A Head of Strategy at a major NZ company described his employer's AI approach in one sentence last week: "Implement Copilot to shut the team up". Deploy it, tick the box with the board, all to keep everyone quiet and the AI noise at bay. No enablement, no M365 licenses for SharePoint semantic context, no Copilot Studio, no training budget.
They're now starting a new role at a competitor in April.
Microsoft's latest Cyber Pulse report (February 2026) paints a picture: 80% of Fortune 500 companies now have active AI agents running, and 29% of employees have already turned to unsanctioned AI agents for work tasks.
The checkbox strategy doesn't stop AI adoption, it creates 'black market' type behaviour, pushing it to the shadows.
This is not an anti-Microsoft argument. Copilot is a valid starting point, and even meeting recaps create value. The argument is that leaders treating deployment as the finish line instead of the starting line is where the strategy dies, because while these organisations have been ticking boxes, the frontier moved.
The frontier is moving every week
While teams have been stuck planning governance frameworks, optimising SharePoint permissions, and waiting for IT to build the perfect agent...
Anthropic released Sonnet 4.6 today with full upgrades across code generation, computer use, long context reasoning, and agentic planning. The average Microsoft user doesn't even know Anthropic models are available to them (or why they're useful).
OpenAI launched Frontier the other week, an enterprise platform where HP, Intuit, Oracle, State Farm, and Uber are deploying autonomous AI agents at scale.
Codex can run end-to-end workflows autonomously for hours.
Claude Code loads 17 context files simultaneously, processes entire project directories, and returns finished work.
Matt Shumer's "Something Big Is Happening" hit 80 million views last week for a reason: the majority of knowledge workers have no idea what the current generation of AI tools can actually do.
Most are still rewriting emails and uploading documents. And copy pasting between interfaces, or building Copilot 'Agents', GPTs, Gems, or Projects.
The real shift is delegating entire tasks and getting finished work back, on tools most organisations have never heard of, let alone sanctioned. I rebuild my complete workflow every few weeks to adapt to the new capabilities as they arrive.
The contrast is not complicated.
One category of tool assists: it sits in the workflow, suggests completions, summarises meetings.
The other category operates: it takes a brief, processes information across multiple sources, executes multi-step work, and returns a finished deliverable.
Most organisations have deployed the first and have no awareness the second exists.
OpenAI calls this the "capability overhang": the widening gap between what AI tools can do and how people are actually using them. The typical power user uses 7x more thinking capabilities than the typical user on the same paid plan. OpenAI's own employees use 15x more. The overhang doesn't close on its own, and the report is clear: the biggest gains come when users move beyond basic chat into repeatable, multi-step workflows. From asking questions to delegating work.
Microsoft's own research calls the top 5% of companies by AI adoption "Frontier Firms". Their workers send 6x more AI messages than the median, and 71% say their company is thriving vs 37% globally (Microsoft Work Trend Index, 2025).
These organisations didn't just deploy a tool and call it a day. They've intentionally embedded AI into how work gets done, across functions, not just within IT.
The rise of the Frontier Team.
The prescription is not ripping out Microsoft and starting over. The prescription is building a Frontier Team: a small, inter-functional or cross-functional group given access to frontier tools inside a governance framework. Skip the IT skunkworks and the data science team running experiments in isolation. This needs representatives from marketing, operations, finance, customer service, and product: people who understand the actual work.
The Frontier Team gets three things most organisations withhold:
Budget for frontier tool licences (Claude Pro, Cursor, Codex)
Protected time to test real workflows, not theoretical use cases
A mandate to report back on what works and, importantly, to share the experimentation failures
They map the terrain so the broader organisation doesn't have to navigate it blind.
This borrows from what Microsoft's Frontier Firms and OpenAI's Frontier platform are already demonstrating at enterprise scale, but microscoped down to something a mid-market business can actually execute. Five to eight people, one quarter, governed access, real work, license to learn.
The critical piece most conversations miss: this is not exclusively a leadership problem.
Leadership creates the conditions, provides the budget, removes the blockers, builds the governance framework. But capability doesn't develop by mandate.
BCG (June 2025) found employees want approximately five hours of hands-on training, and only 36% say the training provided is enough. 78% of AI users are already bringing their own tools (Microsoft Work Trend Index, 2025).
The curiosity already exists but the infrastructure to channel it productively does not.
The individual piece is equally important. Waiting for the company to provide the perfect AI environment is a losing strategy. The people building capability now, experimenting on their own time, learning to work with frontier tools, those are the people leading teams in 18 months. They showed up curious and built the muscle before it was required.
Most people are still stuck in the void.
Using web-based ChatGPT, copy-pasting documents into Copilot, or waiting for IT to build something. That is not the future of AI-augmented work.
The future looks like delegating complex, multi-step workflows to tools that run autonomously, grounded in organisational context, governed by clear frameworks, and accessible to knowledge workers across every function.
The Head of Strategy I spoke with last week understood all of this. His employer's response was to deploy Copilot and walk away. He's taken initiative, found a new path.
The board may have AI in the quarterly report but the person who understood what to do with it will be building capability at a competitor in to erode market share before the next quarter begins. There's a talent leakage problem materialising in every organisation that doesn't shift its focus to enabling frontier teams.
Start small.
Pick five people across five functions.
Give them frontier tool access and one quarter to test real work.
The cost is negligible compared to the Copilot licences already sitting unused.
Every organisation gets to decide: is the AI investment a starting line, or just a receipt that kept everyone quiet for another quarter.
Passionate about all things AI, emerging tech and start-ups, Mike is the Founder of The AI Corner.
Subscribe to The AI Corner
