On 24 September 2026, Arrowfly’s AI Engineering Board held its inaugural meeting, bringing together leaders from engineering, manufacturing, healthcare and research to discuss how organisations can turn artificial intelligence into measurable business value. The discussion focused on moving AI beyond experimentation towards practical applications with measurable results.
The board brings together industry representatives, including Alex Zhavoronkov, founder, CEO and chief business officer of Insilico Medicine, alongside engineering leaders from companies such as BP, Procter & Gamble, Tesla and DuPont. Arrowfly’s editorial leadership convenes and supports the board. Members represent organisations across manufacturing, energy, technology and healthcare, providing perspectives on the practical challenges of applying AI in engineering.
AI and the Engineering Board
Companies typically evaluate new technologies against factors such as cost, implementation time, operational fit and expected return on investment. However, the rapid pace of AI adoption can encourage companies to invest before they fully understand the potential benefits or likely return on investment. The board’s discussions returned to fundamental questions: which problems are worth solving, how success should be measured, and how engineering knowledge can be preserved.
While much of the AI debate focuses on advances in models and software, the board highlighted another important factor: people’s willingness to change established working practices. This cultural adjustment is proving as impactful as the technological developments themselves.
The discussion emphasised using AI to complement human expertise and critical thinking, rather than treating it as a prerequisite for innovation or a replacement for workers. Organisations need people willing to explore new possibilities, alongside experienced practitioners who can assess their practical value.
Beyond standalone tools to integrated workflows
Organisations are also reconsidering their expectations for AI capabilities. Organisations are increasingly interested in moving beyond standalone chat interfaces towards AI tools that connect engineering systems and coordinate work across disciplines. This reflects growing interest in agentic AI systems that can interact with existing software and carry out tasks across established workflows.
Protocols such as the Model Context Protocol (MCP) provide a standardised way for AI applications to connect with external tools and contextual information. In engineering, such integrations could connect natural-language instructions and AI-generated code with workflows for creating or modifying design geometry.
The opportunity extends beyond individual design tasks. Board members identified three priorities: reducing routine work for specialists, providing context across disconnected information systems, and coordinating processes across those systems. Achieving these goals can be difficult because of legacy equipment, limited infrastructure and incompatible vendor systems.
Prioritising the problem over the tool
With new AI models and tools emerging rapidly, experimentation can easily become a series of trials without a clear objective. The Arrowfly AI Engineering Board refocused the discussion on fundamental questions before organisations committed to specific implementations. Key considerations include the intended user, their objectives, and the specific bottlenecks they face.
These questions are not new; they have underpinned established methodologies such as lean manufacturing and Six Sigma for decades. Applying these principles to AI projects helps teams define success metrics before a pilot begins and assess whether the system delivers a genuine improvement.
Balancing innovation with business discipline
The ease of building AI tools allows employees to improve their own workflows. However, an application that benefits one person or a small team may not justify the cost of deploying and maintaining it across an organisation. The discussion highlighted a shift from broad experimentation towards closer scrutiny of AI usage, resource consumption and costs.
This increased oversight leads to demands for clearer use cases before resources are committed. For engineers still exploring AI’s potential, requirements to justify each project may feel restrictive. Yet, without a clear connection to business needs, activity can expand faster than its demonstrated value.
One practical approach is a lightweight review process in which employees explain the proposed application, its intended users and the problem it is designed to solve before development begins. This method aims for deliberate choices without subjecting every experiment to a lengthy approval cycle. Costs extend beyond AI usage. Organisations also need staff to develop, test and maintain supporting applications. Limited engineering capacity can therefore constrain promising projects, particularly when employees must also maintain existing systems.
Deciding which projects not to pursue is as important as selecting those worth developing. This concept, while not new, gains particular relevance in the rapidly evolving AI landscape, where strategic selection is crucial for tangible progress and avoiding dispersed efforts.
Data readiness and knowledge preservation
The discussion moved beyond data availability to a more practical question: whether the information is sufficiently clean, complete and reliable for the intended AI application. Engineering organisations often hold decades of designs, revisions and historical records. The challenge is determining which information is relevant and reliable enough for a particular application.
Key questions include whether older designs remain relevant, which version of a record is authoritative, and whether a dataset accurately represents the work it is intended to inform. These questions make data selection an engineering judgement rather than a simple retrieval exercise. Validation is equally crucial; access to information does not automatically guarantee its accuracy.
The board cautioned against making data preparation so extensive that it delays useful work. Focusing on a particular application first can narrow the required information, making data and project readiness a more manageable task. For example, a design-review application could begin with a defined set of checklists rather than requiring the organisation to structure every engineering record in advance.
