The most-repeated question about AI and jobs — 'will the role still exist?' — is the wrong one. The right one: what does the role become when the routine 40% is gone, and is your organisation designed for that shape?
How should roles be re-designed when AI absorbs 30–60% of routine work?
The reversal
The public debate frames AI's workforce impact as displacement. The operational reality inside live deployments is reshaping: routine work compresses, judgement work expands, coordination work re-emerges, and specialised human review becomes the highest-value part of the workflow. The World Economic Forum's 2024 Future of Jobs report puts the reshaping at roughly 50% of roles by 2027 — a scope that outruns most current role designs.
The insight stack
What actually moves the P&L
Judgement rises to the top of the role
When routine execution is absorbed by agents, the human role compresses toward judgement calls the agent cannot make: ambiguous exceptions, ethical or reputational trade-offs, cross-functional coordination. Role designs still built around execution volume misprice this shift; role designs built around judgement density capture it.
Coordination re-emerges as a first-class skill
Agents multiply throughput but not coherence. The coordination work required to keep multiple agents, human specialists, and external stakeholders aligned becomes a distinct skill set — increasingly the highest-leverage skill in an AI-enabled operating team. Job descriptions that omit coordination systematically underprice the role.
Human-in-the-loop is a job, not a control
Compliance frameworks treat human-in-the-loop as a checkbox. Operating reality treats it as a role: someone reviewing agent outputs, correcting drift, generating labelled data, and escalating novel patterns. Under the EU AI Act, high-risk systems require documented, competent human oversight — which is a role definition problem, not just a governance one.
Case example
A £24M contact-centre operator
A contact-centre operator deployed a triage agent that handled 62% of inbound tickets within a quarter. Rather than reduce headcount, the operator redesigned agent-facing roles into three tiers: routine triage was fully automated, tier-1 humans became exception handlers and coordination specialists with expanded authority, and a new tier-2 'model steward' role emerged — accountable for auditing agent outputs and feeding corrections into retraining. Twelve months later, cost-per-contact was down 34%, first-contact resolution had risen from 71% to 88%, and voluntary attrition in the redesigned roles had fallen by 40% — the routine work had been the driver of exit, not the judgement work.
Mini-playbook
Role redesign, five moves
Split every role into routine, judgement, and coordination components — measure the proportions.
Automate the routine share deliberately, sequenced with capability transitions for the humans doing that work today.
Rewrite job descriptions around judgement density and coordination scope, not execution volume.
Create explicit 'model steward' roles where agents operate at scale.
Move compensation design toward outcomes rather than throughput — throughput is now an agent metric.
How Strategy Labs installs this
Anchored to Process re-engineering
CAE runs role redesign against the process re-engineering artefact, so the redesign is anchored to the actual workflow — not to job-family taxonomies inherited from the pre-AI operating model.
PDC hosts the role-composition analysis, transition roadmap, and model-steward job architecture, so redesign accumulates into a coherent workforce strategy rather than a series of ad-hoc job rewrites.
Frequently asked
Related questions executives ask
- How much of a typical role is genuinely routine?
- Depends on function. Cross-industry future-of-work benchmarks2 put the automatable proportion of most knowledge-work roles at 30–60%, concentrated in structured, repetitive, and information-retrieval tasks. Judgement and coordination components are considerably lower.
- When should we create model-steward roles?
- As soon as any agent operates at material scale — typically once it is handling more than 20% of a workflow, or once its outputs touch regulated or reputationally sensitive decisions. Retro-fitting stewardship after drift emerges is more expensive than designing it in.
- Does this reduce total headcount?
- In some functions yes, in others no. The consistent pattern is composition change: fewer routine executors, more judgement specialists, more coordinators, and new model-steward roles. Net headcount is usually less dramatic than gross composition change.
Over to you
If 40% of your team's routine work were absorbed by agents inside 18 months, is your current role architecture designed to catch the reshape — or will it force redundancies you could have avoided?
Discussion
(…)Comments are moderated before appearing. Your email is only used for moderation and is never shown publicly.
Loading discussion…