The Classical Development Toolkit Is Running Out of Mechanisms

●9 ●28
calendar_today ago • schedule5 min read

Based on my KSCCN article The Exhaustion of the Classical Development Toolkit: Aid Transfers, Export-Led Catch-Up and the Labour Ladder under Sovereign AI (DOI 10.5281/zenodo.24125614).

The questions survive; the mechanisms do not

Development economics was born as a theory of transformation. Its founders asked how a poor, agrarian economy could be moved within a generation onto a path of industrial growth, and they answered with a handful of powerful mechanisms: a coordinated big push financed by external savings (Rosenstein-Rodan), the absorption of surplus labour into low-wage manufacturing (Lewis), export-led catch-up under open markets, and aid that closes the gap between what a poor country can save and what it needs to invest (Chenery and Strout).

My argument is deliberately narrower than a proclamation that development economics is dead. Its questions, how capabilities accumulate and how structural change happens, are still the right ones. What is exhausted is a specific set of transmission mechanisms through which the classical answers were expected to work. I trace five of them.

1. Aid as a net resource transfer

According to the OECD's preliminary data, aid from Development Assistance Committee members fell to USD 174.3 billion in 2025, 0.26 per cent of their combined national income, down from USD 215.1 billion in 2024. The 23.1 per cent real-terms fall is the largest annual drop on record, and a further 5.8 per cent decline is projected for 2026.

Shrinking transfers matter more when they meet rising repayments. The World Bank's International Debt Report 2026 records that low- and middle-income countries paid USD 1.4 trillion in principal and interest in 2025, and that between 2022 and 2025 they paid USD 436 billion more in debt service than they received in new financing.

A stylised identity makes the point. Net developmental transfer equals gross disbursement, minus what intermediaries keep, minus the mark-up of tied procurement and foreign technical assistance, minus debt service, plus the returns on earlier investment in sovereign productive capital. The classical model quietly assumed the deductions were small. When they are not, net transfer can turn negative while gross aid keeps flowing. Intermediation compounds this: direct funding from Grand Bargain donors to local and national actors was USD 174 million in 2023, just 0.6 per cent of their first-level funding, against a 25 per cent commitment. Aid should be judged by a test the classical model never posed: its own progressive obsolescence.

2. Export-led openness without connectivity

The second pillar is the belief that openness plus a competitive exchange rate delivers catch-up. Formal openness without physical, productive, financial and relational connectivity systematically fails to deliver stability or structural transformation for economies without technological advantages or dense industrial ecosystems. Smaller exporters end up buying competitiveness through the exchange rate rather than earning it through productivity.

They also face a double bind. Residual trade and investment rules limit the industrial-policy tools that supported late development, while major markets deploy tariffs, regulatory barriers and carbon-border measures that compress the cost advantages smaller exporters still rely on. The result is neither free trade nor coherent strategic protection; it is constrained agency under heightened external pressure. Connectivity, in logistics, payments, standards and regional value chains, has to be built deliberately. Openness alone does not create it.

3. The automation of the labour ladder

The Lewis mechanism needs the modern sector to absorb labour as it grows. My process-level simulation of a representative 1,500-worker ready-made garment factory in a major South Asian exporter tests what AI and automation do to that requirement. These are model outputs, not forecasts:

Scenario Jobs lost Reduction Women among displaced
Partial automation (current trend) 506 33.7% 338
Advanced automation + AI 876 58.4% 581
Near-full automation (theoretical maximum) 1,126 75.1% 739

The partial scenario rests on technology already diffusing in larger factories. Displacement falls hardest on helpers and lower-skilled operators, and women are about two-thirds of the displaced in every scenario.

The arithmetic is simple. Employment equals the labour coefficient times export output, so employment growth is roughly export growth plus the (negative) growth of the labour coefficient. The ladder works while export growth comfortably outruns the fall in workers per dollar of exports. Independent sector data already show the squeeze: exports more than tripled between 2010 and 2024 while employment stayed near four million, and workers needed per USD 1 million of exports fell from about 220 to about 90. The first rung remains, but it no longer carries the weight of surplus labour. This sits alongside Rodrik's evidence on premature deindustrialisation.

