AI and work in brief
The issue. AI changes tasks before it necessarily changes whole jobs. The leadership question is which activities move to machines, which remain human-led and how roles, accountability and skills change around that boundary.
What is driving it. More capable agents, pressure for productivity, skill shortages and the cost of organisational complexity are moving the debate from tool adoption into job design and decision rights.
How the thinking is developing. Human-led redesign dominates the sources that meet both Plot axes. Baringa and FTI strengthen the human-partnership case, while McKinsey’s agent-manager and HR pieces make role compression and organisation redesign harder to ignore. Five useful labour-market or usage sources remain context because they do not express one sufficiently precise position on both axes.
What we can learn
Augmentation can still change roles, layers and staffing. Human-led does not mean preserving the existing organisation.
Decision rights matter more than job labels. A workflow needs a named human boundary for approval, escalation and accountability.
Transition costs belong in the business case. Training, supervision, exception handling and role redesign can absorb part of the apparent productivity gain.
Baringa, FTI, Accenture, Microsoft and WEF all emphasise redesigning work around human capability rather than treating AI as a simple labour-replacement switch. The common ground is about the design of work: roles, workflows, skills and decision rights change materially even where people remain central.
This cluster is not a promise to preserve every existing role, nor evidence that human-led redesign has already become normal enterprise practice. Augmentation can still alter management layers, staffing and the mix of skills required.
Where the thinking differs
The plotted sources disagree less on human partnership than on how far redesign should travel. Microsoft, Accenture and several McKinsey pieces move toward enterprise orchestration and changed decision rights; Baringa and PwC stay closer to end-to-end workflow redesign.
Labour-market and usage research from OpenAI, Anthropic, IMF and OECD remains useful context, but it answers related rather than identical questions. It has therefore not been forced into the two-axis map.
Plot’s interpretation: The crowded human-led side should not be read as ‘jobs are safe’. Plot’s interpretation is that much of the selected advice wants fewer unchanged jobs and more redesigned roles in which humans retain judgement, exception handling or accountability.
In practice
Baringa’s operational-excellence piece makes the distinction concrete: combining automation with human expertise is an end-to-end process redesign question, not simply a tool rollout. FTI adds the organisational guardrail that people, structures and processes should be designed around responsible collaboration. The practical test is to name what AI executes, what people review and who handles exceptions.
What we still don’t know
The NBER / Atlanta Fed firm survey remains useful context because it reports limited employment and productivity effects so far for most respondents. That asks a different question from a consultancy recommendation or a supplier study. Amazon and Cloudflare actions provide company context, but there is still not enough matched evidence to claim operators consistently sit apart from advisers.
Ideas to watch
Manager layers: whether agent coordination genuinely removes coordination roles or creates new supervisory work.
Outcome quality: whether faster work survives checking, exception handling and client acceptance.
Skill shifts: which capabilities become more valuable when routine production and coordination are automated.
More revealing than another adoption statistic would be evidence that organisations revised an AI work model after experience. Newly added authors do not themselves establish movement over time.
Where does your organisation sit?
Plot a particular function: is the primary emphasis strengthening human capability or substituting a defined human activity? Is AI being added to existing work or prompting a redesign of roles and workflows? Comparing that position with published thinking can reveal an assumption worth examining without prescribing the organisation’s choice.
Follow this Plot for new evidence, revised arguments and any substantiated movement. Plot your organisation to compare its assumptions with the published thinking.
Evidence note: 13 thought pieces are plotted at high editorial confidence; five additional sources are held as context because they do not support both axes from one coherent stance. The set does not establish a company-specific employment forecast.
