The Retraining Trap

Agentic AI isn't just
eliminating jobs
it's outpacing retraining itself

By the time a displaced Australian worker completes retraining for a new occupation, agentic AI has already learned to do that job better. This isn't a future risk. It's happening now.

2.54M
Australian workers displaced by 2041
6–12
Months average retraining time
$0
Current government income bridge

What is Agentic AI — and why is it fundamentally different?

Previous automation replaced physical labour. This wave replaces cognitive work — and it does so autonomously, adaptively, and at a speed no retraining programme can match.

Definition — Agentic AI

Agentic AI refers to artificial intelligence systems that can autonomously plan, reason, and execute multi-step tasks with minimal human intervention — going beyond answering questions to actively taking actions, using tools, browsing the web, writing and running code, managing files, and coordinating complex workflows in pursuit of a goal. Rather than waiting for instruction at each step, an agentic system breaks a high-level objective into subtasks, adapts when it encounters obstacles, and loops until the work is complete.

What sets agentic AI apart from every previous technological disruption is the breadth and speed of its reach. The steam engine displaced farm labourers. Computers displaced clerical workers. Each wave targeted a relatively narrow band of tasks, and it moved slowly enough for society to adapt. Agentic AI targets the full spectrum of cognitive, coordinative, and communicative work — and it is accelerating.

⚠️

This time the goalposts are moving mid-sprint

In past technological transitions, displaced workers could retrain into stable destination occupations. With agentic AI, the destination roles are themselves being automated — often before the retraining programme even concludes. Australia's policy settings have not kept pace with this reality.

Australian Named-Firm Evidence — Q1 2026

AI-attributed redundancies are already happening at named Australian employers

AI-attributed redundancies have begun at named Australian and Australian-listed employers. In February 2026, ASX-listed software firm WiseTech announced approximately 2,000 Australian job cuts attributed to AI automation of engineering and customer-service work. The Commonwealth Bank of Australia confirmed approximately 420 Australian role reductions citing AI-driven workflow automation. Atlassian announced approximately 1,600 global roles affected as AI agents replace manual engineering and support workflows. Block — Afterpay's parent — announced 4,000 global role eliminations, with hundreds of Australian Afterpay staff affected, in a restructure that the company's CEO attributed entirely to AI. The Australian Computer Society's Information Age (March 2026) consolidates these and other announcements into a confirmed Q1 2026 Australian technology-sector total of 9,238 AI-attributed role eliminations, placing Australia second globally for tech job losses in the quarter.

Sources: Computerworld, 24 February 2026; Information Age, Australian Computer Society, 2026 (consolidated reporting and named-firm announcements).
Australian Government Data — Labour Market Preview

Commonwealth department previews gradual change in AI-exposed occupations

On 28 April 2026, the Department of Employment and Workplace Relations previewed early findings from a forthcoming descriptive-data report tracking how the Australian labour market has changed since the launch of ChatGPT in November 2022. The previewed findings: graduate employment outcomes remain positive; there is no elevated rate of occupational compositional change; but there is a slight softening in the growth of the most AI-exposed occupations, such as filing clerks and keyboard operators. The Minister characterised the data as "reflective of a particular point in time, and not predictive". This directionally supports the gradual displacement trajectory underpinning the COAD demand model.

Source: Ministers' Media Centre (DEWR), ministers.dewr.gov.au. (Note: ministerial preview only — the full report is forthcoming.)

The Retraining Treadmill

A displaced worker who retrained as a project manager six months ago made a reasonable choice. By the time they certify, agentic AI can synthesise project data, generate risk registers, assign tasks, draft stakeholder reports, and flag schedule variances — faster, cheaper, and without sick days.

📞
Telemarketer

Role eliminated by AI-powered outbound calling systems. Income lost.

📋
Enrols in Project Management

6–12 month cert. Cost, time, stress, no income support.

🤖
AI Takes That Role Too

Agentic AI now plans, tracks, and reports projects autonomously.

🔄
Retrain Again?

Into what? AI is advancing faster than any retraining cycle can follow.

This scenario is not hypothetical — it is the structural reality of agentic AI adoption. The International Monetary Fund estimates 60% of jobs in advanced economies have significant AI exposure. In Australia, modelling projects 2.54 million workers displaced by 2041. The compounding problem is that displacement is happening simultaneously across sectors — meaning workers are all retraining toward a shrinking pool of destination roles at the same time. Retraining alone cannot be the policy answer.

The occupations facing displacement today

These are not low-skill roles. Many require years of training, professional certification, or specialist expertise. Agentic AI is advancing on all of them simultaneously — this is what makes the current wave unlike any that preceded it.

📞
Telemarketers & Call Centre
Critical exposure

Outbound AI agents already outperform human callers in conversion and consistency.

📊
Data Analysts & Clerks
Critical exposure

Agentic AI queries, analyses, and presents data end-to-end without a human intermediary.

