A technology developed around the unusually demanding engineering requirements of fusion research is moving into commercial industry, as Singular Machines spins out from the UK Atomic Energy Authority with investment from four backers and a contract to test its agentic AI platform inside global engineering consultancy Arup. Rather than applying generative AI primarily to documents or office workflows, the company is targeting complex engineering projects where decisions need to remain traceable and professional judgement cannot simply be delegated to a model.

Singular Machines is developing coEngen, a multi-agent engineering platform intended to coordinate work across different technical disciplines through a shared data model. UKAEA describes the system as designed to preserve traceability, assurance, and engineering judgement while automating parts of complicated project workflows. The company says its target markets include complex machines, facilities, and infrastructure, taking an approach that emerged from fusion engineering into industries with similarly demanding design and assurance requirements.

The company has completed a pre-seed investment round involving the UK Innovation and Science Seed Fund, managed by Future Planet Capital, Oxford Science Enterprises, engineering group Arup, and Japan’s Miraisozo Investments. The value of the round has not been disclosed. Arup has also agreed a separate contract to use coEngen and will provide technical, commercial, and strategic support while helping Singular Machines test the platform against real engineering workflows.

That customer relationship gives the spin-out a more useful test than a funding announcement alone because engineering automation has to work within processes where an apparently plausible output is not sufficient. Designs for energy systems, infrastructure, robotics, or industrial equipment pass through requirements, calculations, reviews, interfaces, safety cases, revisions, and formal approvals. Any AI system operating inside that chain has to leave enough evidence for engineers to understand how a proposal was produced and why it should be trusted.

Agentic AI enters high-assurance engineering

Agentic systems are designed to pursue tasks through sequences of actions rather than responding to every step through a direct human prompt. In an engineering setting, that could allow specialised software agents to work across requirements, geometry, analysis, component selection, documentation, and verification while coordinating information that is currently passed between teams through a mixture of specialist tools, meetings, files, and manual checking. The productivity opportunity lies in reducing coordination overhead rather than attempting to remove engineering expertise from the process.

Singular Machines’ origins in fusion research are relevant because fusion projects concentrate many of the conditions that make engineering automation difficult. Systems combine extreme physical environments with intricate mechanical, electrical, materials, control, and safety requirements, while design changes in one subsystem can propagate through many others. Engineers also need to establish why decisions were made and how a design evolved, particularly where components are expensive, bespoke, or difficult to change after construction.

Innovate UK has already recognised coEngen through its Agentic AI Pioneers Prize, describing it as a multi-agent platform that brings engineering disciplines together through a shared data model and supports traceable optimisation of complex systems. That emphasis on traceability distinguishes the proposition from much of the first wave of enterprise generative AI, where adoption has focused on writing, summarisation, coding assistance, and information retrieval. Engineering automation requires outputs to connect back to formal requirements and evidence rather than merely appearing credible to a user.

Professional liability also changes the deployment model. An engineer signing off a design cannot reasonably defend a decision by saying an AI system generated it, while organisations operating safety-critical infrastructure need to preserve accountability even when software carries out more of the analytical work. Systems therefore need mechanisms for human review, version control, verification, and escalation, alongside clear boundaries around which decisions software can make independently.

Public research looks for a commercial route

The spin-out forms part of UKAEA’s attempt to turn technology and intellectual property developed through national fusion research into businesses serving markets beyond fusion. Its 2026–2030 strategy places greater emphasis on commercialisation, reflecting the fact that decades of public investment have built expertise in robotics, materials, advanced manufacturing, simulation, control systems, and complex engineering management that can have nearer-term industrial applications even while commercial fusion remains under development.

That route gives public research organisations a mechanism for moving technology into markets that a laboratory is not structured to serve directly. Spin-outs can recruit commercial teams, raise outside capital, and sell products while preserving links to the scientific and engineering expertise from which the technology emerged. The difficult stage is usually the transition from an internally useful capability to something general enough that other organisations will pay to adopt it.

Arup’s involvement provides Singular Machines with an early environment for that transition. The engineering consultancy will use real workflows to help shape and test coEngen while exploring applications across complex projects. Its investment also means the relationship sits somewhere between customer, development partner, and shareholder, which can accelerate product development but will still need to demonstrate that the resulting platform works beyond the requirements of a single strategic partner.

Commercial engineering software is already crowded with established computer-aided design, simulation, product-lifecycle, digital-twin, and project-management systems, so Singular Machines is not entering an empty market. Its opportunity lies in coordinating work across those boundaries rather than attempting to replace every specialist tool, particularly where AI agents can connect information and decisions that currently move slowly between disciplines.

The Arup contract will provide an early indication of whether that proposition survives contact with production engineering, where reliability and auditability carry more weight than an impressive demonstration. If coEngen can automate coordination while preserving the evidence engineers need to approve work, fusion research will have produced a commercial technology aimed not at generating another engineering document faster, but at changing how highly interdependent design decisions are managed across an entire project.