Are you becoming a frontier firm, or simply adding AI to broken work?

Claudia Correia de Araujo, Business Development Lead at Global Micro Solutions.

---

Management wants an AI strategy and executives want a measurable return, and that pressure is producing rapid investment across South African enterprise and government. Global Micro Solutions argues that the evidence of transformation sits somewhere less visible. Inside the work itself.

"I have watched organisations spend heavily on new technology while their people continue working around the same broken processes, poor governance and fragmented data and then watched leadership teams describe that as digital transformation and progress," says Claudia Correia de Araujo, Business Development Lead at Global Micro Solutions. "It is one of the more expensive habits in business, and AI has made the trap easier for companies to fall into."

The questions that reveal the truth are operational ones. How quickly can someone reach the information needed for a decision? How much time disappears into disconnected systems and approvals? Has AI removed effort, or added another responsibility?

Microsoft's June 2026 Becoming a Frontier Firm defines the destination through four objectives, which are to enrich employee experience, reinvent customer engagement, reshape business processes and bend the curve on innovation. Its 2026 Work Trend Index shows the raw capability is already here, with 58% of AI users producing work they could not have produced a year ago, rising to 80% among frontier professionals.

What the numbers cannot show is whether the organisation around that capability is actually changing. In Global Micro Solutions' experience, the difference comes down to a sequence of decisions, each one setting up the next, and it starts with where leadership looks first.

Most AI programmes begin with a hunt for tasks and use cases to automate. It is visible and fairly easy to demonstrate, which is precisely why it can be dangerous. Why? Because weak processes are preserved instead of being reconsidered, enhanced or, even better, re-engineered entirely.

"I have sat in enough digitisation programme reviews to recognise the pattern," says Correia de Araujo. "Leadership sees new technology and assumes service delivery, customer engagement and revenue management must be improving. But when you trace the work with the people who actually do it, you often find the same broken process underneath.

"AI does not repair that process. It accelerates it, scales it and makes its failures visible in the output. Inefficiencies that people once compensated for manually begin to surface as inconsistent results, while weak data, security vulnerabilities and governance gaps become impossible to ignore. The result is not transformation. It is a faster, more visible version of the same problem."

Technology can accelerate a workaround and also spread its risks. Leaders need to follow the work from beginning to end, asking where a service request stalls and why staff have to reopen it. A frontier firm begins by asking why the work happens this way and what outcome it is trying to improve, and the people best placed to answer are the ones doing the work themselves.

Yet those are the very people most digital programmes overlook. Employees are expected to experiment and adopt AI while carrying the same targets and workload, and then watch every recovered hour fill immediately with something else.

"What leadership interprets as resistance is often a perfectly rational response to very poorly organised change," says Correia de Araujo. Only 13% of AI users surveyed in Microsoft's 2026 research said they were rewarded for reinventing work with AI, a contradiction that will stall transformation regardless of how capable the technology becomes.

The same research cites a 17-point lift in reported AI value when managers actively model its use. That matters more than another adoption presentation, and it leads to the harder question.

Once agents take on real work, where do people stay involved?

Continuous human involvement is a design decision. Global Micro Solutions sees four broad patterns of human-agent collaboration. A person may remain the author, act as an editor, direct a complete task or orchestrate several agents while handling exceptions. Each process needs a pattern that fits its risk profile. Repeatable work can be delegated, while sensitive or costly decisions need a person close to them who can act with confidence.

In South Africa this is also a legal and compliance question. Section 71 of POPIA generally restricts certain decisions with legal consequences or a substantial effect on a person when based solely on automated profiling. The Act provides exceptions and safeguards, including an opportunity for the affected person to make representations and, in relevant circumstances, receive sufficient information about the underlying logic.

"Human approval at the end offers little protection when the reviewer lacks authority to challenge the result AI has produced," Correia de Araujo notes. "Accountability has to be designed through the process. The organisation should be able to say what an agent may do, what remains under human control and who answers when something goes wrong."

The answer to whom answers gets harder as agents multiply; we now have agents managing agents. As they take on more execution, people move into supervision, which raises a management question that deserves far more attention.

How many agents can one person responsibly direct and supervise?

"No executive would hand a manager 40 direct reports without serious consideration, yet that is how many organisations are deploying agents, minus the job descriptions," says Correia de Araujo.

Microsoft named this the human-agent ratio and was careful to call it critical and task-specific. No published benchmark exists, and the right span depends on the agent's autonomy, the consequences of failure and the authority of the person who remains accountable. An agent summarising approved documents presents a different supervisory burden from one changing citizen records or releasing payments, and review time, access changes, exception handling and retraining all consume real capacity.