4. From capital constraint to compute and sovereignty

The classical toolkit treated capital as the binding constraint. At the frontier, that constraint has shifted to compute, and frontier capability is being rationed rather than diffused: scarce accelerators, high-bandwidth memory and firm power are pre-empted by incumbents, and export controls have shown that capability and permission are different assets. A developing country cannot borrow its way to the frontier the way the financing-gap model assumed it could buy machinery.

The ITU reports 2.2 billion people still offline in 2025, 96 per cent of them in low- and middle-income countries. UNCTAD reports that by the end of 2023 only about 30 per cent of developing countries (excluding LDCs) and 12 per cent of LDCs had a national AI strategy, against about two-thirds of developed countries.

This also changes what foreign direct investment means. An AI zone that imports megawatts and exports residual heat is an enclave, not a development engine. Its productivity depends jointly on energised power, rule certainty, complementary organisational investment, permeable human-capital spillover and domestic offtake. These are complements: omit any term and the jurisdiction is running a warehouse with a national-branding budget. Where the classical model counted capital inflows, this one counts megawatts, trained cohorts and domestic use.

5. Institutional voids as design spaces

Classical policy read weak institutions as deficits to be fixed by transplanting templates through conditionality. A better reading treats voids not as permanent deficits but as design opportunities: adaptive, agile institutions built natively on digital public infrastructure, with AI and distributed ledgers substituting for missing intermediaries in verification, traceability and finance, and tiered, development-linked "fluid rulemaking" in trade governance. This is the most speculative strand, and its evidence is still mostly prototypes and case studies.

What would prove me wrong

A thesis of exhaustion has to say what would disprove it. East Asia's convergence used exactly this toolkit, but under conditions that have changed: growing aid, hard-to-automate assembly and licensable technology. Human capital and macroeconomic stability remain essential foundations. Aid may rebound, and the World Bank reports net transfers on long-term debt turned positive in 2025. Automation may create compensating demand for labour. Open-weight models may diffuse capability faster than compute concentration restricts it.

If, over five to ten years, net aid transfers recover, labour coefficients in export manufacturing stabilise and a broad group of low- and lower-middle-income economies converge without building domestic research and compute capacity, the thesis should be abandoned. The article sets out an indicator set to track exactly that, using existing sources: net aid transfer after debt service; the share of aid through country systems and local actors; debt service as a share of revenue and aid receipts; the labour coefficient of export manufacturing; R&D intensity and researchers per million; sovereign compute capacity; and export complexity and domestic value added.

If the indicators move as predicted, development cooperation will have to measure itself less by what it transfers and more by what it leaves behind: research capacity, fiscal space, sovereign compute and institutions that recipient countries own.

Read the full article, with its references, equations and indicator tables:
Zenodo (DOI): https://doi.org/10.5281/zenodo.24125614
Academia: https://www.academia.edu/180707833

KSCCN

🔥 Join developers growing publicly
Share your knowledge, build in public, and grow your developer presence with a global community.

More Posts

How AI Is Changing Web Development, and What Developers Still Need to Understand

Daniel Decker - Nov 10

Dozens of email accounts: learning AI coding tools as a student

rahuldk1105 - Oct 27

Reimagining Official Development Assistance: Sovereign AI and Structural Cooperation

Coder2100 - Nov 6

Rails, Compute and Patient Capital: The New Engines of Structural Transformation

Coder2100 - Nov 14

Technology as Power: Acceleration, Divergence, and the Reconfiguration of Agency

Coder2100 - Nov 4
chevron_left
1.2k Points • 37 Badges
Dhaka, Bangladesh • drmuntasir.github.io
18Posts
4Comments
3Connections
Dr. Syed Muntasir Mamun is Chief Innovation Officer, and Director General of both the International ... Show more

Related Jobs

View all jobs →

Commenters (This Week)

2 comments
1 comment
1 comment

Contribute meaningful comments to climb the leaderboard and earn badges!