⚖️
Paralegals & Legal Assistants
Critical exposure

Document review, contract summarisation, and legal research increasingly automated.

📝
Junior Accountants
Critical exposure

Reconciliation, data entry, and financial reporting are routine AI tasks.

📋
Project Coordinators
High exposure

Scheduling, reporting, and stakeholder communication increasingly handled by AI agents.

🧾
HR & Recruitment Officers
High exposure

CV screening, interview scheduling, and onboarding documentation are automatable today.

🏥
Medical Coders & Administrators
High exposure

Clinical coding, patient scheduling, and insurance processing increasingly AI-handled.

📰
Copywriters & Content Producers
High exposure

Routine content, product descriptions, and report drafting now AI-generated at scale.

What the numbers tell us

Australia's AI displacement is not a projection based on speculation. It is a convergence of independent empirical datasets that all point in the same direction.

2.54M
Australian workers projected to be AI-displaced by 2041
COAD Assumptions Log INI-004 v5 | OSCA 2024
60%
of jobs in advanced economies with significant AI exposure
IMF World Economic Outlook 2024
55K
US jobs eliminated by AI in early 2025 alone
Challenger, Gray & Christmas (2025)
$0
Dedicated federal income support for AI-displaced workers in Australia today
Budget Papers 2025–26
"The pace of machine development may permanently outpace the speed of human adaptation — compressing the adjustment window that workers and institutions would normally rely on."
— COAD Policy Rationale, 2026

Australia's Future Fund — with a corpus of AUD 269.1 billion (31 March 2026 portfolio update) — is already earning approximately $19 billion annually in investment returns. An AI Headcount Tax on firms substituting workers with AI agents could generate a further $35 billion by 2041. Combined with bond financing within Australia's exceptional fiscal headroom (22.5% debt-to-GDP against an OECD average of 68%), a fully funded income bridge for every displaced Australian worker is not just possible — it is eminently affordable.

Context matters: The experts cited above don't all have the same incentives. Some profit from alarm; others from reassurance. We've mapped the commercial interests behind every major AI forecast.

Read the bias analysis →

Why Australia needs COAD — Citizen Offset for AI Displacement

COAD is not a universal basic income. It is a targeted, graduated income bridge for workers whose livelihoods are displaced by AI automation — funded by the industries and systems that are doing the displacing.

🛡️

A financial floor when retraining fails to keep pace

COAD provides a guaranteed income bridge so displaced workers are not left destitute while the job market transforms around them. No worker should face homelessness because an AI took their job before they could adapt.

🚀

Freedom to retrain without existential pressure

With a stable income floor, displaced workers can make thoughtful retraining decisions — not desperate ones. They can pursue education, caring roles, or community work without the financial terror of immediate destitution.

💼

A built-in work incentive

COAD is set at 70% of minimum wage (~$50,284), creating a genuine 43% earnings premium for employment. Evidence from the Stockton SEED trial found recipients were 28% more likely to find full-time work — not less.

💰

Funded by those who benefit from automation

COAD's three-pillar model draws on Future Fund returns (~$19B/year), an AI Headcount Tax on firms substituting workers with AI (~$35B/year), and government bonds within Australia's AAA fiscal headroom (~$36B/year).

Year Annual COAD Payment PPP Equivalent (2041 dollars) Eligible Population Total Programme Cost
2027 $20,000 AUD 0.96 million workers ~$19.2B
2030 $25,000 AUD Rising
2035 $30,000 AUD Rising
2041 $35,000 AUD $52,000–$58,000 PPP 2.54 million workers ~$76B (vs $90B capacity)

Source: COAD Assumptions Log INI-004 v5 (February 2026). Three-pillar funding capacity of $90B exceeds projected peak demand of $76B by an 18% buffer. All figures audited to COAD-PRIME Bulletproof standard.

What happens if Australia does nothing?

The question is not whether AI will displace Australian workers — the IMF, the World Economic Forum, and Australia's own Treasury modelling agree that it already is. The question is whether Australia will be the country that designed a response, or the country that improvised one after the damage was done.

27
2027

960,000 displaced workers, no income bridge

Without COAD, nearly a million Australians face unemployment from AI displacement with no dedicated support mechanism — relying on a JobSeeker system not designed for structural technological transition.

30
2030

The retraining treadmill accelerates

Workers who retrained in 2027 for roles that are now also AI-exposed face a second displacement wave. Political pressure mounts. Ad hoc emergency measures are introduced at greater cost and less coherence than a designed system.

41
2041

2.54 million displaced — without the $14B buffer COAD preserves

At peak displacement, 2.54 million Australians require income support. A reactive system costs more, delivers less, and leaves communities damaged in the interval. COAD's designed surplus buffer of ~$14 billion disappears in a fragmented response.

Every Australian should understand what's coming

Share this page with your MP, your workplace, your union, your community. The window to design a proper response to AI displacement is narrowing — and public understanding is how policy gets made.