Workflow Redesign Around AI

The primary organisational value of AI is not captured by automating individual tasks but by redesigning entire workflows around human-AI collaboration — a distinction that most organisations are not yet making.

What It Is

McKinsey (2025) draws a sharp distinction between task automation (replacing individual steps in an existing workflow with AI) and workflow redesign (reimagining an entire process from scratch around what humans, agents, and robots do best). Task automation can improve efficiency at the margin; workflow redesign can unlock step-change improvements in productivity and output quality.

The practical difference: task automation typically requires little organisational change and can be implemented bottom-up by individual workers or teams. Workflow redesign requires leadership commitment, cross-functional coordination, role restructuring, and changes to performance metrics, culture, and incentive systems. McKinsey’s $2.9T value estimate for 2030 is contingent on organisations performing the harder version — workflow redesign — at scale.

The UC Berkeley “Measuring Agents in Production” study (2025) offers a ground-level complement: surveying 306 practitioners across 26 domains, it finds that production AI agents are typically built with “simple, controllable approaches” — organisations prefer reliability and auditability over sophistication.

Why It Matters (for Organizations)

The workflow redesign lens reveals a key competitive dynamic: organisations that redesign workflows (rather than just automating tasks) will disproportionately capture AI’s productivity gains. Those that don’t will spend money on AI tools while achieving only marginal efficiency improvements. This creates a bifurcation between “AI leaders” — who rebuild processes around intelligent systems — and “AI laggards” — who bolt AI onto existing processes.

For leaders specifically, McKinsey argues they need to engage directly with AI rather than delegating, invest in the human skills that matter most in a redesigned workflow (framing, interpretation, oversight, quality assurance), and balance efficiency gains with responsibility, safety, and trust.

Evidence & Examples

  • McKinsey identifies “complex, high-value workflows that rely on unstructured data” as the primary opportunity for redesign — legal contracts, medical records, customer service interactions, financial analysis (agents-robots-and-us-skill-partnerships-in-the-age-of-ai.pdf)
  • UC Berkeley study found production agents primarily built for: increasing productivity, reducing human task-hours, automating routine labour, and increasing client satisfaction — in that priority order (2512.04123v1.pdf)
  • 306 practitioners across 26 domains surveyed; 20 in-depth case studies conducted; finding: simple, controllable approaches dominate production deployments, not the sophisticated multi-agent orchestration seen in research contexts (2512.04123v1.pdf)
  • Top development challenges in the Berkeley study: evaluation and measurement of agent performance, handling edge cases, maintaining reliability, and managing user trust (2512.04123v1.pdf)
  • McKinsey: the skills growing in demand as part of workflow redesign include quality assurance, process optimisation, and teaching — humans taking on the oversight and improvement roles that AI creates (agents-robots-and-us-skill-partnerships-in-the-age-of-ai.pdf)

Tensions & Open Questions

  • The gap between aspiration and practice: McKinsey describes an aspirational state of comprehensive workflow redesign; the Berkeley study describes what organisations are actually doing in production — simpler, more controlled, more cautious deployments. The path from the latter to the former is the core change management challenge.
  • What does “redesign” mean for knowledge work? Workflow redesign is straightforward to conceptualise in manufacturing or logistics (where processes are already documented). In knowledge work — consulting, legal, investment analysis — processes are often informal, tacit, and person-dependent. How do you redesign a workflow that has never been written down?
  • Middle management as bottleneck: The “conductor vs. craftsman” shift (a concept seeded in AGENTS.md) suggests middle managers will increasingly need to orchestrate AI-human workflows rather than perform technical tasks themselves. Whether middle management is up to this role — and whether organisations will invest in developing it — is a major open question.
  • 🔴 TODO: Find case study evidence of specific organisations that have successfully completed full workflow redesign (not just task automation) around AI. What did the process look like? What were the failure modes?

Skill Partnerships Human-AI · AI Agents in Production · AI Delegation and Multi-Agent Systems · AI Skill Formation and Deskilling