Preserving reasoning, not just records
A critical question emerged during the discussion about data readiness: what engineering knowledge never makes it into formal records? Research datasets may contain numerical results but omit the observations, experimental conditions and contextual information needed to interpret them correctly. These details, while obvious to the original worker, are indispensable for training an AI agent.
Pressure to innovate quickly can worsen this problem. Faster documentation, including AI-generated records, does not necessarily preserve all the details needed for later interpretation. This concern connected with a framework discussed during the meeting: artifacts, dependencies and decisions. Artifacts are the objects themselves and their associated metadata.
Dependencies describe relationships between objects—for example, how a component relates to a requirement or how a requirement maps to a test. Decisions record the rationale behind a particular choice. Each adds a different kind of context.
Knowing which records exist does not reveal how they relate to other work, and those relationships do not necessarily explain why a decision was made.
Decision rationale can be particularly difficult to recover because it is often scattered across emails, meeting notes and design reviews rather than captured in structured records. For organisations aiming to leverage AI for accumulated engineering experience, simply making records accessible is insufficient. Preserving the relationships and reasoning that render them meaningful becomes imperative.
Physical AI and safety considerations
As AI becomes more integrated into robotics and other physical systems, the consequences of system behaviour become more immediate and tangible. The discussion highlighted questions about safety guarantees, system latency, integration and governance. A robot’s ability to perform a task is only one factor in determining whether it can operate safely and effectively around people and equipment.
Similar concerns apply to AI-assisted engineering tools used to develop safety-critical products. Because AI systems can produce probabilistic or incorrect outputs, their use in consequential engineering decisions requires appropriate safeguards, validation and human review. Safety and governance must therefore be built into the engineering process, with human oversight retained where decisions carry significant consequences.
This repeatedly underscored the importance of human expertise across industrial applications and physical AI. Professionals are essential for assessing AI output, recognising its limitations, and determining the sufficiency of evidence. They provide the critical human layer required to ensure responsible and effective deployment of AI in physical environments.
Arrowfly’s broader “AI for Engineers” initiative
The AI Engineering Board forms part of Arrowfly’s wider AI for Engineers initiative, launched on 30 September 2026. This year-round resource aims to provide engineers and technical leaders with practical knowledge, peer connections, and professional guidance as AI reshapes the engineering profession.
Arrowfly, which owns engineering media brands including Design World and The Robot Report, says its network reaches 2.1 million verified industry readers.
The initiative comprises four components: a dedicated News Desk, an annual conference, an original research programme and an advisory board. The News Desk will cover AI tools, workflows and implementation decisions across engineering disciplines. The inaugural AI for Engineers conference is scheduled for 20–22 September 2027 at the M Resort Spa Casino in Henderson, Nevada, near Las Vegas. Arrowfly expects approximately 700 engineers and technical leaders to attend.
The conference programme will feature five tracks covering engineering design and automation, medical devices and medtech, robotics, energy and electronics, and industrial operations and smart manufacturing. The programme is designed to include practical AI workflows and structured peer-to-peer discussions. The research programme will include the 2027 Outlook of AI, a survey examining AI adoption across five engineering disciplines.
Arrowfly plans to publish the survey findings in a report during the fourth quarter of 2026, covering adoption, implementation barriers and expectations for AI through 2030. Eligible engineers who complete the survey will receive a complimentary pass to the 2027 conference, subject to Arrowfly’s qualification criteria. The advisory board is intended to help ensure that editorial coverage, conference programming and research priorities reflect practitioners’ needs rather than vendor interests alone.
Towards practical and evidence-based AI adoption
The value of the inaugural meeting lay in bringing together different perspectives to identify practical ways of improving AI implementation. Participants discussed the potential of connected engineering workflows alongside the practical challenges posed by legacy infrastructure.
The discussion also balanced the freedom to develop new applications against the resources required to test, maintain and support them. Calls for enhanced data access evolved into deeper questions about comprehensive documentation, relational data, and the underlying decision rationale. This approach emphasises grounding AI implementation in operational requirements rather than theoretical capabilities alone.
For engineering professionals evaluating AI projects, the discussion offers several practical starting points. These include defining problems precisely, understanding the specific information required for each application, and rigorously preserving the context that gives that information genuine meaning. The emphasis is firmly on practical application and demonstrable results.
The board also emphasised validating AI outputs against defined objectives and using evidence, rather than novelty, to guide further investment. The meeting also demonstrated the potential value of a cross-industry forum in which practitioners can examine and challenge ideas while drawing on experiences from different sectors.
The goal is not simply to keep pace with AI developments, but to help engineers make informed decisions about where and how to apply the technology. This approach places the emphasis on practical value, sound engineering judgement and responsible AI adoption rather than experimentation for its own sake.
