Notes — Jennifer Pahlka on Reforming Government

Notes on Jennifer Pahlka in conversation with Tyler Cowen — Conversations with Tyler (https://conversationswithtyler.com/episodes/jennifer-pahlka/), 9 April 2025.

Four questions [Adler frame]

Q1 — What is it about as a whole? The founder of Code for America and former US Deputy CTO argues that American government’s core failure is procedural, not moral or partisan: decades of accreted process have severed the feedback loop between what lawmakers intend and what actually happens when policy is implemented. Pahlka moves across Congress’s passivity, the DOGE-era disruption debate, the imbalance between bureaucratic ‘go energy’ and ‘stop energy’, accountability’s collapse into process, preparing government for AGI, public-sector unions, comparative state capacity (Singapore, Estonia, the UK, US states), and AI-driven procurement and sovereignty questions — returning throughout to one diagnosis: America has never learned to subtract policy once added.

Q2 — How is it argued? Mostly through concrete operational example rather than abstract theory: the IRS Direct File team, USDS’s early institutional weakness, the 212-question California food-stamp form set against Maine’s leaner one, New York City’s Moses-era civil service exams, unemployment insurance dating to the 1935 Social Security Act. Cowen repeatedly pushes provocations (accountability as the problem; DOGE-style chaos as the only workable path; AGI adjudication as fairer than human judgement), and Pahlka’s characteristic move is partial agreement followed by a reframing — she rarely rejects a provocation outright, more often relocating the diagnosis (e.g., ‘it’s not that we have too little state capacity, it’s that we mismeasure what capacity even is’).

Q3 — Is it true, in whole or part? The operational examples (Direct File, the food-stamp comparison, the 212-question form, Moses-era hiring rules) are drawn from her direct professional experience running Code for America and USDS and read as reliable testimony, though none is independently sourced within the conversation. [?] Her broader claims — that government accountability collapses into process more than the private sector does, that Congress’s passivity stems specifically from a broken feedback loop rather than from partisanship or polarisation, and the DOGE insourcing rumour she explicitly flags as unconfirmed — are stated with more confidence than evidence in this transcript. [?source] Her refusal to name a concrete non-defence cut when pressed by Cowen is a notable evasion that Cowen does not let go unremarked.

Q4 — What of it? A first-hand, mechanism-level account of why competent, well-intentioned reform efforts (reinventing government in the 1990s, USDS, DOGE) keep failing to change the underlying system, distinct from both a ‘just cut government’ account and a ‘just fund government better’ account. It gives the wiki a state-capacity vocabulary — go energy/stop energy, the bureaucratic rigidity cycle, procedural bloat versus resource scarcity — that is genuinely new territory, and it extends the existing Sovereign AI concept with a governance-specific boundary case: a nation whose entire public administration runs on a foreign AI stack.

Glossary

Waterfall process — Pahlka’s term for the traditional legislative-to-implementation pipeline: a law is written once, then handed down stepwise through the bureaucracy with no return path to check whether it achieved its intent. Contrasted with an iterative, feedback-driven model. [§ Congress, the courts, and the waterfall problem]

Go energy / stop energy — her operating metaphor for bureaucratic imbalance: ‘go energy’ is the effort of teams trying to build or ship something better (her example: the IRS Direct File team); ‘stop energy’ is the roughly tenfold larger apparatus of rules, infrastructure, and process arrayed to prevent things happening. Reform means identifying and removing stop energy, not simply attacking the people doing go-energy work. [§ Go energy, stop energy, and the bureaucratic rigidity cycle]

Bureaucratic rigidity cycle — the pattern by which reformers create new, unencumbered units (USDS, the Defense Innovation Unit) without ever dismantling the incumbent infrastructure those units were meant to replace, so the new unit ends up merely added on top rather than substituting for anything. [§ Go energy, stop energy, and the bureaucratic rigidity cycle]

Accountability-to-process (vs accountability-to-outcome) — Pahlka’s diagnosis that because government outcomes are heterogeneous and long-term, measurement defaults to fidelity to procedure rather than to actual results — a Goodhart’s Law dynamic she says government suffers more acutely than the private sector, which retains profit as a clarifying outcome signal. [§ Accountability, process, and Goodhart’s Law]

Procedural accretion — her term for the decades-long, rarely-reversed accumulation of regulation on top of a program’s original simple design; her standing example is US unemployment insurance, which traces to the 1935 Social Security Act and is now governed by thousands of pages of active regulation despite a simple underlying purpose. [§ State capacity: Singapore, Estonia, the UK, and America’s 250-year accretion problem]

Key claims by section

Congress, the courts, and the waterfall problem [§ Congress, the courts, and the waterfall problem]

Congress’s passivity stems from members sensing, without wanting to admit it publicly, that the laws they pass do not produce the effects they intend.
The Loper Bright decision and other current disruption may push Congress away from the waterfall model, though Pahlka has ‘no simple answers’ on what replaces it.
Too much residual power has flowed to the courts, which cannot act fast enough to sustain iterative government — a problem she expects to bite especially on the political left, which is relying on litigation to check the Trump administration.

