Singular Machines has been spun out from the UK Atomic Energy Authority with backing from the UK Innovation and Science Seed Fund, Oxford Science Enterprises, Arup, and Japan’s Miraisozo Investments. Arup has also contracted to use and help develop coEngen, the company’s agentic AI platform for coordinating complex engineering work.
The technology grew from engineering activity around fusion programmes, where mechanical structures, electrical systems, cooling, controls, materials, diagnostics, safety requirements, and maintenance constraints have to be developed in parallel. Changes made by one discipline can alter loads, interfaces, access, thermal behaviour, cost, or compliance elsewhere in the design, leaving substantial work in reconciling models and documentation as a project evolves.
CoEngen is intended to automate part of that coordination while maintaining an auditable engineering trail. The platform uses multiple software agents working around a shared data model rather than treating engineering as a sequence of isolated text prompts, with Singular Machines focusing on the provenance of inputs, assumptions, and outputs as designs change.
That is a different requirement from general-purpose generative AI. An engineering result may influence equipment dimensions, structural loading, power requirements, control architecture, procurement, or safety documentation, and a plausible answer is of limited use if a reviewer cannot determine how it was produced. Engineers remain responsible for final decisions, so any useful automation layer has to expose its assumptions rather than conceal them behind fluent output.
The company’s first substantial commercial relationship gives that proposition a broader test. Arup has invested in Singular Machines, signed a contract to use coEngen, and will provide technical, strategic, and commercial support while the platform is developed against real engineering workflows. That moves the software beyond its fusion origins into projects where disciplines, software environments, clients, data standards, and assurance requirements differ considerably.
UKAEA’s Culham environment nevertheless provides a demanding starting point. Fusion engineering involves large, long-lived facilities in which designs evolve while specialist teams work concurrently, and late discovery of an interface conflict can affect equipment, schedule, cost, and buildability. A system capable of reducing repetitive coordination work without weakening configuration control has an obvious application beyond fusion.
Singular Machines was incorporated in 2024 and remains a comparatively young business. The value of its latest pre-seed financing has not been disclosed, so the more informative measure of progress will be how much of coEngen’s automation can be repeated across different engineering programmes rather than how large a funding headline can be attached to the company.
The platform also enters an engineering-software market that already contains simulation packages, product lifecycle management systems, digital twins, optimisation tools, requirements-management platforms, and specialist design applications. Large organisations carry years of project data and established approval processes inside those systems, making wholesale replacement improbable. An agentic platform is more likely to succeed by connecting existing engineering tools and reducing manual transfers between them.
That integration brings its own technical burden. Permissions, version histories, nomenclature, design baselines, review gates, and change control all have to remain intact if automated agents are allowed to move information or initiate engineering tasks. Outputs also need to be sufficiently transparent for specialists to challenge them, particularly where data are incomplete or several engineering objectives conflict.
Singular Machines has already received external recognition for the approach, with coEngen winning the advanced-manufacturing category of Innovate UK’s Agentic AI Pioneers Prize earlier in 2026. The award provides validation of the concept, but commercial deployment with Arup will be the more demanding test because the platform will have to save engineering effort under normal project constraints rather than within a technology competition.
For UKAEA, the spinout adds another route for converting publicly developed engineering capability into an independent industrial business. For Singular Machines, the immediate task is narrower: show that agents can coordinate meaningful engineering work while preserving the evidence trail on which design reviews and professional responsibility depend. Automation becomes considerably less impressive once nobody can explain why the machine changed the design.
