Agentic AI: opportunities and challenges – and how to navigate them responsibly

1. Start small, scale responsibly

Select a contained, non-critical use case with clear success measures. Run pilots with human oversight before handing over more autonomy. Learn and adapt. 

2. Keep humans in the loop

Even when the AI can act independently, build escalation protocols for high-stakes decisions. In other words: automation should never mean abdication. 

3. Plan for governance early

Establish cross-functional AI governance committees, with representation from compliance, legal, operations and ethics. Define policies for: 

• Acceptable use cases 

• Data handling and privacy 

• Escalation and incident response. 

4. Design with people in mind

Map the user journey, including employees, customers and partners. Factor in accessibility standards and cultural differences. Use frameworks that make AI reasoning visible so users can trust decisions. 

5. Test like it matters

Use diverse, structured testing with users of different profiles, roles and needs. Include scenario-based ‘red-teaming’ to stress-test against failures, biases, or adversarial inputs. 

6. Monitor and adapt

Deploy monitoring systems to track AI decisions and performance over time. Hold regular stakeholder reviews and be prepared to pause or roll back functionality if risks emerge. 

One creative tool we’ve explored is the ‘AI-sona’ – a twist on the design persona concept, used to explain the role and capabilities of AI agents to non-technical stakeholders. This simple reframing can prevent over- or under-estimating what the AI can do and helps teams think about agents as part of the workforce.