The number is only half the question. The people directing agents need real technical depth, solid design thinking and business analysis capability, and a working grasp of the legal and compliance exposure attached to every output they approve.

"These AI managers are constantly in investigative, verification and validation mode," says Correia de Araujo. "They must trace sources, corroborate evidence, distinguish fact from inference and understand the consequences of error before trusting an output. Defence, intelligence and forensic professionals already work this way, combining human, signals and open-source intelligence into a shared picture and assessing it at tactical, operational and strategic levels. AI supervision will require a similar discipline, yet most organisations have not begun training for it."

Global Micro Solutions advises organisations to test this capacity by simulating exceptions and measuring how quickly a supervisor can intervene. "A successful run tells you that the workflow worked once," says Correia de Araujo. "An exception tells you whether the management model works when it matters the most."

BNY provides a useful example. Its key digital AI employees are treated like staff, with defined identities, supervisors and auditable activity. In one payment-validation process, work that took five or six minutes fell to under 30 seconds, while open investigations dropped by nearly 80%. The result came from narrow scope and controlled access, backed by clear ownership, governance and leadership.

None of this shows up in usage and adoption figures. Someone can use AI daily while correcting their work and moving information between systems. Leadership should follow the complete process and compare lifecycle time with rework. "If every efficiency becomes a heavier workload, employees will understand what the promised benefit was really for," says Correia de Araujo.

AI should also create capability the organisation did not have before. At Levi Strauss, knowledge was spread across more than 400 technology systems. In one day, a finance agent analysed 1 100 standard operating procedures and catalogued 18 000 tasks, work that would otherwise have taken multiple people close to a year.

Microsoft's research found that organisational conditions account for 67% of reported AI impact, against 32% for individual factors, which places the larger determinant of value in how the organisation is managed. A better workflow should be reused and a failed output should change the next attempt. Everything the organisation teaches AI about its standards and judgement becomes intelligence that compounds.

A frontier firm can name the outcome it is pursuing, point to a process that works better and show where human judgement remains non-negotiable. Three questions give a quick temperature reading on business evolution.

Do your AI reports measure active users, or does the effort actually remove from core processes? Can the supervisors closest to your agents state each agent's limitations and their own span of responsibility? And when an AI-assisted decision is challenged, can you show the audit trail of who held the authority to change the outcome?

"If those answers come easily, then you have a good foundation for AI transformation and it will show in the numbers," says Correia de Araujo. "If they do not, then you need to go back to basics as the work underneath has not changed. That gap will not close on its own."

Twelve months from now the distance between the organisations that redesigned their way of work to be frontier and the ones that simply equipped it will be visible in cost, in risk and in the quality of their decisions. "That distance is being set now, in choices that look small," she says. "Which process to follow end-to-end, which supervisor to give real authority, which number to put in front of the management. Decisioning intelligence is taking on a whole new meaning in this new world of human-AI collaboration."

Every question raised in this article can be answered with evidence. Global Micro Solutions' Frontier Firm Assessment translates these disciplines into a practical maturity assessment aligned with Microsoft's three-phase Frontier Firm journey: human with assistant, human-agent teams, and human-led, agent-operated work. It assesses each priority workflow across the operating model, workforce and technology.

The assessment establishes how work actually flows, where people carry friction, which processes require redesign before AI is introduced, how human judgement and agent supervision will coexist, the data governance and integration needed and what the current evidence proves. It delivers an honest baseline, identifies the highest-value opportunities and provides a sequenced roadmap for turning agentic AI into measurable business value.

The process begins with a 30-minute discovery call. That is usually enough to establish which questions the organisation can already answer and where its most significant maturity gaps exist.

Claudia Correia de Araujo is a Microsoft 2026 Frontier Club Most Valuable Professional and Frontier Captain, a Microsoft Sales Titan and a Certified Microsoft AI Transformation Leader. Connect with her on Linked-In or Email to discuss your organisation's frontier firm strategy, solutions and results.

Sources

Microsoft: Becoming a Frontier Firm methodology, June 2026

Microsoft 2026 Work Trend Index Annual Report

Microsoft 2025 Work Trend Index: The year the Frontier Firm is born

Microsoft WorkLab: The making of a Frontier Firm at BNY

Microsoft WorkLab: How Levi's is refashioning its company with AI

Protection of Personal Information Act 4 of 2013, section 71