Possible future 1 of 6 · W01
AI helps; humans stay in charge
AI becomes a normal workmate, but people still make important decisions and remain responsible.
- Forecast target
- 31 Jul 2031
- Evidence cut-off
- 2 Aug 2026
- Edition
- 2031-2026-08-02
- Status
- Working forecast
This is a scenario, not a prediction or probability. It is a deliberately testable possible condition used to see how product demand changes.
How this future develops
Now
Where we are
AI can already draft, analyse and organise work, but it still makes mistakes and often needs checking.
World shift
What would change
AI improves and spreads, but not enough for organisations to trust it alone with important decisions.
2031
What life could feel like
Most people still have jobs. AI prepares and checks the work; a named person decides, explains and fixes what goes wrong.
Why the apps would change
The most useful apps help people work faster while making responsibility, approval and recovery clear.
The condition the research tested
If cross-domain reliability remains brittle or costly to assure through 2030, and institutions keep a named human responsible for consequential action, machines mainly prepare, check and coordinate work. Paid employment remains the main income, status and time allocator in July 2031, while physical, climate, care and education constraints keep gains local and uneven.
Causal chain
- Reliability and recovery remain costly.
- Audit, liability and security keep authority scoped.
- Organisations redesign tasks around supervised assistance.
- Wages and payroll-linked protection remain central.
- Care, teaching, maintenance, climate adaptation and security retain human duty.
- Work structures time and status, while contribution and belonging also depend on family, community and civic institutions.
- Relational AI sometimes bridges to people and sometimes substitutes for reciprocal contact.
- Physical and fiscal limits keep outcomes regionally uneven.
What we had to assume
- Independent audit and liability remain influential.
- Employers and educators preserve supervised expertise pipelines in some capable systems.
- No single compound shock destroys managed interdependence everywhere.
What nobody knows yet
- Whether constrained delegation reflects capability or institutional choice.
- Whether complementarity improves bargaining or intensifies surveillance.
- Whether gains finance care, education and climate adaptation.
- Whether relational systems bridge to people or displace reciprocity for particular groups.
Evidence used to frame this world
observation · W01-E01
At the cutoff, agent progress coexists with short dependable horizons, jagged transfer and self-validation gaps.
Source records: AC-06, AC-09, src-international-ai-safety-report-2026, src-metr-long-tasks-2025
Important limit: Evaluations overrepresent digital benchmark environments and do not prove failure everywhere.
observation · W01-E02
Exposure measures do not establish realised displacement, and bounded studies show mixed productivity.
Source records: EWF-04, EWF-07, SC-WORK-01
Important limit: Occupation exposure and bounded studies do not settle economy-wide employment.
projection · W01-E03
Growing compute demand meets grid, transformer, water and local-permission constraints.
Source records: PC-S01, PC-S02, E05
Important limit: Projected totals are not commissioned local capacity.
policy-intent · W01-E04
AI rules, public compute, workforce programmes and assurance plans point toward supervised adoption.
Source records: AC-25, AC-30, AC-33, AC-36
Important limit: Plans and laws do not prove delivery, enforcement or equity.
inference · W01-E05
Costly review, rescue, security and liability preserve accountable human roles and supervised entry paths.
Source records: EWF-07, src-ilo-exposure-indicators-2026, src-who-multimodal-health-guidance-2024
Important limit: This causal link is inferred, not an observed 2031 result.
scenario-condition · W01-E06
Independent tests do not jointly cross long-task, low-rescue, matched-cost and broad-diffusion gates by 2030.
Source records: src-oecd-ai-trajectories-2026, src-metr-long-tasks-2025
Important limit: The conjunction is a scenario classifier, not a forecast.
unknown · W01-E07
Net wages, work intensity, entry careers, informal work, unpaid care and relational-system effects remain unsettled.
Source records: EWF-04, EWF-05, SC-AI02
Important limit: No global causal series joins these outcomes.
Forecast apps most affected by this future
RemedySwitch
Challenge a machine-made decision and reach a qualified human who can stop it, explain it and put the result right.
CaseCairn
Stop being bounced between services: one qualified human carries your case until the next service genuinely acts.
DutyPool
Call an independent qualified duty officer to pause a high-stakes machine action and organise a fair repair.
HumanHarbour
Step back from an embedded synthetic relationship without losing your records, support or route to a real person.
ShiftWitness
See whether a machine can finish the whole shift, including every rescue, delay and repair.
ClaimTide
Make old proof expire the moment a machine, workplace or rescue plan changes in a meaningful way.
HumanLift Ledger
Count the human effort hidden behind an automated claim and give workers power to challenge it.
CareKeel
No silent cancellation: keep the agreed care minimum moving through an outage.
LoadCovenant
Approve a huge new load without quietly making homes and public services pay the price.
RecoveryLoom
Practise the failure nobody expects—before your team has to recover a real hospital, grid or delivery system.
Mendstep
Fix real homes and local services, get paid while learning, and climb into skilled work one safe repair at a time.
AppStore2031