14th April 2026
Machines Can Do the Work. Only Humans Can Do This.
The goal of AI-led work redesign is not to move human work to machines. It is to ensure humans do human work. That distinction sounds simple. Most organisations are failing to make it.
HR has a specific and urgent responsibility here: to codify where human judgement leads, to invest in the capabilities that make that judgement reliable, and to protect job quality and inclusiveness as AI adoption scales. Without this, the gains from AI come at a hidden cost to fairness, trust, and long-term performance.
This builds on an earlier argument: that HR’s new job is to redesign work, not manage people; and that the gains come from end-to-end workflow redesign, not tool rollouts.
Where human judgement must lead in AI-augmented work
High-stakes, ambiguous, and ethical decisions. Novel contexts where data is sparse or biased. Culture, belonging, and employee well-being. Machines are genuinely weak in these zones, the research on this is clear and consistent. The answer is not to hope that good judgement emerges naturally as AI scales. It is to codify decision rights and build human judgement as an explicit organisational capability.
If you’re still working out where to begin, the previous piece on why AI pilots stall covers workflow selection and decomposition in detail. Here, we focus on what HR needs to institutionalise once that work is underway.
Four things HR should institutionalise across the enterprise
The framework breaks into four practical pillars:
1. A human-work canon
A short, published list of decision domains that are human-first with machine advice — sensitive talent calls, customer remediation, safety-critical trade-offs, ethics-laden choices in pricing or claims.
2. A judgement curriculum
Explicit capability building in ethics, bias detection, uncertainty handling, and sense-making for leaders and frontline teams alike.
3. Skills transformation at scale
Material reskilling and redeployment, particularly for entry-level roles, with internal mobility and skill adjacency treated as first resorts, not afterthoughts.
4. Outcome-based metrics
Evaluate redesigns on cost per outcome, exposure reduction, auditability, fairness, accessibility, and time-to-productivity, not volume and SLA counts.
Making AI adoption employee-centric
On adoption, be radically employee-centric. Adjust performance expectations during rollout. Create structured practice time. Have leaders model AI use visibly. This is both an enabler of adoption and an opportunity to make work more sustainable.
The stakeholder question every redesign must answer
Before finalising any redesign, ask: which groups: customers, regulators, communities, or employees would be most affected if human judgement were diluted in key decisions? And how will you demonstrate that your redesigned workflows improve fairness, safety, and accessibility while lowering cost and risk?
The organisations that will get AI adoption right
They are not the fastest movers. They are the most deliberate ones. They know where humans must lead in AI-augmented workflows, they build that capability, and they can evidence it to stakeholders.
If you want to develop a human-work canon and judgement curriculum for your leadership team, let’s talk.
Justin Miles
Categories: Designing Organisations
