The dominant narrative about artificial intelligence ( AI ) is move fast or fall behind. Governments are urged to adopt the technology quickly, scale it aggressively and regulate it lightly, as if speed itself were a development strategy.

This assumption is as wrong as it is dangerous. In reality, the main risk for many developing economies is embracing AI too early, before they have the digital infrastructure, institutional capacity, labour market absorption mechanisms, and productive capabilities required to ensure that automation delivers broad-based gains.

I call this risk premature automation, which mirrors a phenomenon that Dani Rodrik described as premature deindustrialization: the erosion of manufacturing employment in developing countries before they could realize its full growth potential. Hasty AI adoption will likely produce a similar outcome: destroying jobs, eroding capabilities and hindering development rather than fostering transformation.

In advanced economies, AI can help supplement an aging labour force and boost white-collar workers’ productivity. By contrast, many emerging economies have much younger populations and limited employment opportunities. Each year, an estimated 12 million young Africans enter the labour market, of which only three million secure formal employment. Against this backdrop, rapid automation in service industries such as customer support, logistics and finance, or even in public administration, could displace workers before alternative pathways emerge.

Latin America and parts of Europe face the same threat: embedding AI into fragmented and outdated administrative systems and sectors with stagnant productivity growth will most likely eliminate middle-skill jobs without creating new engines of growth. If economies with weak fundamentals try to leapfrog into AI-enabled automation, the effort will only amplify their dysfunction.

When countries have paper-based public registries, fragmented payment systems, and weak data governance, AI models are trained on poor inputs and built on brittle infrastructure. Errors scale, and bias hardens, resulting in institutional overload.

Examples abound: Automated eligibility systems for public services have excluded legitimate beneficiaries because the underlying registries were incomplete; algorithmic decision-making tools were introduced before appeal mechanisms existed; and predictive systems were implemented without interoperable data. These are sequencing failures, not technological breakdowns.

In the absence of deliberate sequencing, countries may end up exporting raw data and importing algorithms, platforms and governance systems designed elsewhere, creating new dependencies. Value capture is concentrated upstream, while local firms and workers are relegated to marginal roles or displaced entirely. The technology may be different, but the goal is still extraction.

In this environment, data governance is industrial policy. Countries that fail to devise strategies for interoperability, ownership, and standards will be at the mercy of AI companies. To avoid that outcome, they should focus on sequencing – in other words, embracing frontier technology only after building the necessary foundations.

In practice, sequencing means digitizing records and developing interoperable digital public infrastructure before automating decisions. It also means determining the pace of AI adoption, so that innovation is sustainable. Sandboxes, sector-specific pilots, and labour impact assessments allow governments to learn, adapt, and correct course.

Crucially, such an approach would prevent economic dislocation by ensuring that automation complements, rather than replaces, human labour. When organizations can still learn by doing and accumulate capabilities, AI becomes a development accelerator.

Late-mover status can even be an advantage, because it allows countries to design guardrails before diffusion accelerates, avoiding advanced economies’ mistakes. Pix, Brazil’s government-backed payments system, illustrates how deliberate sequencing and strong public infrastructure can help a latecomer not just catch up, but set global benchmarks.

Current debates about AI governance focus heavily on ethics, safety, and frontier risks. While these issues are important, the more immediate challenge for most countries is the misalignment between AI’s capabilities and institutional readiness. To address this problem, policymakers and stakeholders must consider who captures value; how productive capabilities are built; what happens to workers; and how the technology interacts with existing state capacity.

Premature automation, like premature deindustrialization, will leave economies more fragile and dependent. Africa does not need to win a race it never signed up for. Nor do the many middle-income – and even advanced – economies confronting stagnant productivity and social problems. Instead, they need an AI strategy grounded in logic: digitization precedes automation, capacity precedes scale, and governance precedes diffusion.

In an era increasingly shaped by geopolitical competition over data, semiconductors and cloud infrastructure, sequencing becomes a tool of sovereignty by allowing countries to engage with technology on their own terms. AI need not be a destabilizing force. But it almost certainly will be if policymakers rush to adopt systems they cannot govern.

Mark-Alexandre Doumba is the minister of digital economy and innovation of Gabon.

Copyright: Project Syndicate