About - Rogue Bureaucrat

"Rebellion is born when rulers forget they are meant to serve." — Confucius

"The greatest calling for rebels at work is to help organizations evolve from protectors of accepted orthodoxy to discoverers and promoters of new ideas." — Rebels at Work

"The stories they don't want you to read? Read them twice. The truths they tell you are too complex? Learn them deeply. That's how systems begin to crack." — Lawrence Nault

I'm Marika Olson, and I spent 15 years at USAID managing development portfolios from Indonesia to Libya. Afghanistan, Georgia, Libya, Mali, Indonesia — $70M to $250M at a time, across 20+ countries. I was RIF'd in 2025 when the Administration dismantled the Agency.

I'm a rebel bureaucrat because the bureaucratic craft is worth defending. I was one of the people who built and ran the administrative machinery for foreign assistance — and I'm not ashamed of that word. Bureaucracy, done well, is how a country honors its commitments to the people who depend on them. It is patient, it is specific, it is unsexy, and it is the thing that gets public money to the farmer in Mopti or the hazelnut cooperative in Kakheti. The Administration treated that craft as disposable. I don't.

I write in the tradition of guerrilla government — Pinchot opposing Taft from inside Forestry, Hickel criticizing Vietnam from inside Interior, civil servants who use administrative knowledge as a form of resistance when leadership loses the plot. I'm outside the Agency now, but the craft isn't. It lives in me and in every other RIF'd colleague who still knows how these instruments actually work.

What happened to USAID was a destruction, not a restructuring. DFC survived at reduced scale. State absorbed remnants. The bilateral architecture that took decades to build is largely gone. Private capital, philanthropic mobilization, and locally-led initiatives are picking up pieces but have not reconstituted what was lost.

Rebellions are built on hope. This publication is where I keep the knowledge accessible, usable, and honest while something new gets built — grounded in actual portfolio experience, not think-tank abstraction. What worked, what didn't, what's been lost, what's being built, and who's doing the building.

