AI Adoption: The Last Mile

Integrating AI at work is as much a social and behavioral problem as a technical one. Leaders need a mindset that focuses adoption on real business challenges, adding value, while maintaining employees’ trust and engagement.

In a recent MIT Sloan Executive Education webinar, Senior Lecturer Paul McDonagh-Smith, considered the final journey organizations must take to move beyond model-building toward fruitful adoption of AI-assisted systems, embedding AI into operations, culture, and behavior—the last mile.

Models to mindset

It is not enough to have good algorithms. Developing models—be they from Open AI, Microsoft, Google, Meta, or others—is only a starting point. For AI adoption to produce sustainable impact, an organization’s leadership must develop a mindset that encompasses technological, organizational, and human adaptation.

………………………………………………………………………………………………….

Join Paul McDonagh-Smith and colleagues on MIT Sloan’s: ‘AI Essentials: Accelerating Impactful Adoption’ program

Dates: Dec 10–11, 2025 ⃒  Format: 2 days, in-person ⃒  Location: Cambridge, MA

 ………………………………………………………………………………………………….

McDonagh-Smith opens by considering the model, asking those leading AI adoption to reflect on the evolution of AI (from original AI to machine learning to deep learning to generative AI). This reflection has two purposes: first to understand what is out there now and what we might need for our organization’s specific circumstances; and secondly to raise awareness of the evolving a nature of AI and appreciation of how the various constraints that faced the technology at each stage led to its ongoing development.

With this comes his key message that leaders need to embrace a mindset of “exploration, experimentation, and evolution,” to facilitate “organizational natural selection.” As biological evolution moves from simplicity to complexity, companies need to “define simple organizational rules so they can better react to the evolution process we are going through.”

 Metrics to impact

Continuing the Darwinian theme, McDonagh-Smith suggests we are in a period of ‘punctuated equilibria,’ when long periods of stability are interrupted by short, rapid bursts of intense evolutionary change. In this situation, legacy KPIs will not be sufficient to “shift the mindset.” New metrics of value will be needed, measuring for example: how people are performing better due to AI, or the velocity from decision making to implementation, or autonomy and the percentage of work run automatically vs. that requiring human intervention.

Last mile AI engineering

It is in the final stages of AI design and adoption that organizations often fail. To address this, McDonagh-Smith introduces the concept of ‘last mile AI engineering.’ This he says is “fundamentally about reducing the space between AI’s potentiality and real-world impact.” This means connecting AI technology to human capabilities—creativity, curiosity, critical thinking, collaboration, emotional intelligence, passion, etc.—to “unlock AI’s potentialities.”

5 key principles of last mile AI engineering:

1. Decision first: Success depends on framing and targeting the problem to be solved accurately at the outset, and creating a disciplined process that integrates human and AI capabilities to deliver value.
2. Human-centered co-design: The AI project team must involve everyone—cross-function peers, senior leaders, front-line workers, and other stakeholders—in the design and adoption process.
3. Context above compute: Leaders should see AI not only as a technical enabler, but as a transformation tool, thinking not what AI could do but on what it can practically do in their context—using but augmenting existing knowledge.
4. Duality of thinking: Think small scale and big-picture. Adopt iterative, experimental approaches. Take small steps, learn fast, and scale up where initiatives are getting traction.
5. Trust by design with continual oversight. Apply good governance to AI implementation through accountability, alignment with organizational values, and risk mitigation. Crucially governance is also about enabling continuous improvement.

Mind the gap

“Don’t fall between the capabilities of AI intelligence and human intelligence. Bring the two together,” warns McDonagh-Smith. This involves understanding the correct context for AI in your organization. To avoid push-back leaders need to recognize when people accept and welcome AI and when they will reject it believing the human touch is essential.

Principles for strengthening AI adoption:

1. Fit to context: Deploy AI where it outperforms humans and use human intelligence for higher-level, strategic, creative, or judgement-based work.
2. Earn trust with truth: Avoid hype and overpromising, set realistic expectations, and rely on accurate metrics to move from pilot to adoption.
3. Human-centered orchestration: Clarify the roles and accountability between humans and AI in the workflow, defining the points of intersection and handoffs, and making overrides and optouts easy.
4. Sharing the dividends: Where there have been measurable productivity gains through deploying AI, share the dividends with employees—in terms of less time pressure, bonuses, upskilling.
5. Governance by design: Ensure good data pipelines, infrastructure, monitoring, and mechanisms to get feedback from users to continually improve models and their use.

………………………………………………………………………………………………….

This article is based on a recent MIT Sloan Executive Education webinar: From Models to Mindset: The Last Mile of AI Adoption

………………………………………………………………………………………………….

Dates: Dec 10–11, 2025 ⃒  Format: 2 days, in-person ⃒  Location: Cambridge, MA

 ………………………………………………………………………………………………….

Join MIT Sloan expert faculty for the 2026 ‘AI Executive Academy’

Dates: Mar 23–Apr 3, 2026 ⃒  Format: 10 days, in-person ⃒  Location: Cambridge, MA