AI Strategy Needs a Practice, Not an Adoption Plan
Over the past two weeks I have been exploring how AI intensifies work rather than reducing it, and how it shifts the bottleneck from production to human judgement. Feedback tells me a lot of people recognise these patterns in their own experience. The intensification. The growing queue. The subsequent time constraints driving to an acceptance of “good enough.”
But recognising a pattern is only useful if it changes what you do next. And for most organisations, what comes next is still framed as an adoption challenge. Get more people using the tools. Measure usage. Track hours saved. The logic feels sound. The problem is that it addresses the wrong question.
The Challenge to Adoption as a Plan
Gallup’s workplace data from Q4 2025 tells an interesting story. Just over a quarter of US workers now use AI at least weekly. Almost half have never used it at all. The headline adoption figure has barely moved. But the intensity among committed users keeps climbing. The people who have found value are going deeper. The question is whether organisations are designing around their experience or still trying to push the half who have not started.
Here in Australia, the RBA’s 2025 survey of firms found that AI adoption has been largely piecemeal. Often employee-led rather than employer-led, returns are described as mixed. Many firms are still focusing on AI as an enabler of “cost-out”. This puts further pressure on delivering against the wrong metrics. Some firms are prioritising the identification of high-impact use cases, whilst many are still searching for where AI fits. Few are building around where it is already working.
The gap is also widening between those who set AI strategy and those expected to execute it. Gallup found that 69% of leaders reported using AI, compared to 40% of individual contributors. They conclude that this is a use-case problem, not a technology problem. It measures whether people are using AI. It says nothing about whether AI is making the work better.
From Adoption to Practice
The distinction matters. Adoption asks: are people using the tools? Practice asks: how does AI fit into the way work actually gets done?
If AI intensifies work, then getting more people to use it without redesigning how work flows will just intensify further. If AI moves the bottleneck to human judgement, then the organisational response cannot simply be faster review. It has to be a different workflow. One that protects human capacity and applies it where it creates the most value. It also has to include the quality of the human input itself. How well people brief AI, how effectively they provide context, how sharp their judgement is about what constitutes a good output. These are learnable skills. And as they improve, the upstream quality of what AI produces rises, which directly reduces the rework and review burden that builds when AI is bolted onto unchanged processes.
PA Consulting argued recently that the real shift happens when leaders stop asking “where can we add AI?” and start asking “if AI is a permanent capability, how would we redesign our processes from scratch?” PwC’s 2025 Global AI Jobs Barometer makes the same case from the data. Rather than accelerating adoption, their recommendation is to redesign workflows from a blank sheet, deliberately sequencing AI and human effort rather than layering one on top of the other.
Research published this week by Harvard Business School and Microsoft identifies the same pattern. They call it the ‘last mile’ problem: pilots succeed but gains do not scale because the organisation around them has not changed. The obstacle is rarely the technology. It is operating model design.
This is the shift from a plan to a practice. A plan says “we will adopt AI across the organisation.” A practice says “we will redesign how this workflow operates, with AI handling production and humans focused on the judgement that determines whether the output is good enough.” The first is a technology rollout. The second is an operating model change.
Leaders Cannot Design What They Have Not Experienced
There is a clear prerequisite to building an AI practice. The people responsible for designing how AI fits into workflows need to have personally experienced what AI does to their own daily rhythm and workflows.
If you have not felt the bottleneck building, the queue of AI-generated output waiting for your review, you will not understand why workflow redesign matters. If you have not experienced the pressure to approve rather than engage, you will not grasp why protecting review capacity is a strategic priority.
Ibarra and Jacobides identified in HBR five critical skills leaders need in the age of AI. The foundational one was developing AI fluency through personal, daily use. Not delegating it or reading about it in a briefing note. But using it, visibly, in meetings and decision-making. McKinsey’s 2025 State of AI survey reinforces this: organisations that are high performers in AI are three times more likely to have senior leaders who actively role-model its use. The connection between leadership engagement and outcomes is measurable.
I have felt the bottleneck. I understand why it matters. A leader who has opened Copilot twice does not have that understanding. And without it, the workflows they design will likely default to bolting AI onto existing processes and hoping for efficiency.
What Practice Design Looks Like
In practical terms, practice design starts with a single workflow. One process where AI is already generating output and a human is making judgement calls on that output.
Map it. Identify where AI is producing and where a human is reviewing. Ask whether the human is positioned to engage properly, with adequate time and context, or whether they are processing a growing queue. If the answer is the queue, you have an adoption strategy. You do not yet have a practice.
The measurement framework from last week applies here. Track decision quality, not hours saved. Monitor rework rates as a signal of how well AI is being applied. Watch the review burden on your most experienced people. These are the diagnostics that tell you whether AI is improving outcomes or just increasing volume.
Where I Am With This
I am still working this out for myself within my own work. What has changed is that I am more deliberate about where I apply my review effort and more focused in how I brief the production side, so that what comes back for review is closer to where it needs to be. This is the part of the practice that is easiest to overlook. The quality of what arrives for review is shaped by the quality of the interaction that produced it. Getting better at that interaction, better prompting, better context, clearer direction, is not a soft skill. It is the mechanism that relieves the bottleneck.
I am also using AI to help me examine how I am working. Which workflows are structured well and which ones leave me processing a backlog? It turns AI from the thing that creates the bottleneck into a tool for redesigning around it. This is very much a work in progress.
That shift, from reactive to deliberate, is small. But it is the difference between being shaped by AI and shaping how AI fits into the work. The companies that get this right will not be the ones that adopted AI fastest. They will be the ones that built a practice around it.
Sources
Gallup (2026). “Workplace AI Tracker, Q4 2025.” January 2026.
Reserve Bank of Australia (2025). “Technology Investment and AI: What Are Firms Telling Us?” RBA Bulletin, November 2025.
PA Consulting (2026). “AI in 2026: How Organisations Will Change as They Continue to Embed and Scale.” February 2026.
PwC (2025). “2025 Global AI Jobs Barometer.” June 2025.
Lakhani, K.R., Spataro, J. and Stave, J. (2026). “The ‘Last Mile’ Problem Slowing AI Transformation.” Harvard Business Review, March 2026.
Ibarra, H. and Jacobides, M.G. (2025). “5 Critical Skills Leaders Need in the Age of AI.” Harvard Business Review, October 2025.
McKinsey (2025). “The State of AI in 2025: Agents, Innovation, and Transformation.” November 2025.