The Skills Gap That AGI Creates Inside Your Organisation Is Not the One You Think

Most workforce conversations in enterprise AI circles orbit the same question: which roles disappear first? It is the wrong question, or at least an incomplete one. The more consequential challenge for technical leaders right now is identifying which human capabilities increase in value precisely because AI handles execution, and then building L&D investment strategies that do not depreciate faster than the models they are meant to complement. Getting this wrong does not just waste training budget. It leaves your organisation structurally exposed at the exact moment when competitive differentiation depends on what your people can do that AI cannot replicate cheaply.

Why Execution Skills Are the Wrong Investment Target

AI systems are improving fastest at the layer of work that is most legible: writing code to a specification, summarising documents, generating test cases, translating requirements into structured outputs. These are execution skills, and they have historically consumed a large share of engineering L&D budgets precisely because they are teachable, measurable, and immediately visible in output quality.

The problem is that commoditisation of execution creates a ceiling, not a floor. When any competent engineer with a capable model can produce a working implementation in half the time it previously took, the differentiating value shifts upstream, toward the person who can define what should be built and why, and downstream, toward the person who can assess whether the output actually solves the right problem.

Organisations that continue investing primarily in execution-layer upskilling are, in effect, training their teams to compete with AI on AI's own terms. That is a diminishing-returns strategy, and the rate of diminishing returns is accelerating.

The Lindy Effect Applied to Capability Investment

The Lindy effect is a useful heuristic here. Capabilities that have been valuable for decades are more likely to remain valuable for decades more. Judgment, institutional context, the ability to decompose ambiguous problems, and the skill of recognising when a technically correct answer is commercially wrong: none of these are new, and none of them are becoming easier to automate.

This matters for L&D roadmapping because it suggests a durability test. Before committing budget to any capability development programme, the question worth asking is whether the skill being developed would still be valuable if the underlying tooling changed entirely in eighteen months. For most prompt engineering curricula, the answer is no. For structured thinking about system trade-offs, the answer is yes.

The practical implication is that your L&D roadmap should be anchored to capabilities with long half-lives, and treat tooling fluency as a hygiene factor rather than a strategic investment. Teams need to know how to use current tools effectively. That is table stakes. The compounding returns come from investing in what sits above the tool layer.

What Compounds: Judgment, Context, and Institutional Knowledge

Judgment Under Uncertainty

Judgment is the capacity to make sound decisions when information is incomplete, constraints are in tension, and the cost of getting it wrong is asymmetric. AI systems can surface options and model scenarios, but they cannot carry accountability for the decision or hold the organisational history that makes one option obviously wrong in a specific context.

Engineering leaders who invest in developing judgment in their teams are building a capability that becomes more valuable as AI handles more of the analytical groundwork. The human role shifts from analysis to evaluation, and evaluation requires judgment that cannot be transferred from a model.

Context-Setting and Problem Definition

The ability to frame a problem correctly is upstream of everything else. A well-framed problem with a mediocre solution frequently outperforms a poorly framed problem with a technically excellent one. As AI accelerates execution, the quality of the initial framing has a larger multiplier effect on outcomes.

Developing this capability inside engineering teams means creating structured practice in requirements articulation, stakeholder translation, and constraint identification. These are not soft skills in the pejorative sense. They are precision instruments that determine whether AI-assisted work produces value or produces polished noise.

Institutional Knowledge Preservation

Institutional knowledge is the accumulated understanding of why systems are built the way they are, which decisions were reversed and for what reasons, and where the bodies are buried in the architecture. This knowledge does not live in documentation. It lives in people, and it is extraordinarily difficult to reconstruct once it walks out.

As AI tools accelerate onboarding and reduce the time junior engineers spend on execution tasks, there is a real risk that the natural process by which institutional knowledge transfers, through working alongside experienced engineers on hard problems, gets short-circuited. Preserving this transfer mechanism requires deliberate design, not just good intentions.

How to Structure a Capability Audit

A useful starting point is a two-axis assessment of your current L&D investments. One axis measures the durability of the capability being developed, using the Lindy test described above. The other measures how much of that capability is already being absorbed by AI tooling in your stack.

Investments that score low on durability and high on AI absorption are candidates for deprioritisation or significant reduction. Investments that score high on durability and low on AI absorption are where you should be concentrating budget and organisational attention.

The third category, capabilities that score high on durability but are partially supported by AI, is the most interesting. These are areas where AI handles the mechanical layer but the human skill required to direct and evaluate that mechanical layer is increasing in importance. Structured code review, architectural decision-making, and cross-functional technical communication all sit in this category.

Building Development Programmes That Do Not Depreciate

The structural principle here is to separate tooling fluency from capability development in your programme design. Tooling fluency should be continuous, lightweight, and largely self-directed. Capability development should be deliberate, cohort-based where possible, and measured against outcomes rather than completion rates.

Mentorship and apprenticeship structures are underinvested in most engineering organisations and have strong durability characteristics. They transfer judgment and institutional knowledge in ways that formal training cannot replicate. The cost is time from senior engineers, which is a real constraint, but the return is compounding capability rather than depreciating certification.

The organisations that will be best positioned as AI capabilities continue to advance are not those that upskilled fastest on the tools. They are those that preserved and developed the human judgment layer that gives those tools meaningful direction. That is the capability gap worth closing.

Companion piece to our broader work on organisational readiness for AI systems. See What Long-Running Agents Expose About Engineering Team Readiness for a practical analysis of the workflow and trust-calibration gaps that surface when teams move from AI assistance to autonomous agents.