Karl Fischer, CMO, Executive Head: Strategic Digital Services at DVT. Recall the scene in Fantasia: Mickey Mouse, acting as the sorcerer's apprentice, enchants a broom to carry buckets of water. The plan is brilliant. But things turn: the broom multiplies. And multiplies again. Before long it's a flood, the brooms are well past anyone's control and Mickey has no way to stop them. Foreseeable? Probably. What's the lesson here for AI? What may start as a useful (and necessary) experiment can escalate very quickly at scale, particularly when that scale and velocity are well beyond the typical management capability of existing processes and the responsible human beings.Most organisations are actively implementing AI. They are running pilots, deploying tools and experimenting with automation and agents. There is real pressure (and FOMO) to move fast. One question often comes too late, and sometimes not at all: who owns the decisions once AI is making them? The question is even more critical when those decisions are part of an agentic-enabled solution; the system doesn't just recommend, it acts. When decision-making is automated with AI, it happens continuously and at machine speed. Human behaviour naturally tends to limit the scale of mistakes. People hesitate. They may have "a feeling that something is off". Hence, HITL (human in the loop) is often a recommended safeguard as you evolve your agentic automation.Such measures are essential to address risk that may lie in factors previously perhaps not given as much consideration in technical solution implementation: bias, ethics, policy and regulations. If a decision turns out to be wrong, the system does not carry the responsibility. That lies with the leaders who approved its deployment.Leaders in organisations that succeed with AI consequently ensure:AI governance and policies are in place to ensure ethical and responsible AI use.Risk management is given focus alongside technical experimentation and innovation.Change management and cultural alignment are actively addressed throughout the AI journey.Readiness is addressed with respect to foundational elements: data, platforms, tools, security, skills and talent.Vision and commitment are clear (and measurable): what, where and how will value from AI be achieved.Holistic process and cross-sector views are considered to find, prioritise and realise value from AI and automation.These elements effectively put in place the answers to "how" you will undertake the AI journey in your organisation.In a recent McKinsey webinar, Robert Levin, Senior Partner, McKinsey, stated: "[... ] the differentiating capability is not the 'what' of a transformation. Peers in a given industry usually share a similar view on how to create value through AI. The differentiation is the 'how engine', your ability to predictably, consistently turn your attention to the important capabilities and know that you'll be able to build, adopt and scale them to value.""Rewired to win: Reimagining the enterprise with tech and AI" (https://www.mckinsey.com/featured-insights/mckinsey-live/webinars/rewired-to-win-reimagining-the-enterprise-with-tech-and-ai#)Apart from the guidance the "how" elements provide, essential risks associated with AI adoption and process automation are also addressed. As examples:ElementHelps to AddressMitigatesAI Governance and PoliciesUse of unsanctioned AI toolsExposure of IP, customer data and corporate data in the public domain.Risk managementCompliance and security requirementsProtection of PII, adherence to GDPR, POPIA (South Africa), AI legislation. Actively addressing attack vectors that may surface through public facing or connected agentic interfacesChange management and cultural alignmentFailure to achieve value from investment in AIFailing to achieve effective and productive adoption of AI tools and capabilitiesExperimentation with AI tools in support of both learning and establishing a foundation in a business is essential. That experimentation is far more effective and valuable when it is guided by clearly communicated answers to:Why is AI being used?What AI is planned and authorised for use?Who owns AI success, and what are the meaningful measures of success?What are the time-frames for adoption and enablement?What are the anticipated impacts and what are the outcomes to be avoided?Leaders who have the answer to "who is responsible for AI decisions" will be the ones who put governance and ownership for AI in place early. It is also essential that those leaders demonstrate personal commitment to the process. Ultimately, it will prove far easier to manage a few well-understood, well-behaved "brooms" than to attempt to regain control once they have multiplied beyond your very human limits.