance. Any team trying to build something better — her example is the IRS Direct File team — operates inside ten times as much infrastructure, rules, and process designed to stop things. Reform requires deliberately finding and removing stop energy rather than simply attacking the people doing go energy work.
3. Accountability has collapsed into process, not outcomes. Because government outcomes are heterogeneous and long-term, measurement defaults to fidelity to procedure rather than to results — a version of Goodhart’s Law that Pahlka says afflicts government more than the private sector, which still has profit as a clarifying signal.
4. Resource abundance can produce worse service than resource scarcity. California, with vastly more money and tech talent than Maine, built a food-stamp enrolment form with 212 questions and had one of the lowest participation rates in the country; Maine, spending far less, ranked near the top. Over-specification and risk-aversion, not underfunding, was the barrier.
5. America’s deepest capacity problem is 250 years of unremoved policy. Unlike Estonia or Singapore, which built digital government on comparatively greenfield policy, the US has ‘added and very rarely subtracted’ since 1935 — leaving even simple programmes like unemployment insurance buried under decades of accreted regulation that digitisation alone cannot fix.

Content

Congress, the courts, and the waterfall problem

Cowen opens on why Congress has become passive. Pahlka, who came to Congress-watching late after years inside executive-branch agencies, argues that members increasingly sense the disconnect between the laws they pass and what actually happens on the ground — ‘turning the steering wheel, but it’s not connected to the wheels of the car’. She traces this to a ‘waterfall’ process inherited from industrial-era government: legislation is written once, handed down through the bureaucracy, and rarely revisited against its original intent. The alternative is an iterative model with a tight feedback loop between policymakers and implementers — closer to how anything that works well in ordinary life actually operates. She is sharply critical of the amount of residual power that has flowed to the judiciary as Congress and the executive branch have failed to close that loop: courts cannot act quickly enough to sustain iterative government, a problem she says will bite hard on the political left as it looks to litigation to blunt the Trump administration.

Is DOGE the best available shock?

Cowen presses Pahlka on whether a chaotic, break-things approach — DOGE, or something like it — is simply what large-country reform requires, given that past efforts (the second Bush administration, Al Gore’s ‘reinventing government’) tried and largely failed. Pahlka’s answer is qualified regret rather than agreement: ‘I just feel sad that we didn’t try’ a more orderly disruption. She distinguishes the 1990s reinventing-government push, which was a 20th-century, industrial-era effort, from the 21st-century digital transformation she believes is actually needed. She recounts a senior Air Force official telling her a defence cut needs to be at least 50 percent to ‘shock the system’ enough to change behaviour — and admits her own instincts, formed by four years on the Defense Innovation Board, ‘rhyme’ uncomfortably with Vivek Ramaswamy’s proposal to cut bureaucrats by Social Security number. Her distinguishing claim is that she wants government to work, not merely to shrink, though she concedes ‘it could be smaller and be better’.

Go energy, stop energy, and the bureaucratic rigidity cycle

Asked why the US doesn’t simply build new, unencumbered agencies and let old ones wither — as with Palantir and Anduril substituting for parts of the defence bureaucracy, or Christopher Kirchhoff’s Defense Innovation Unit — Pahlka names the failure mode the ‘bureaucratic rigidity cycle’: new units get created but the old infrastructure (CIOs, procurement frameworks) is never actually dismantled, so the new unit ends up added on top rather than replacing anything. She experienced this herself building the United States Digital Service: even years in, USDS had far less institutional power than the incumbent procurement and IT structures it was meant to disrupt. Her operating metaphor is an imbalance between ‘go energy’ (the Direct File team building something cheaper, faster, better) and ‘stop energy’ (the roughly tenfold apparatus of rules and infrastructure around them). The reform task is to identify and remove stop energy deliberately — not to cut go energy first, which is what she fears is happening under DOGE’s less discriminate approach. Pressed for a concrete target, she names the Defense Department as needing a shock to the system, though she resists naming a single non-defence cut, preferring ‘far more discriminate[ion]’ than the current pace.

Accountability, process, and Goodhart’s Law

Cowen puts a provocation to Pahlka: that accountability is not the solution to bureaucratic dysfunction but the problem, because outcome measurement inevitably morphs into process measurement. Pahlka agrees in a qualified sense: public servants are held accountable to metrics that are proxies for procedure, not for the real outcomes people care about. She accepts Cowen’s observation that this also afflicts large private firms (he cites declining US construction productivity) but argues government suffers it more acutely, because the private sector still has a clarifying signal — profit — that government lacks. She invokes Goodhart’s Law (‘once a measure becomes a target, it ceases to be useful’) to explain why 1990s New Public Management-style goal-setting erodes over time, and argues the fix is not abandoning targets but making them revisable — a digital-transformation habit of iterating on the target itself once it stops tracking the intended outcome.

