Large language and generative AI foundation models have been improving by between 30% and 100% every eight months, depending on the metric, with notable gains across comprehension of domain data, reliability, planning, workflow generation, and analytical reasoning.
In the workplace today, work is increasingly being offloaded to these models.
So far, this has mostly been informal, with prompts like “write me a response” or “pull the requests from this email,” but it is now being embedded directly into enterprise applications. Examples include extracting key steps from project specifications, summarizing service reports, and suggesting next actions from maintenance logs.
Together, these represent the creation and enrichment of vital business data from semi-structured inputs, making AI a direct participant in enterprise workflows rather than just an assistant outside them.
Increasingly, the reasoning faculty of foundation models is being used to stretch AI application to wider tasks. You ask a model to solve a business problem, check its reasoning, and let it use tools to carry out the plan it creates.
This goal-oriented, tool-using software is known as an agent. Nearly every enterprise technology company is developing them, though their levels of autonomy and reasoning differ. Given that these systems underpin manufacturing, logistics, finance, and customer operations, it is natural that automation is the next step.
The next 20 years will re-define the relationships between people, assets, data, processes and the enterprise systems which coordinate them entirely.
Enterprise software will stop just being the place where work is done and increasingly become the thing doing the work, reasoning and executing. And why wouldn’t it? Large-scale data centers are already approaching the electricity consumption of small countries, with global demand expected to reach nearly 1,000 TWh by 2030, around three times the UK’s usage in 2023. It’s inevitable that some of that intelligence will be directed toward enterprise work. The real question is how we prepare for it, and how we decide what thinking belongs to people and what can safely be delegated to software.
The Shift Starts with Context Engineering
A lot of what we call digital transformation has been about connecting systems. What comes next is about connecting context.
Context engineering is the practice of designing systems so that data, applications, and AI models can understand and act on the same reality, the same underlying data. It is what allows information to flow with meaning. It is what lets one agent’s action inform another, what allows AI agents to reason across boundaries, and what ensures humans can stay in the loop when it matters most.
Key elements include:
- Data accessibility
- Semantic alignment
- Security boundaries
- Relevance optimization
- Human oversight
- Feedback loops
A lack of context engineering, or poor implementation, can severely limit agents. Knowing which tool to use means little if they can’t identify the right data, understand its relationships, or act on it correctly. Add in security, technical, or legal restrictions, and you risk agents that fail or worse, act unpredictably.
Much of what we are already doing at IFS supports this goal. We have moved to an API-first architecture, using RESTful APIs and OData for integrations, interfaces, and data exposure. Alongside this, our Model Context Protocol (MCP) servers provide context with data, making enterprise applications orchestra table. Together, these steps create accessible data for users, partners, internal teams, and agents, making systems connected, connectable, and ready for the next generation of Industrial AI. They engineer the context.eri bağlantılı, bağlanabilir ve yeni nesil Endüstriyel Yapay Zeka’ya hazır hale getirir. Bağlamı tasarlarlar.
From Interfaces to Interactions
A clear sign of the shift towards letting systems do some of the reasoning and orchestration on our behalf is the slow erosion of the traditional user interface. The UI is an artefact of how we used to work with software. Many of our preferred systems start with an instruction or search bar which we fill with natural language to start a process, saving clicks.
Our digital workers, our agents, reduce the need for UI even further. They communicate through MCP and act as orchestration nodes in an event-driven architecture. They would much rather call well documented APIs than remember the 15 clicks required to get something done with UI-loadtime slowing them down.
From an onboarding and change management perspective, learning a UI has become an unnecessary burden in most change projects. If we can replace it with natural, human interaction, systems we can talk to, guide, and collaborate with, we make technology simpler and much more effective.
When agents can handle enough goals, tools, and context, will we still need a UI at all?
Industrial AI vs Commodity AI
Not all AI carries the same level of risk. In consumer or low-stakes business contexts, getting it mostly right can still be acceptable. In industrial operations, that is not good enough. Industrial AI is high risk by nature because decisions affect assets, safety, productivity, and sometimes lives.
Whilst we will need to see improvement in areas beyond context, consistency and predictability being two, this is another reason why context matters.
Industrial AI must be grounded in system-of-record logic, shaped by industry domain, asset type, geography, and regulation. An agent managing maintenance schedules or dispatching field crews operates inside a world of compliance and consequence. Context engineering ensures that world stays consistent, traceable, and explainable. It is the difference between useful autonomy and unacceptable risk.
The Human and AI Balance
Over the next decade, industrial companies will be run by AI and humans, but always led by people. Digital workers will take on structured, repeatable, and time-sensitive tasks, while humans focus on creativity, ethics, and innovation, the things that make us human: teaching, persuading, leading, designing, and storytelling. Give it another five to ten years and even the environments we work in will change. The only enduring reason for a screen may be to talk to someone who is not sitting next to you. Everything else will be driven by intent, conversation, and context with digital workers taking instruction seamlessly and executing autonomously on our behalf. Offices will return to being places of discussion and creativity, not rows of people clicking through applications.
The Next Era: Context Engineering as the Bridge
We began by building systems to record what happens, the first digital ledgers of business. Then came systems to plan what should happen, coordinating resources and processes at scale. Over time, those systems grew dense with features, integrations, and custom code, powerful but often rigid and hard to change. Each generation brought more capability, but also more complexity. The next generation of ERP will make things happen, autonomously completing work based on live data, events, and policy.
In industries like manufacturing, construction, aerospace, and energy, this shift will redefine how work, value, and resources interact. Factories will adjust production based on AI-managed energy availability. Projects will reschedule around materials and weather. Predictive maintenance will become coordination between digital workers, robots, and engineers. Aircraft will plan maintenance around performance and demand. Energy grids will balance for cost and carbon. Every process will form part of a system that senses, decides, and acts as if guided by a set of musicians, each aware of the others’ rhythm.
This is the payback from context engineering: the connective tissue that allows intelligence to move safely and meaningfully through and between enterprises. It bridges the gap between the system of record and the system of action, between data and decision, between today’s ERP and tomorrow’s autonomous enterprise.
The One Big Takeaway
The biggest advantage any industrial organization can take right now is to start investing in context engineering, making systems and data connected and connectable so intelligence can flow where it is needed. Strengthen data foundations. Prepare systems for event-based interaction. Pilot digital workers. Build trust within workforces by showing these technologies are explainable and well governed. This is the groundwork for high-impact Industrial AI. The choices made today will decide how confidently we step into the next era of human and digital collaboration.
SOURCE: Matt Ely (2025 November 24) Context Engineering: From ERP to AI and the Future of Work.IFS Blog. https://blog.ifs.com/context-engineering-from-erp-to-ai-and-the-future-of-work/
