Author: Matt Robson
Key Takeaways:
- LLM-based AI agents are moving beyond text generation to autonomously execute complex engineering tasks, running simulations and iterating on designs to dramatically accelerate development cycles.
- In R&D, LLMs are acting as creative "co-scientists," synthesising vast amounts of information to propose novel materials and hypotheses, reducing discovery timelines from years to days.
- Company-tuned co-pilots are embedding compliance criteria and institutional knowledge directly into engineering workflows.
Introduction
Imagine waking up to find a full day's worth of complex engineering simulations already completed, optimal design candidates highlighted, and the next detailed analyses automatically launched.
This is the near future of engineering, enabled by large language models (LLMs) evolving from passive predictors into active co-pilots. LLMs are already streamlining tasks such as content creation, code generation, and the summarisation of complex documents… and maybe even helping to write blog posts... But in engineering, the real opportunity isn’t just faster text or code generation, it’s embedding these AI systems directly into design and validation workflows.
At Monolith, we see this as the decisive next step in the integration of AI into industry.
In hyper-competitive fields like battery development and automotive manufacturing, AI co-pilots are already planning manufacturing and testing, operating specialized software, documenting critical decisions, and contributing to creative problem-solving on par with human experts.
Agentic execution: elevating engineers from tool operators to AI directors
This transformation is driven by the AI agent; by combining sophisticated reasoning with the ability to act, these advanced LLMs execute tasks, review outcomes, and determine next steps toward a user-defined goal.
In software development, tools like Claude Code already develop entire applications autonomously from simple design briefs using LLM agents.
However, hardware engineering presents greater challenges due to the complexity of translating language to physical designs and data scarcity.
Nevertheless, agentic workflows are automating traditionally time-consuming simulation tasks, including computational fluid dynamics (CFD) and finite element analysis simulations, with reported accuracies from simulation benchmarks exceeding 80% in simple cases.
This is achieved through multi-agent systems, where a team of specialised AI agents collaborates in a 'division of labour' approach to the workflow.
This approach unleashes continuous, autonomous iteration. Engineers define objectives and constraints, and an AI co-pilot runs simulations overnight, analysing results, identifying promising designs, and launching further simulations. Engineers arrive at a full workday's progress already completed, dramatically accelerating development.
The potential of these techniques also extends far beyond automation of single-task simulation workflows.
Imperial College London researchers offer a glimpse into this future with a new framework for mechatronics design, which coordinates a team of specialised agents covering planning, mechanical CAD, CFD simulation, electronics, and software alongside a human engineer. This system successfully designed a functional water-quality monitoring vessel, demonstrating how LLM agents can manage entire design processes, spanning physical components, multi-domain constraints, and verification.
Illustrative multi-agent engineering workflow for the design of a simple, hypothetical brake duct. The planner coordinates Design, Airflow/CFD, Materials, and Manufacturing agents; the right panel shows a short example chat.
These advancements profoundly democratise highly specialised technical skills, particularly in materials R&D. Complex simulation techniques like DFT and MD, critical for discovering new materials for batteries and fuel cells, were once exclusive to PhD-level scientists.
That barrier is now disappearing as LLM agents begin to master these techniques. In one case, a multi-agent system performing DFT analysis on crystal lattices achieved average errors of less than 1% compared to human experts.
Ultimately, these developments will fundamentally change how engineers interact with their software:
- From manual to conversational: Moving beyond clicking menus and writing scripts to a goal-oriented partnership.
- From operators to directors: Engineers transition from expert tool operators to creative directors, responsible for guiding development, verifying AI outputs and continuously improving system performance.
- Accelerated design: Guiding intelligent systems to explore, validate, and execute complex designs with unprecedented speed and scale.
Creative partners: uncovering insights hidden in plain sight
Beyond automating well-defined tasks like simulation setup, LLMs are beginning to function as genuine creative partners in R&D. Their ability to process and synthesise vast quantities of information, from academic papers to the immense datasets they were trained on, allows them to identify patterns and draw connections that human researchers might miss.
In vast and fast-moving fields like battery research, where the volume of new information is impossible for any individual to track, this capability is becoming critical to advancing the field.
Google has recently demonstrated an AI "co-scientist," a Gemini-based system that reviews scientific literature to generate, rank, and propose testable hypotheses. In a collaboration with Imperial College London, it independently reproduced the discovery of a new antimicrobial-resistance mechanism in just two days, a feat that took human researchers nearly a decade to achieve.
