If your strategic framework for national AI capability assumes that AI engineering skills (building and deploying AI systems) and AI literacy skills (using AI-enabled tools) grow in roughly parallel proportions, the 2025 LinkedIn AI Skills Diffusion Index shows the framework is wrong almost everywhere. In most countries, AI literacy is growing meaningfully faster than AI engineering. India and the US both show this pattern. The exceptions (UAE, Chile, and South Africa) show the opposite, with engineering skills climbing faster than literacy skills. The divergence is now substantial enough that "AI talent supply" needs disaggregation into literacy and engineering layers, and the strategic implications cut against intuition.

AI Skills Diffusion Index by country, literacy vs engineering, 2016–2025:

The LinkedIn AI Skills Diffusion Index is a new measure introduced this year. It tracks AI skills growth within a country relative to that country's own baseline, distinguishing between:

  • AI engineering skills: technical capabilities for building and deploying AI systems (e.g., machine learning frameworks, neural network architecture, deployment infrastructure)
  • AI literacy skills: familiarity with AI-enabled tools and applications (e.g., AI prompting, Microsoft Copilot Studio, AI agents from a user perspective)

The index measures growth, not level. A country that started with a small AI workforce and scaled rapidly will show high diffusion scores. A country that started large and scaled at the same pace will show lower diffusion scores. The metric is therefore best read as a trajectory indicator.

The pattern in the data: in most of the 47 countries in the LinkedIn sample, the AI literacy line is climbing steeper than the AI engineering line. The interpretation: the workforce is acquiring AI tool fluency faster than it is acquiring AI system-building skill. The technology is diffusing as user capability faster than it is diffusing as builder capability.

The three exceptions (UAE, Chile, and South Africa) show the inverted pattern. AI engineering skills are growing faster than AI literacy skills. The interpretation: the workforce in these countries is acquiring builder capability faster than user fluency. This is structurally unusual and worth examining.

[CHART fig_744_2026 — Fastest growing AI literacy and engineering skills in the US, 2025]

The US-specific data adds detail. The fastest-growing AI literacy skills in the US in 2025 are AI prompting and Microsoft Copilot Studio. The fastest-growing AI engineering skills are AI agents, AI productivity, and AI strategy. The literacy skills are user-facing tools. The engineering skills are higher-order capability terms: "agents" and "strategy" are themselves overlapping with literacy-level usage in many contexts. The boundary between literacy and engineering is fuzzy in the US data.

Three contested interpretations of the literacy-outpacing-engineering pattern are visible, and the choice between them matters for talent strategy.

Interpretation 1: the pattern reflects rational labour-market adaptation. The argument: AI tools are now available to non-technical workers, and adopting these tools produces immediate productivity benefits. AI engineering capabilities take longer to develop, have higher educational prerequisites, and offer benefits over longer horizons. The workforce is rationally optimising for the higher-immediate-return capability acquisition. Literacy will continue to grow faster than engineering because the marginal return on literacy is higher in the short term. Engineering growth will remain modest because the supply of new engineering-trained workers is constrained by educational pipeline limits.

Interpretation 2: the pattern reflects a structural under-investment in engineering capability that will produce capability gaps in 2027–2030. The argument: AI engineering capability is the prerequisite for building competitive AI systems. A workforce heavy on AI literacy and light on AI engineering can use AI but cannot build it. National AI competitiveness rests on engineering capability; literacy alone doesn't produce competitive position. The countries that have prioritised engineering (the three exceptions) are building structural advantage that will appear over a 5–10 year window.

Interpretation 3: the literacy-engineering distinction itself is more porous than the LinkedIn metric allows. The argument: AI agents, AI strategy, and AI productivity (the US "engineering" growth categories) are not pure engineering skills; they include strategic and product capabilities that overlap with literacy. The metric may be measuring labelling conventions in LinkedIn profiles more than underlying capability composition. Under this view, the literacy-engineering divergence is partly an artefact of how workers describe their skills rather than what they actually do.

The data does not adjudicate between these interpretations. The strategic planning implication, however, is similar across interpretations 1 and 2: organisations and countries should treat AI literacy supply and AI engineering supply as separate variables rather than as a single "AI talent" supply. The interpretation 3 view, that the distinction is fuzzy, is a methodological caveat but does not eliminate the strategic distinction.

Three structural implications for talent strategy follow.

The first implication: AI engineering capability is now a scarce variable in most countries. The literacy supply is growing fast enough that "find someone who can use AI tools" is no longer a binding constraint in most labour markets. The engineering supply is growing more slowly, which makes "find someone who can build AI systems" the binding constraint for organisations whose AI strategy depends on internal building rather than external tool adoption. Organisations need to know which constraint applies to their AI strategy and source accordingly.

The second implication: the country variation is now strategic. The UAE, Chile, and South Africa pattern, engineering outpacing literacy, represents national workforces with structurally different AI capability composition. Organisations with operations in those countries access an engineering-heavy supply that is different from the literacy-heavy supply in most of the rest of the world. The strategic question for global organisations is whether this distinction is consequential for their AI strategy; if it is, geographic talent sourcing decisions follow.

The third implication: the literacy-engineering ratio is moving fast enough that strategic plans need to monitor the ratio rather than fix it. The current ratio (literacy growth much greater than engineering growth in most countries) may shift as engineering education catches up, as the literacy supply approaches saturation, or as the boundary between the two categories evolves. Strategic plans built on the 2025 ratio without monitoring will be operating from a fixed-point reading of a moving variable.

The contested question for strategy planning: which of the two interpretations (rational adaptation vs structural under-investment) is correct? The answer affects whether the 2026–2030 trajectory continues the current pattern (literacy continues to outpace engineering) or reverses (engineering investment catches up). The data does not yet show enough years for high-confidence inference. The recommendation: plan for both, with explicit signal monitoring. The cost of mis-calibration is asymmetric. Interpretation 2 (structural under-investment) carries the larger cost if correct, because the resulting capability gap is harder to close on short notice. Strategic plans that hedge toward interpretation 2's caution are more robust to which interpretation prevails.

For workforce strategy teams setting global AI talent direction, the planning anchor needs to incorporate the literacy-engineering distinction explicitly. The "AI talent" variable that most plans currently use is too coarse to capture the relevant supply dynamics. Disaggregating into literacy and engineering, and tracking each as a separate variable with separate growth rates and separate country distributions, produces a planning view that matches the structural picture the data is now showing.

Sources