Specialisation is one of the foundations of institutional capability. Complex organisations depend on people who know different things deeply, from legal interpretation and financial analysis to engineering, public health, data science, procurement and frontline service delivery. As knowledge becomes more specialised, institutional capability can increase substantially because problems that once exceeded generalist competence become tractable. Yet specialisation also changes the architecture through which the institution must think. More expertise usually means more interfaces.

The relationship is structural. A specialist capability becomes institutionally useful only when its knowledge can enter processes involving other specialists, decision-makers and operational actors. Adding a new domain of expertise therefore creates additional points at which knowledge must cross boundaries, be translated, reconciled and integrated. The organisation gains a node of cognitive capability and simultaneously increases the number of relationships required to make that capability usable.

This is not an argument against specialisation. The benefits are often indispensable. It is instead a reminder that local cognitive improvement can create relational demand elsewhere in the system. A legal team may become more sophisticated, but operational staff must know when legal input is required and how to incorporate it. A data-science function may generate more advanced analysis, but policy teams must understand how to use its outputs, challenge assumptions and communicate implementation realities back to technical specialists. LOCAL EXPERTISE ≠ INSTITUTIONAL USABILITY OF EXPERTISE.

The effect becomes more pronounced as expertise differentiates. An organisation with three broadly defined professional domains has relatively few major interfaces to manage. If each domain later divides into several specialised subfields, the potential interaction architecture expands rapidly. Not every specialist needs to communicate with every other specialist, but the institution must still determine where integration is necessary and who owns the connections.

This is why more expertise can paradoxically produce more fragmentation. The institution may contain far greater knowledge than before while finding it more difficult to assemble that knowledge around a common problem. Individual units become smarter in their own domains, yet cross-boundary cognition becomes more demanding.

The distinction from knowledge distribution is important. Distributed expertise means no individual can know the whole, which is often entirely normal in a complex institution. Interface burden concerns the additional architecture required to make that distribution collectively usable. DISTRIBUTED EXPERTISE ≠ EFFECTIVE INTEGRATION OF DISTRIBUTED EXPERTISE. The first describes where knowledge resides; the second describes whether the institution can connect those locations when necessary.

This also differs from ordinary coordination. Coordination can align activities even when deep knowledge does not need to cross boundaries. Expertise interfaces require something richer because specialised reasoning must often be translated without losing the distinctions that make it valuable. A project can be well scheduled and formally coordinated while specialists still fail to understand one another’s constraints.

The problem is especially visible in interdisciplinary policy. A complex infrastructure decision may involve engineers, economists, environmental scientists, lawyers, procurement professionals and local authorities. Each group can perform excellent work inside its own domain, yet the institutional decision depends on the relationships between those domains. One specialist’s assumption may become another specialist’s constraint, and one team’s optimisation may create a problem elsewhere.

The more expertise the institution acquires, the more important boundary-spanning capability becomes.

Some organisations respond by creating integrator roles, joint teams or cross-functional processes. These mechanisms can reduce interface burden because they establish recurring routes through which specialised knowledge is connected. Yet integration itself requires capability. People working at boundaries need enough understanding of multiple domains to recognise when translation is failing, when assumptions conflict and when expertise cannot simply be aggregated into a common answer.

Artificial intelligence can change this architecture in two directions. AI systems may reduce some interface burden by translating technical information, summarising specialist evidence or making expertise more widely accessible. A policy official can query complex data more easily, while specialists can generate explanations adapted to different audiences. These capabilities may genuinely improve interoperability.

At the same time, AI introduces new expertise domains. Institutions now require model developers, data engineers, cybersecurity professionals, domain experts, legal specialists, procurement teams, auditors and operational users to work together around systems that combine technical and public-governance consequences. The addition of computational capability therefore creates new interfaces even as technology helps manage existing ones.

This means the net effect of technology cannot be inferred simply from its ability to automate knowledge work. A new AI capability may reduce the number of human handoffs inside one workflow while creating new governance, verification and accountability interfaces around the workflow as a whole.

The same systems principle applies beyond AI. CAPABILITY ADDITION ≠ INTERFACE REDUCTION. Sometimes the more capable an institution becomes locally, the more sophisticated its relational architecture must become globally.

Interface burden can eventually become a bottleneck. Specialists spend more time translating for one another, meetings multiply, decisions require larger groups and simple questions trigger long routing chains because nobody owns the combination of expertise needed to answer them. The institution may respond by hiring still more specialists, unintentionally increasing the very integration demand that is constraining performance.

The solution is not to reverse specialisation but to treat interface architecture as a first-class design problem. Institutions need to know where specialised knowledge must interact, where common vocabularies are useful, where differences should remain explicit and where integrator roles provide disproportionate value. Not every boundary should be eliminated, but important boundaries should be governable.

A mature institution therefore understands expertise as both a cognitive asset and a relational commitment. Every new specialist capability creates potential value, but realising that value often requires new connections through which the capability becomes usable by the rest of the organisation.

This changes how institutional capacity should be measured. Counting experts tells us something about what knowledge exists. It tells us much less about whether the institution can bring that knowledge together at the moment a decision requires it.

Specialisation makes institutions capable of knowing more than any individual could know, but that advantage survives only if the organisation develops enough interfaces to turn distributed expertise into usable collective cognition.