Shadow AI is already running in your business

Leaders who treated AI as something to ship have pulled ahead of leaders who treated it as something to get their head around. The real use is happening on employees' screens, outside the official pilot programs.

Your employees are already "Secret Cyborgs"

Your employees are finding that AI makes them far more efficient. Your company is probably reporting only "small to moderate gains."

A large part of your workforce is probably using AI in secret.

Employees have rational reasons for the secrecy.

• Employees worry that showing their new efficiency will lead to headcount cuts for themselves or their colleagues.
• Some people who produce "incredible outputs" with AI hide their methods to protect a reputation for brilliance. They do not want to credit their AI partner.
• Many expect efficiency to bring more work without more pay or recognition.
• Vague threats about improper AI use and "terrifying blanket policies" push adoption into the shadows. That creates risks for data security and governance.

Your workforce may already be changing while your organization sees none of it.

Employees now run agents

The pattern matured in 2025. Secret cyborgs pasted prompts into ChatGPT in a closed browser tab. In 2026 the problem is shadow agents: employees spin up autonomous agents with Claude Code, custom GPTs, Copilot Studio or MCP-connected tools. These agents read company databases, call internal APIs and take multi-step actions, all outside IT governance.

This is a different class of risk. A leaked prompt is embarrassing. An agent with stale credentials that reads customer data and writes to your CRM is a breach waiting to happen. The AI-use policies most companies wrote in 2024 don't cover any of it.

Employees already run AI in production without your knowledge, outside the official pilots.

Leaders have to use AI themselves before they can judge it

Leaders have to close this gap. Ethan Mollick, an author and researcher, argues that leaders who delegate AI use, get reports back or only read about it do not get it. He puts it bluntly: unless you've had an existential crisis... you haven't got AI.

Real understanding comes from personal use that changes how you see work. Start with advanced prompting techniques or try the range of AI models yourself.

The deeper problem is that for decades, organizations have given up on radical redesign. The org chart was invented in 1855 to run a railroad, and it has changed little since.

Companies have handed this work to consultants, yet "nobody has special information about how to best use AI at your company." The playbook for this new world does not exist yet. You write it yourself, often starting with clearly defining user needs before you consider the technology.

AI behaves less like software

• Unlike traditional software, which only executes commands, AI "works like a person." It can argue, refuse tasks, and sometimes you have to justify to the AI why it should do a step. Managing it takes the same skills as managing a person.
• AI can narrow creative output. In a 2023 BCG experiment with GPT-4, consultants who used it for creative ideation produced ideas that were 41% less diverse as a group. Frontier models have improved, but the homogenization risk hasn't disappeared. AI can also raise the bar: AlphaGo beat the world's best Go players and inspired them to invent moves never seen before.
• AI can coach as well as do tasks. In a field experiment with Kenyan entrepreneurs, high performers who got advice from a GPT-4 mentor did just over 15% better, while low performers did about 8% worse. AI can also act as a "brutally honest team coach," analyzing meeting dynamics and communication patterns. One use is stopping a meeting to ask an AI whether it should continue.

AI changes how talent develops and what work is for

• Juniors who use AI to do their work may stop building the skills that normally come from doing it themselves. Managers find that AI "does a better job and never complains." Talent may stop growing through the hierarchy the way it used to.
• Employees will work out how to use AI on their own, so HR has to design the organizational structures, reward systems and coaching models for an AI-augmented workforce. HR also needs to understand how AI agents and workflows are reshaping roles.
• Teams need time set aside to reflect, or AI only helps them do bad or pointless work faster.
• The purpose of work is shifting. When a research report that took weeks takes minutes, "the bottleneck isn't the research anymore, it's figuring out what research to do." When code can be written instantly, as in projects like the AI-coded Desktop Assistant, the challenge becomes "understanding what to build." AI forces you to rethink why you do each task.

AI raises the average, so curation becomes the job

• AI raises average performers. Bad writers become average, and average developers become good. AI will force a recalibration of what counts as average, which pushes true experts to find new ground to stand out.
• Creators are becoming curators. They now select from multiple AI-generated options instead of starting from scratch. That raises a question: "Do we value curation as much as creation?" The risk is a "diminished sense of responsibility" for the output, so humans must act as orchestrators and evaluators who own what ships.

Leaders should start with three actions

A deliberate change management plan helps you shape an AI-first future where humans orchestrate the work. The CEO playbook and the operational framework are where to start. Use resources for AI development and consultation to get started.

• Use AI yourself before you judge its effect, because Ethan Mollick argues that leaders who delegate AI use or only read reports do not get it.
• Govern the agents as well as the chatbots, starting with the credentials each one holds and the systems it can write to.
• Own the unit economics of AI, because token use, growing context windows, agent loops and multi-model fallback each add cost, and those costs compound before anyone budgets for them.

If you want help with this, get in touch.