Is DOGE the best available shock? [§ Is DOGE the best available shock?]

Pahlka regrets that past reform efforts (second Bush administration, Al Gore’s reinventing government) were ‘too respectful of old ways of working’ rather than genuinely disruptive, but stops short of endorsing DOGE’s approach as correct.
A senior Air Force official told her defence cuts need to reach roughly 50 percent to genuinely ‘shock the system’; she notes uncomfortably that this rhymes with Vivek Ramaswamy’s proposal to cut bureaucrats by Social Security number.
Her distinguishing claim against pure downsizing: she wants government to work, not just to shrink — though she allows government ‘could be smaller and be better’.

Go energy, stop energy, and the bureaucratic rigidity cycle [§ Go energy, stop energy, and the bureaucratic rigidity cycle]

New reform units (USDS, the Defense Innovation Unit) get created but the old infrastructure they were meant to disrupt — CIOs, procurement frameworks — is never actually dismantled, so gains stay marginal.
Her go energy/stop energy metaphor: any team building something better operates inside roughly ten times as much infrastructure and rule-making designed to stop things; the reform task is to find and remove stop energy deliberately.
Pressed to name a cut, she names the Defense Department as needing a ‘shock to the system’ but declines to specify a non-defence target, preferring more discrimination than DOGE’s current pace. [?]

Accountability, process, and Goodhart’s Law [§ Accountability, process, and Goodhart’s Law]

She partially agrees that accountability itself is part of the problem: public servants are held accountable to process proxies, not to the outcomes people actually care about.
Government suffers this more than the private sector because the private sector retains a clarifying signal (profit) that government lacks.
Invokes Goodhart’s Law to explain why 1990s New Public Management-style target-setting erodes over time; her fix is not abandoning targets but making them revisable as circumstances change.

Preparing government for AGI [§ Preparing government for AGI]

Much current civil-service labour, especially in the social safety net, is low-value paperwork and compliance rather than the human contact the safety net is meant to provide; AGI should displace the paperwork, not the human contact.
She resists Cowen’s claim that AGI adjudication would simply be more fair and therefore acceptable — her concern is less about AGI’s eventual capability than about testing rollout at limited scale before wide deployment.
She concedes AGI’s arrival is ‘disrupting her thinking’ even within the conversation itself, in response to Cowen’s pressure.

Public-sector unions and the case for civil-service independence [§ Public-sector unions and the case for civil-service independence]

Unions often advocate for policies against the broader interests of their own members — especially protecting the small minority of true underperformers, to the frustration of the rest of the civil service.
She partially accepts Cowen’s asymmetric-information account (concentrated losers protest loudly; diffuse costs to everyone else stay quiet) but argues institutions should still be designed against that dynamic rather than treated as fixed.
Citing Philip Howard, she nonetheless departs from him: civil servants need some protection from being fired for inconvenient expertise (‘the science says X’), so knowledge persists across administrations.

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

She resists Cowen’s hypothesis that US ‘military state capacity’ explains American dysfunction, noting recruitment struggles and lost war games against China, but accepts the underlying mechanism that excess resources can themselves cause dysfunction.
Central diagnostic: California’s food-stamp enrolment system, over-resourced into 212 questions, produced the second-lowest US participation rate; Maine, spending ‘orders of magnitude less’, ranked near the top. [?]
The US cannot simply copy Estonia or Singapore because those states digitised comparatively greenfield policy; the US has added policy since 1935 and ‘very rarely subtracted’, and lacks their advantage of no federalism.

AI, procurement, and national sovereignty [§ AI, procurement, and national sovereignty]

Government procurement buys software ‘the way we buy pencils and cars’ — fixed-scope purchase after years of requirements-gathering — when AI requires ongoing internal ownership and iteration.
She reports, with an explicit caveat that it is unconfirmed, that parts of DOGE may actually favour insourcing technical talent rather than further privatisation. [?source]
On whether a nation running its education, military, and social services through American AI remains sovereign (Cowen’s Peru example), she cannot answer, but notes this is the first transformational technology not originating from US government investment (unlike the internet, out of DARPA/DOD) — leaving government with less institutional leverage over what it now depends on.

The best-run American states, and closing with Niskanen [§ The best-run American states, and closing with Niskanen]

She declines to name a single best-run state but points to Pennsylvania (Governor Shapiro, citing a Don Moynihan paper on faster permitting) and Colorado (Governor Polis) as places tackling procedural accretion directly.
Her four-pillar prescription for New York City: fix Moses-era civil-service hiring rules; reduce procedural bloat; invest in digital infrastructure; close the feedback loop between policymakers and implementers.
She now leads a state-capacity team at the Niskanen Center, which she says was the only think tank doing serious state-capacity work when she finished her book.