The conversation around AI in business is shifting. The core challenge for leadership is no longer understanding how the algorithms work. Instead, the real work lies in figuring out where the technology creates value without breaking the company or alienating the workforce.

To guide an organisation through this transition, we might look past the marketing hype and manage six interconnected realities:

  • The Strategic Shift: Moving from a technology-first to a problem-first mindset.
  • Depth vs. Surface Adoption: Recognising the difference between using a tool and redesigning a workflow.
  • The Automation Curve: Understanding that AI integration is a spectrum, not a binary switch.
  • The Data Foundation: Recognising that an AI strategy is often a data strategy in disguise.
  • Governance & Risk: Balancing enterprise innovation with considered guardrails.
  • The Human Element: Redefining the social contract to ease the fear of replacement.

1. The Strategic Shift: From Hype to Problem-Solving

Leaders often confuse buying a new tool with transforming their business. True optimisation requires shifting the lens away from the technology itself and focusing on the business model.

Instead of asking "What is our AI strategy?", organisations might ask "What are our operational bottlenecks, and can this technology solve them?" Success is better tracked through cycle-time compression, cost reductions, or the generation of new revenue streams — not vanity metrics like "hours saved."

Buying enterprise licences for tools like ChatGPT or Microsoft Copilot is surface-level adoption. It speeds up individual tasks but leaves the core business model unchanged, creating a false sense of security where leadership assumes they have addressed the AI transition.

2. Depth vs. Surface Adoption

Grassroots adoption is valuable. Across organisations right now, proactive teams are using AI to streamline admin and reclaim hours of their week. But the challenge for leadership is taking those individual gains and translating them into deep, systemic value.

Klarna's AI rollout illustrates the difference. Had they stopped at surface adoption, they would have deployed a chatbot to answer FAQs. Instead, they integrated AI deeply into their enterprise APIs, enabling it to autonomously process refunds and manage returns — resolving two-thirds of their customer service chats and reducing resolution times from 11 minutes to under 2 minutes.

3. The Automation Curve

  • Level 0 — Programmed Automation: Traditional rules-based software. Rigid; breaks when context changes.
  • Level 1 — Assist: AI as a cognitive sidekick. The human drives the workflow. This is where the "bolt-on" trap lives.
  • Level 2 — Assemble: AI does the heavy lifting of preparing multi-step workflows. The human reviews and approves.
  • Level 3 — Authorise: AI is given decision-making authority within strict boundaries.
  • Level 4 — Autonomise: AI operates against long-term strategic goals, handling daily adjustments while humans audit quarterly performance.
  • Level 5 — Networked Autonomy: Cross-functional AI agents communicate directly with each other.

The leap from Level 1 to Level 2 is where most organisations stall, because it requires internal systems to be connected and readable by machines.

4. The Data Foundation

An AI strategy is often a data strategy in disguise. AI cannot function if internal knowledge is locked in messy formats or isolated silos. Auditing and cleaning data architecture is a prerequisite to moving up the automation curve.

5. Governance, Risk, and Guardrails

This means governing how proprietary data is used when interacting with external language models, implementing testing and human-in-the-loop protocols for high-stakes decisions, and staying ahead of compliance requirements around automated decision-making.

6. The Human Element: Culture and Fear

When people fear for their jobs, they may use AI for shallow tasks to look compliant while actively avoiding deep integration into their core work. Employee fear typically spikes at Level 3 — the moment the tool stops being a sidekick and starts executing tasks independently.

Leadership language might pivot from "efficiency" (which employees often translate as "layoffs") to "augmentation." To achieve deep adoption, the company needs to provide a personal incentive for the employee — such as upskilling or moving to higher-value work.

Conclusion

The organisations that adapt successfully are unlikely to be those that simply purchase the highest volume of software. Advantage will belong to those that deliberately align these tools with their core business problems, build the necessary data foundations, and thoughtfully bring their people along on the journey.

The shift from human-driven tasks to agentic workflows presents a tangible change in the social contract of work. Navigating it asks us to focus closely on the psychology of responding to change — not just the mechanics of the technology.