Picture an operations manager at a mine or a plant at 7am on a Tuesday morning. Before she has made a single decision, she has opened the ERP, the maintenance system, a production dashboard, two spreadsheets and a WhatsApp thread from the night shift, and reconciled them in her head to work out what actually needs her attention today. Down the corridor, a procurement clerk is chasing the same overdue purchase orders by e-mail that he chased last week. Upstairs, a salesperson is rebuilding a customer's history from six systems before a 10am meeting.
None of this is an AI problem. It is friction, and the organisation has quietly made its people the integration layer between its systems.
That is the picture I keep returning to after 4Sight's recent From AI Vision to Frontier Business event. The ambition in the room was real, and it is real everywhere. In Microsoft's 2025 Work Trend Index, 82% of leaders call this a pivotal year to rethink strategy and operations, and 81% expect agents to be integrated into their AI strategy within 12 to 18 months. The same survey finds 80% of the global workforce short of the time or energy to do their work, with employees interrupted every two minutes. The distance between those two findings is where I think the real opportunity sits, because AI amplifies whatever it is built on. Strong foundations get amplified. So do broken ones.
Digitised is not connected, and connected is not intelligent
Most businesses have spent a decade digitising. They own ERP, CRM, HR and operational platforms and sit on enormous amounts of data. But digitised does not mean connected, and connected does not mean intelligent. Fragmented data produces fragmented AI. A process that 10 people run 10 different ways gives an agent nothing consistent to follow. And a capability people do not trust will not be adopted, however good the demo.
Our ERP session framed this as four foundations for trusted AI: data, process, operations and people. Is there one version of the truth? Is the process standard enough to hand to an agent? Can a recommendation become a transaction? Will people actually use it? Those are unglamorous questions, and I would argue they are the questions. They turn “Where can we put AI?” into “What problem are we solving, what stops us today, and what outcome would prove we succeeded?”
The encouraging part is that the answers are usually local. Not a transformation programme, but the Tuesday morning reconciliation, the overdue purchase orders, the customer history. Fix the data, process and ownership behind one of those moments and the intelligence has something solid to stand on.
From systems that record work to systems that help do it
For decades, enterprise technology has recorded what happened: transactions, customer activity, workforce information, plant data. The shift AI makes possible is from systems of record to systems of execution: technology that does not just flag a problem but interprets it, assembles the context for a decision, starts the right workflow and brings a person in where judgment or approval belongs.
The procurement scenario we demonstrated made this concrete. A request typed in plain language is clarified by an agent, completed with the missing details and prepared for approval; a second agent watches purchase-order lines and surfaces the ones that need chasing. No supplier is contacted and no requisition is raised without a human saying yes. The clerk is not replaced. He stops chasing and starts deciding.
Two things make that safe rather than reckless. The first is context. A general model knows what a purchase order is, but not your suppliers, approval limits, stock position or who is allowed to see what. We showed how open standards such as Model Context Protocol let an agent work inside the ERP's existing security model, so that access which does not exist in the ERP is not conjured up by the AI. The second is a division of labour I thought was neatly put at the event: AI provides knowledge, people provide expertise and governance provides the control that turns the two into execution.
Islands, bridges and the person in the middle
The risk I see most often is not too little AI but AI islands: a chatbot in one department, a Copilot in another, an ERP add-on, an RPA bot, a spreadsheet macro. Each improves a task; the process between them stays broken; and the operations manager is still doing the reconciling at 7am, now between smarter tools.
The alternative is connected AI, and the principle behind it fits on one line: systems provide context, AI interprets and reasons, automation executes reliably and people stay in control of the decisions that matter. It changes the question from “What can this tool do?” to “What outcome can we deliver end to end?”
Customers feel the difference first. They experience one company, not a marketing platform, a sales platform and a service desk, and the failures they remember happen in the gaps between departments rather than inside them. Employees feel it next. At the event, we showed leave, payslips, tax certificates and policy answers reached through Microsoft Teams and WhatsApp, because nobody should need to understand the application landscape to complete an everyday task.
The last bridge is to the individual. As organisations grow, knowledge scatters across systems, documents and departments, and we ask each person to find what is relevant to their role. 4brain, our vision for role-based organisational intelligence, turns that around: the organisation's intelligence understands the role and brings the relevant context to the person. A finance leader, a salesperson and a shift supervisor need different signals. The aim is not more information. It is a clearer view of what matters, why it matters and what should happen next.
Governance is the accelerator, and the plant is part of the business
As assistants become agents that complete tasks, governance matters more, not less. Access must respect permissions, actions must be accountable and auditable, and answers must come from approved sources. The governance agent we demonstrated does something modest and useful: it gives employees natural-language answers on HR and company policy, grounded in the approved documents. Governance is often cast as the brake on AI. My own view is the opposite. Done well, it is what lets a leadership team say yes with confidence, and it starts with data, because every Copilot, agent and automated decision inherits the quality and governance of the data beneath it.
For a South African business, the opportunity also does not stop at the office door. In mining and industrial environments, enormous value sits in operational information: equipment condition, production, energy and water, safety and quality. Our OT session connected AI to asset reliability, process productivity and remote and autonomous operations. The ambition is to bridge the systems of record and the plant on trusted data, not to optimise them separately.
Frontier is a direction, not a finish line
Microsoft describes the destination as the Frontier Firm, a business that runs on intelligence on demand through human-agent teams, and sketches three phases on the way: AI as an assistant, agents as digital colleagues working at human direction, and finally, humans setting direction while agents run whole processes. The report is candid that the journey is not linear and that many organisations will be in all three phases at once. That is why we speak of a Frontier Business rather than a Frontier Firm: the same journey, applied to the whole enterprise, plant included.
The Work Trend Index also reports that 71% of leaders at Frontier Firms say their company is thriving, against 39% of workers globally. One can read that as cause or as effect, and the firms that qualify were already mature, so I would hold the number lightly. What I take from it is narrower: the organisations that have embedded intelligence in how work gets done are not the ones that ran the most pilots. They are the ones that were ready to scale what worked. Others will argue that speed matters more than readiness, and there is something in that. In my experience, speed without foundations mostly produces impressive demonstrations.
So the most practical idea from the event was also the smallest: start with one moment that matters. Find somewhere a customer, an employee or a shift team feels friction. Fix the data, process and ownership behind it. Add the intelligence or automation that fits. Measure. Then decide what deserves to scale. The boardroom conversation changes with it: not “Are we doing AI?” but did customers get answers faster, did managers decide earlier, did the plant run better, did we create capacity for growth?
Because AI is not the outcome. A better-performing business is. The organisations that hold on to that distinction are, I believe, the ones most likely to make the journey from AI vision to Frontier Business. And it begins, more often than not, with someone's Tuesday morning.
Source: Microsoft, 2025 Work Trend Index: The Year the Frontier Firm Is Born, April 2025 (31 000 workers, 31 countries).
Share
Editorial contacts