Every year brings a wave of technology predictions. Most restate the same themes with different adjectives. This is a shorter list — five things that practitioners in regulated industries should actually be paying attention to in 2023, and a note on how to separate signal from noise in each area.
Large language models moving from experiment to production
2022 was the year that demonstrated LLM capability at scale. ChatGPT's release in November compressed three years of gradual awareness into six weeks. For most enterprise technology leaders, the question shifted from "is this real?" to "what do we do about it?" in a single quarter.
2023 is the year the governance questions get real. For regulated industries — healthcare, manufacturing, financial services — the question is not whether LLMs are useful. It is how you evaluate them, audit their outputs, and design the human-in-the-loop workflows that meet regulatory expectations.
The organisations that will benefit in 2023 are not necessarily the ones that move fastest. They are the ones that move with appropriate rigour — piloting use cases where the failure mode is containable, building evaluation frameworks before scaling, and treating the governance questions as design inputs rather than compliance afterthoughts.
Specific use cases worth the attention in regulated industries: document summarisation with human review, classification of incoming communications, first-draft generation for routine structured content, and knowledge base search with citation. These are high-volume, well-bounded, and have clear evaluation criteria. That makes them the right entry points.
Data mesh gaining traction
Data mesh is not new — Zhamak Dehghani's original paper is from 2019 — but it is transitioning from architectural theory to enterprise implementation. The organisations that succeed with it are treating it as an organisational change programme that requires technology support, not a technology project that affects organisation.
The practical question for 2023 is not whether data mesh is the right model. It is whether your organisation has the domain team maturity, the platform engineering capacity, and the governance discipline to implement it. For most large enterprises, the answer is "partially, and transitionally." The most realistic implementations in 2023 will be hybrid models — central data engineering providing platform infrastructure while domain teams gradually take on data product ownership.
The failure mode to watch: organisations that adopt the data mesh vocabulary (domain ownership, data as a product) without the organisational investment needed to make it real. Domain teams that are named as owners but not resourced, supported, or held accountable for data quality end up with the same data governance problems as before, plus the added complexity of a distributed operating model.
Platform engineering becoming mainstream
The internal developer platform concept — building a paved road to production that product engineering teams can use without deep infrastructure expertise — is gaining traction as a response to the cognitive load that cloud-native development has imposed on application teams.
The value proposition is real. Product teams that spend 30 percent of their time on infrastructure toil — configuring pipelines, managing environments, navigating cloud permissions, debugging deployment failures — are not building product. Platform engineering teams that absorb that complexity and surface a well-designed developer experience measurably improve product team throughput.
The 2023 dynamic to watch: platform engineering is becoming a distinct discipline with distinct hiring profiles and tooling ecosystems (Backstage, Port, Cortex, and others). Organisations that conflate platform engineering with DevOps or SRE will staff it wrong. The discipline requires product thinking applied to internal tooling — a different capability than infrastructure management or reliability engineering.
Sustainability as a cost question
Cloud cost and carbon footprint are converging. This is not coincidental — the same architectural decisions that reduce compute cost tend to reduce energy consumption. Efficient workload scheduling, right-sized instances, reduced data transfer, and regional deployment close to compute consumers all reduce cost and carbon simultaneously.
For regulated industries in the UK and EU with Scope 3 reporting obligations (Streamlined Energy and Carbon Reporting, Corporate Sustainability Reporting Directive), cloud infrastructure is increasingly a reportable item. The cloud providers' sustainability calculators and carbon dashboards are becoming inputs to the reporting process.
The practical implication: sustainable architecture is increasingly just good architecture. Teams that have been applying cloud cost engineering principles — idle resource elimination, workload scheduling, architecture tiering — are already most of the way there. The gap is measurement and reporting, not fundamental redesign.
IoT and OT convergence
In manufacturing and environmental services, the gap between IT systems and operational technology is narrowing. OT — the PLCs, SCADA systems, and sensor networks that control physical processes — was historically isolated from enterprise IT for security and reliability reasons. That isolation is becoming harder to maintain as real-time operational intelligence becomes a competitive requirement.
The convergence is happening at the edge: compute capacity is moving onto the factory floor, into the water treatment plant, and onto the field sensor network. This makes it possible to process operational data locally — enabling real-time process control decisions — while selectively forwarding aggregated and filtered data to cloud systems for analytics, reporting, and ML model training.
The security and reliability requirements for OT are not the same as for enterprise IT. Patching cycles, availability requirements, and failure modes are different. Organisations that treat OT/IT convergence as a straightforward IT connectivity project will encounter these differences at the worst possible time. The organisations that succeed in 2023 are the ones that bring OT expertise into the architecture conversation from the start — not after the first network incident.