This signals a fundamental shift to AI as a creative partner, a potential now being realised in materials science, where recent works have shown that an AI can generate novel compositions for battery electrodes and electrolytes by reviewing existing literature and making connections to adjacent fields of research.
However, the impact of these breakthroughs will scale massively when they move from academic literature to the lab, combining with an organisation's own test data and domain expertise.
At Monolith, we believe this is the key: grounding AI's creative hypothesis generation in real-world test results transforms promising ideas into validated solutions.
The logical next step is to fuse AI's creative power with its simulation capabilities and connect them directly to physical automation, closing the loop from initial idea to final experiment.
This blueprint is leading to the rise of the fully automated, 'self-driving' laboratory, where AI co-pilots manage the entire research cycle to dramatically increase research throughput.
Pioneering platforms like the A-Lab, a collaboration between Lawrence Berkeley National Laboratory and Google DeepMind, are already making this a reality. By pairing AI algorithms that design new materials with robotic systems that synthesise and test them, they have created a closed-loop discovery engine.
This approach promises to dramatically accelerate R&D, allowing scientists and engineers to focus more on guiding the direction of innovation.
Expert co-pilots: guarding your engineering integrity
For engineering teams, what matters most is trust: knowing AI isn't a black box but a system they can audit, validate, and rely on. Monolith ensures every AI recommendation can be traced back to data, not just probability.
Beyond vendor tools, a 2025 report from the German Research Centre for Artificial Intelligence (DFKI) showcases four production-oriented deployments where LLM agents are already delivering actionable and auditable intelligence for diverse, real-world challenges.
Examples range from optimising production planning and enhancing drug safety monitoring to ensuring sustainable textile design meets strict standards, proving LLMs are moving into the core of industrial operations.
By embedding organisational knowledge directly into workflows, these expert co-pilots deliver value in three critical areas:
- Automated QA: Co-pilots act as vigilant QA engineers, automatically parsing design changes and flagging gaps against internal standards and external regulations, complete with clause-level citations.
- Design for manufacturability: They ensure new designs are viable from the start by surfacing critical shop-floor data, such as machine capabilities, approved materials, and historical production challenges, guiding engineers towards proven, efficient paths.
- Preserving institutional memory: Most importantly, they can capture the invaluable expertise of senior staff. When an expert departs, their knowledge remains, converted into standard operating procedures and training modules. This transforms decades of experience into a searchable, resilient asset that accelerates onboarding and ensures continuity.
LLMs promise faster, safer answers, reducing rework and audit friction. This leads to designs better aligned with factory capabilities and preserves collective wisdom for future innovators.
Monolith: From AI Models to Engineering Co-Pilots
At Monolith, we believe the engineering co-pilot isn’t just a vision of the future—it’s already taking shape in today’s test labs.
Our no-code AI platform helps engineers test less and learn more, with proven tools for anomaly detection, test plan optimisation, and calibration that cut costly validation cycles and surface insights hidden in data.
Engineering teams at companies like Vertical Aerospace and Nissan are already using Monolith to shorten development timelines and accelerate innovation.
Looking ahead, we’re expanding these capabilities toward the agentic workflows discussed in this article.
Our research focuses on integrating multi-agent frameworks and model context protocols (MCPs) with Monolith’s machine learning tools, building the foundation for autonomous research agents that can plan, execute, and validate engineering workflows.
We’ll be sharing more on this in a series of upcoming webinars—exploring how agentic AI can reshape engineering validation and research. Stay tuned for details and join us as we continue building the next generation of engineering co-pilots.
Conclusion
The journey from predictive text to engineering co-pilot is a rapid ongoing evolution across distinct, interconnected roles. LLMs are emerging as diligent agents accelerating design and simulation, creative partners charting new research frontiers, and expert guardians preserving institutional knowledge and ensuring compliance.
While each capability offers a significant leap in productivity individually, their true transformative power lies in convergence. A future where AI proposes a novel material, autonomously validates it via simulation, designs a product, and ensures regulatory adherence is an implementation challenge being solved today.
By offloading manual execution, exhaustive research, and procedural lookup, LLM engineering co-pilots free human engineers and experts to focus on framing problems, applying critical judgment, and guiding innovation.
The era of the engineering co-pilot amplifies human skills, creating partnerships poised to solve previously intractable problems with unprecedented speed and ingenuity.
“The future of engineering co-pilots will not be won by generic AI platforms, but by tools built for the realities of testing, validation, and compliance. That’s where Monolith is focused — turning vision into practice for engineers today.
