I spent the day at t3 Summit and the topic du jour was, of course, AI, with a focus on governance and security. And I think I found a good analogy: I’ve been likening AI to power tools for a while already—but the distinction between tools and power tools was often unclear. We already give people tools in the workplace, whether that’s a ticket system or an office suite. But taking an analogy to everyday life, these are really more like appliances: coffee makers, not power drills. Certainly not buzzsaws!
An employee bringing a personal table saw into the workplace would invite questions about its suitability, safeguards, and use. Moving company work into a private “shadow workshop” would be recognisable as a change in the working environment.
The digital equivalent can feel like a shortcut: email a confidential company spreadsheet to a private account, then upload it to an unsanctioned ChatGPT account to finish a report.
The governance boundary is crossed before the AI produces an answer. You’ve heard of Shadow IT—be very afraid of Shadow AI.
Increasingly powerful AI tools are available largely independently of technical background. An employee can acquire the capability to build automations or direct agents without necessarily having the knowledge to assess the resulting data flows, access requirements, or failure modes. Friendly desktop and mobile interfaces compound that gap.
The physical power-tool comparison offers other lessons: Table saws have guards; some designs add automatic stops. Industrial machinery uses interlocks, designs safety in, and requires training. Cars have seatbelts, airbags, and other safety systems. These protections exist because competent operators still make mistakes.
AI adoption needs access, competence, and engineered safeguards together. Expecting “100×” productivity—or a fivefold improvement in a specific workflow—requires more than access to a subscription and a prompting course.
Teams need relevant knowledge of information architecture and security, supported tools that meet their needs, and timely specialist help. Otherwise, demanding more output while making the approved route impractical creates incentives for shadow AI.
Governance must therefore cover both the tools and the movement of work:
- Make clear what data can go where, backed by proportionate transfer controls.
- Provide usable managed environments and appropriately scoped access.
- Enforce authorisation for consequential actions outside the model.
- Limit the scale of mistakes and provide effective stopping and recovery mechanisms.
- Assign ownership and support the people operating the workflows.
Broad AI access is almost guaranteed, and can be valuable. But it only becomes dependable when there are safeguards—and when the sanctioned route is practical enough that employees do not feel the need to invent their own shadow workshop.
Leave your buzzsaws at home.