Preparing government for AGI

On how government should prepare for AGI, Pahlka resists the premise that it simply means firing workers. Her example is the social safety net: enormous amounts of current civil-service labour are low-value paperwork and compliance work rather than the human contact — actually helping someone in a tough time — that the safety net is meant to provide. She wants AGI to displace the paperwork and free workers (an eligibility worker for SNAP, for instance) to do the higher-value human work, while conceding displacement will happen regardless. Cowen pushes harder, arguing AGI could eventually adjudicate cases more fairly than humans and that people will simply have to accept that; Pahlka’s resistance is less about capability than about what people currently want from these interactions — empathy, referral to other services — and about the discipline of testing AI on real government needs at limited scale before wide rollout, rather than ‘rolling stuff out willy-nilly’.

Public-sector unions and the case for civil-service independence

Cowen argues public-sector unions will resist both AGI-driven change and accountability generally, and asks whether they should simply be abolished. Pahlka agrees unions often advocate for policies against their own members’ broader interests — particularly their protection of the small minority of true underperformers, which she says drives the rest of the civil service crazy without unions campaigning to fix it. She resists Cowen’s asymmetric-information explanation (concentrated losers scream, diffuse costs to everyone else stay quiet) only partially, agreeing the dynamic exists but insisting institutions should still be designed against it. Citing Philip Howard, whose critique of public-sector unions she has ‘warmed to’ over the years, she nonetheless departs from him on one point: civil servants need some protection from being fired for telling inconvenient truths — ‘the science says X’ — so that expertise persists across administrations rather than the entire civil service being wiped out with each new one.

State capacity: Singapore, Estonia, the UK, and America’s 250-year accretion problem

Cowen probes comparative state capacity — Singapore’s technocratic competence, the UK’s stagnation since the financial crisis, and a provocative hypothesis that America’s high ‘military state capacity’ (willingness to fight and die) is intrinsically bound up with the dysfunctions Pahlka criticises. Pahlka resists the hypothesis (recruitment is in fact struggling, and the US loses war games against China) but concedes the underlying point that excess of a resource can itself cause dysfunction. Her central diagnostic example is food stamps in California versus Maine: California, awash in money and Silicon Valley talent, over-specified its enrolment system into 212 questions and landed second-lowest national participation; Maine, spending ‘orders of magnitude less’, ranked near the top. It was too many resources, not too few, that produced worse service. On why the US cannot simply copy Estonia or Singapore, she argues the real asymmetry is legacy: those states had comparatively greenfield policy environments to digitise, while the US has never meaningfully subtracted policy since the 1935 Social Security Act — unemployment insurance, a program that should be simple, is now buried under ‘thousands of pages of active regulations’. Estonia and Singapore also lack federalism, which multiplies the US problem across fifty separate policy stacks.

AI, procurement, and national sovereignty

Cowen raises the possibility that AI capability forces a de facto privatisation of government through procurement, since the US government has no AI of its own. Pahlka diagnoses the deeper procurement failure: government buys software ‘the way we buy pencils and cars’ — years of requirements-gathering followed by a fixed-scope purchase — when software, and especially AI, requires ongoing internal ownership and iteration. She reports (with a caveat that it is unconfirmed) that parts of the DOGE effort are actually pro-insourcing technical talent, on the reasoning that operating government with so little in-house technical competence ‘is crazy’. Extending the question to other nations — Cowen’s example is Peru running its education, military, cybersecurity, and social services through American AI — Pahlka concedes she cannot answer whether such a country remains meaningfully sovereign, and notes a structural novelty: unlike prior transformational technologies, this one did not originate from US government investment (DARPA, DOD), so the government itself has less institutional understanding of, or leverage over, what it now depends on.

The best-run American states, and closing with Niskanen

Asked to name the best-run US state, Pahlka declines a single answer but points to Pennsylvania under Governor Shapiro (citing a Don Moynihan paper on faster permitting) and Colorado under Governor Polis as places where interesting change is happening — cases of states either tackling procedural accretion directly or never having let as much accrete in the first place. Asked what she would tell New York City’s government, she gives four concrete pillars: fix a hiring-and-firing system still shaped by Robert Moses-era civil service exams; reduce procedural bloat, which she calls her current favourite target; invest in digital infrastructure; and close the feedback loop between policymakers and implementers so that policy is treated as iterative rather than ‘fire and forget’. She closes on her current role leading a state-capacity team at the Niskanen Center — a think tank she says was, notably, the only one working seriously on state capacity when she finished her book — aiming to extend that four-pillar framework to Congress, states, and cities alike.

See also

• Jennifer Pahlka — speaker
• Tyler Cowen — host
• State Capacity — the concept the episode anatomises
• Recoding America — Pahlka’s book on the delivery deficit
• Sovereign AI — a related dependence-and-capacity argument at the level of AI infrastructure