What I'm Exploring Lately

Developing Thoughts

USAID's legacy: USAID created foundations that enabled what's coming next — but 'what's coming next' is now bifurcated. Some successors (GiveDirectly, fintech, market systems) build on USAID's development logic. The bilateral deal model doesn't build on it; it replaces it with a different logic entirely. The 'evolution not rejection' framing needs a harder edge when the replacement model is extractive rather than constructive.
Private-sector integration: Private-sector integration improves outcomes when structured to create bilateral economic value — both parties benefit commercially. The Georgia hazelnut case proves this isn't theoretical.
Technical solutions vs. systems solutions: Technically elegant solutions fail without system-level understanding. I learned this in grad school with a peanut variety that would never reach the people who needed it. Finance, distribution, politics — these are the real constraints.
The fundamental question for interventions: The diagnostic question should be 'Could this work as a business model, not just a project?' — but the follow-on question is now equally important: 'Is the de-risking instrument designed to change the risk calculus, or just to signal commitment?' Both have to be yes for private capital to actually follow.
Blended finance instrument design: The failure mode in blended finance isn't insufficient subsidy — it's instruments designed to signal good intent rather than structurally alter a lender's risk calculus. Junior tranche positioning changes the math. Larger grants just change the optics. The DFC case crystallizes this: 100% guarantee coverage removes the lender's credit judgment entirely rather than improving it, which is why DCA's partial model outperformed on mobilization ratios. The practitioner diagnostic — 'where do you sit when things go wrong?' — is the clearest single-sentence test for whether an instrument is structural or signaling.
Structure vs. subsidy in development finance: OECD evidence now gives empirical grounding to what practitioners felt in the field: where you sit in the capital stack matters more than how much public money you commit. This reframes the entire debate about ODA efficiency.
DCA→DFC institutional knowledge loss: USAID's Development Credit Authority was arguably the most effective development finance instrument the US government ever ran — 542 partial credit guarantees across 74 countries with a 2.4% default rate and ~29:1 mobilization ratio. DCA's model was development-problem-first: identify a market failure, structure a partial guarantee (typically 50% coverage) through local banks in local currency, and train those banks to lend in unfamiliar sectors. The BUILD Act (2018) merged DCA into DFC alongside OPIC, but the merger functioned as OPIC absorbing DCA — OPIC was 10x larger, and its risk-averse, client-driven, dollar-denominated culture dominated. Brookings documented the result: a comparable $15M DCA guarantee in Colombia cost $250K in subsidy; post-merger, a smaller $8M DFC guarantee in Colombia cost $340K. The 'dispersed institutional knowledge' problem in the blended finance post isn't abstract — it's specifically DCA's 20 years of deal-level data on what guarantee structures work in which markets, now diluted into a larger institution that measures success differently.
The 'USAID gap' narrative: The gap framing was wrong in seven ways: (1) the bilateral replacement model means the US remains present under harder terms — no vacuum, different terms; (2) expanded MDB capacity means the multilateral lane is growing; (3) the $19.2B closeout figure shows the gap is program-type specific — worst for institutionally embedded models with no exit design; (4) Food for Peace reversion adds a category where the instrument survives but the development function disappears inside it; (5) Germany-as-accidental-anchor shows the replacement architecture is circumstantial rather than strategic; (6) the DRC Ebola case adds community trust and field coordination capacity as institutional assets that don't appear in any budget line; (7) the PEPFAR spillover evidence adds a final dimension: the accounting of what's being cut will systematically miss the multi-function health system value that was never in the appropriations calculus — not because it wasn't real, but because it was never designed to be measured.
Transactional bilateralism vs. multilateral soft power: USAID's multilateral model built sovereign capacity and accumulated soft power as a byproduct. The bilateral replacement model extracts sovereign concessions — data, targets, reportedly mining access — as a condition of entry. These aren't equivalent instruments. One compounds; the other depletes.
Aid conditionality evolution: Traditional conditionality was policy-based (structural adjustment, governance benchmarks). The emerging bilateral model appears to be asset-based — tying health programming to resource access. This is a qualitative shift, not a scaling of existing conditionality logic.
Technical vs. political change in development finance: A methodology document from a rating agency can be more consequential than a congressional hearing. The S&P MDB assessment overhaul is a case study in how financial infrastructure shifts while political debates consume all the attention. Development practitioners who only watch political channels will consistently be surprised by the most consequential changes.
MDB expansion as USAID gap-filler: Expanded MDB balance sheet capacity is structurally different from USAID's model in ways that matter for reach. MDBs historically concentrate in middle-income markets with clearer credit profiles. The fragile and frontier markets USAID served most heavily are not where MDB risk appetite naturally lands, even with expanded capacity. More lending power does not automatically mean lending in harder places.
The 'who fills the gap' question: The post-USAID gap question now has three distinct candidate answers that need to be held separately: (1) the bilateral replacement model with harder sovereign terms, (2) private capital via blended finance structures, and (3) expanded MDB lending from rating methodology changes. These are not equivalent instruments. They serve different geographies, risk profiles, and development objectives. The field's habit of treating them as interchangeable is analytically dangerous.
DFC guarantees as signaling vs. catalytic instruments: The OECD evidence now gives a structural answer to the DFC question: DFC's predominant 100% guarantee model sits above rather than below private capital — it removes default exposure entirely rather than sharing it in ways that change senior investor math. Compared to DCA's 50% partial guarantee (which left local banks with skin in the game and required them to make the credit judgment), DFC's fuller coverage functions more as political insurance than structural de-risking. The instrument is not worthless, but it's solving a different problem than mobilizing reluctant private capital into frontier markets.
Water and WASH as blended finance frontier: Water is the next sector where the structure-vs-subsidy question will be tested. The political salience is high (World Bank shift, Matt Damon's investor skepticism) but the commercial logic is genuinely contested — unlike mobile money, where network effects created private incentives, water infrastructure has public-good characteristics that may resist market structuring at the last mile. Whether blended finance instruments can bridge that gap or whether WASH requires a fundamentally different model is an open empirical question.
Price shocks and petro-state capital allocation: Temporary oil price spikes are institutionally dangerous for petro-states with known finite reserves precisely because they strengthen the internal voices arguing for doubling down on extraction infrastructure. The incentive runs exactly counter to what the structural reality requires. The 'trap dressed as opportunity' framing isn't rhetorical — it's the accurate description of how price signals interact with political economy inside NOCs operating under a hard reserve clock.
Energy transition timing in fragile petro-states: The solar pivot is hardest to execute at the moment it is most urgent. Political fragmentation, damaged infrastructure, security constraints in high-radiation southern zones, and DFI exits from fossil lending all converge in the same window as the reserve clock. This is a coordination failure with no obvious institutional owner — not a technical problem awaiting a technical solution.
Institutional brittleness as a program design dimension: Termination cost is an underused diagnostic for structural donor dependency. Programs with high closeout costs — long supply chains, embedded expatriate personnel, procured physical assets — were institutionally brittle by design: they could not persist without continuous donor input. Programs with low closeout costs either dissolved because they had no real infrastructure (a different failure) or because they had embedded capacity in local systems (the rarer, better outcome). The $19.2B figure accidentally quantifies the former category at scale.
The $19.2B closeout figure as forensic evidence: The closeout notification is more analytically useful than USAID's own program evaluations because it was never designed to persuade. It's a legal and financial accounting document that reveals what the institution was actually holding together — not what it claimed to achieve. Extension services, seed systems, and community health cadres probably have structurally different termination cost profiles than infrastructure, procurement-heavy logistics, and supply chain management programs. That difference is the real finding: not all USAID programs were equally brittle, and the cost structure tells you which ones were.
Platform dependency as a development programming failure mode: Development programs that achieved scale by attaching to political vehicles with different primary objectives — commodity clearance, farm subsidies, diplomatic instruments — were always structurally vulnerable to political reversion. The development logic is the layer, not the foundation. When the political conditions that enabled the layering reverse, the layers peel off cleanly. This is distinct from donor dependency (where the program needs ongoing funding) — it's architectural dependency: the program couldn't exist at its current scale without the political vehicle, and the vehicle was never really about development.
The 'reversion not elimination' frame for development instrument changes: Some of the most consequential changes to the development architecture aren't eliminations — they're reversions to an earlier, pre-development-layer design. P.L. 480 returning to USDA with explicit reorientation toward US farm interests isn't a policy change in the normal sense; it's the stripping of decades of added development logic to restore the original 1954 architecture. The instrument survives; the development function disappears inside it. This is harder to document and advocate around than program termination — 'the program survived' is technically true and analytically misleading.
Circumstantial vs. strategic anchors in the aid architecture: A circumstantial anchor — an actor holding up the architecture through budget sequencing and arithmetic rather than mandate — is categorically different from a strategic one. A strategic anchor can be argued with, lobbied, and held accountable to commitments. A circumstantial anchor can shift without a policy decision, without a reversal, without a headline. Germany didn't choose this role; it cannot be assumed to keep it. For program designers, this means the 'donor landscape continuity' assumption has now collapsed at both the US level (policy exit) and the systemic level (no strategic replacement).
US public opinion as a development architecture variable: KFF data showing collapsed US public support for global development isn't a political problem to be managed — it's a structural variable that changes what the development architecture can be built on. Mental exit precedes policy exit and survives policy reversal. Even if US administrations change, the public opinion data suggests the political cost of restoring USAID-scale engagement has risen structurally, not cyclically. Practitioners designing programs with 10-15 year horizons need to treat this as a durable constraint, not a temporary one.
Posture inversion as capacity measurement: When a government shifts from deploying capacity toward an outbreak to restricting movement away from it, that inversion is not a policy preference — it is a measurement. The ban posture is only rational if there is no deployable capacity to send. The entry ban IS the response in the sense that it is the complete set of tools the institution currently has available. This makes it a more honest diagnostic than any budget figure: budget figures measure what was allocated, posture measures what can actually be done.
Community trust as institutional asset in outbreak response: Community trust is an institutional asset with fundamentally different characteristics than funding or personnel: it takes years to build through sustained presence, evaporates rapidly when that presence withdraws, and cannot be reconstituted by emergency deployment. Every prior US Ebola response that worked depended on trust infrastructure built during non-emergency periods. That trust infrastructure is not listed in any budget line and does not appear in closeout cost calculations — which means the $19.2B closeout figure systematically undercounts the actual institutional loss in global health response capacity.
Multilateral mechanisms absorbing slack after US withdrawal: CEPI's three-candidate portfolio approach represents a genuinely different design logic than USAID-era outbreak response: diversified bets across vaccine candidates rather than single-pathway deployment, funded multilaterally rather than bilaterally. Whether it can absorb the coordination and community-trust functions USAID held — not just the R&D functions — is the open question. The portfolio model is good at producing candidates; it was not designed to build the field infrastructure that gets candidates to outbreak zones and into arms.
PEPFAR cuts and the accounting problem: When PEPFAR gets cut, the accounting will measure what PEPFAR was designed to measure: ART coverage, viral suppression, PMTCT outcomes. The spillover evidence — measurable improvements in maternal mortality, child health, economic indicators, educational attainment — was never in the appropriations calculus and won't appear in the loss calculus either. The accounting of the dismantlement will miss most of what's actually being dismantled. This is the same structural problem as the $19.2B closeout figure missing community trust assets: mandatory disclosures are analytically honest about what they can count, and the off-balance-sheet losses may be larger.
Intentional vs. emergent architecture in development programs: PEPFAR's spillover value was emergent, not designed: twenty-three years of sustained presence, trained health workers, supply chains, and community-level infrastructure accumulated effects that the program was never asked to produce and never claimed credit for. The practitioner question — can this be designed deliberately? — is genuinely open and important. The honest answer is probably 'no' for the full spillover effect, but 'yes' for some components: integrated programming with explicit health system strengthening mandates generates measurable spillovers at lower intensity. The architecture wasn't accidental in its components; it was accidental in its cumulative effect.
Vertical program design and health system value: The PEPFAR evidence reframes the vertical-vs-horizontal debate in global health. Vertical programs (single-disease focus, narrow metrics, dedicated supply chains) are politically legible and accountable. What the spillover literature shows is that vertical programs run at sufficient scale, over sufficient time, with sufficient community penetration, generate horizontal effects — not because they were designed to, but because health infrastructure is fungible. The implication is uncomfortable: the political case for vertical programs (clear metrics, demonstrable results) systematically undervalues what they actually build.