Doctorow: Australia Risks Losing Irreplaceable IT Skills in the AI Bubble’s Wake

In an interview published Monday by The Guardian, technology critic and author Cory Doctorow issued a stark warning to Australian businesses and policymakers: when the AI investment bubble eventually pops, organizations that replaced workers with AI will discover they’ve lost skills that cannot be bought back. The warning lands at a pivotal moment for Australia, as the federal government refines its national AI strategy and companies accelerate deployments of generative AI tools across IT, customer service, and administration.

What Doctorow Said: The AI Bubble Will Burst, and When It Does, Skills Disappear

Doctorow, whose new book The Reverse Centaur’s Guide to Life After AI released in July 2026, is not arguing that AI is useless. He makes a more nuanced case: the technology has genuine utility, but the financial hype and incentive structures around it are unsustainable. “The question isn’t ‘are we in an AI bubble?’ but ‘when will the bubble burst?’” he told The Guardian.

The bursting of a bubble doesn’t make the underlying tools vanish—just as the dot-com crash didn’t kill the internet, and the cloud survived earlier hype cycles. But it does fundamentally change which use cases prove durable. After a correction, systems that survive will be those that are measurably useful, affordable without endless venture capital, and integrated into real business processes under human governance.

The greatest danger, Doctorow warns, is not a chatbot’s hallucination or a model’s poor benchmark score. It’s the “unrecoverable workplace knowledge” that walks out the door when executives treat AI as a direct replacement for labour. Detailed knowledge of business processes—why a legacy Group Policy Object exists, which line-of-business application breaks after a certificate rollover, how a payroll integration handles exceptions—rarely sits in a document. It lives in the heads of experienced administrators, developers, and operations staff. “You can’t just recover it like the workers who have it when you scatter them to the four winds,” he said.

What This Means for Australian IT Teams — and Their Employers

For Windows administrators, support engineers, and development leads, Doctorow’s warning translates into a concrete set of risks that go beyond job security. The real threat is the erosion of the tacit knowledge that makes Microsoft 365 tenants, Azure subscriptions, Intune policies, Active Directory forests, and complex desktop estates manageable.

For IT admins and operations staff: The tools you maintain daily—PowerShell scripts, Group Policy, service account dependencies, SharePoint workflow integrations—are often under-documented. An AI copilot might generate a plausible runbook, but it cannot recall the one failed disaster-recovery exercise that taught you why an apparently redundant step is critical. If organisations hollow out their senior IT workforce, the institutional memory that prevents multi-hour outages disappears. The result is “automation debt”: systems that appear to run themselves but become impossible to repair when a vendor changes an API, a certificate expires, or a model hallucinates a configuration change.

For developers and engineering managers: “Vibe coding”—generating code with AI and pushing it to production without architecture review, test coverage, or security assessment—creates technical debt at an unimaginable scale. Doctorow calls this a future incident with an unclear author. Low-code tools like Power Automate and Teams bots make it easy for non-technical employees to build business-critical workflows that IT never sees. When those break, there’s no runbook, no ownership, and no rollback plan. The correct response isn’t a ban on experimentation; it’s a clear hierarchy of risk:

This framework aligns with Australia’s voluntary AI safety guidance, which emphasises deployer responsibility and human-centred deployment.

For business leaders and procurement teams: The temptation to replace support desks with automated agents or downsize senior engineering staff is strong when vendors promise cost savings. But after a bubble corrects, the organisation that retained human judgement—the ability to inspect, verify, and intervene when an AI system goes wrong—will be the one that survives. The alternative is a brittle dependency on a vendor’s roadmap and a workforce unable to recover lost operational knowledge.

How We Got Here: Australia’s AI Policy Meets Market Hype

Australia’s official AI stance, as outlined in the government’s response to the Senate committee on AI and the National AI Plan, is more cautious than the most aggressive replacement narratives. The Department of Industry, Science and Resources describes a strategy built on three pillars: infrastructure and domestic capability, worker support and skills, and safeguards. Government analysis cited in that response suggests AI is more likely to augment than replace most work in the near term, while acknowledging the need for active planning with employers, workers, and unions.

Yet the pressure to deploy is immense. Massive investments from sovereign wealth funds and tech giants keep the bubble inflated, and circular investment between chip manufacturers, AI companies, and enterprise customers masks the fragility. Doctorow points to the delay of IPOs from firms like Anthropic and OpenAI as one potential domino. When the bubble does burst, he argues, governments that resisted becoming overcommitted buyers at the peak will be best positioned to acquire hardware, expertise, and open-source models cheaply.

Prime Minister Anthony Albanese recently promised “plain as day” copyright protections to ensure AI companies pay for creative works, following industry lobbying for a text-and-data-mining exception. The government has since stated it is not considering such an exception, but Doctorow cautions that copyright reform alone won’t ensure creators get paid—too often, licensing payments flow to large platforms and rights holders rather than the workers themselves. A broader creative-economy response, including labour law rights over AI use and collective bargaining, is needed.

Four Actions to Protect Your Organisation from Automation Debt

1. Keep a human operating model. Do not automate away the last people who understand a core process. Maintain deliberate human capability for incident handling, quality review, and manual fallback—even if AI handles routine work. For IT teams, this means preserving administration skills around identity, endpoints, networking, backup, security operations, and application ownership. A copilot can accelerate a script; it cannot be the accountable owner of a domain controller recovery.

2. Demand exit paths from AI suppliers. Before committing to a platform, ask: Can prompts, logs, configurations, and knowledge bases be exported? What happens to custom agents if the service changes price or is discontinued? Is there a workable offline or manual process? Who owns generated outputs and fine-tuning data? These are standard enterprise architecture questions, now applied to AI.

3. Measure outcomes, not activity. Many AI projects are evaluated through superficial metrics like prompt volume or licences assigned. Better measures include error rates before and after deployment, escalations, employee workload and satisfaction, cost per completed business outcome, and the ability to continue operating if the AI service becomes unavailable. A pilot should be allowed to fail without becoming a political embarrassment.

4. Invest in skills that last across market cycles. Australia’s public strategy already recognises that AI capability is about more than buying access to offshore chatbots. The National AI Plan highlights infrastructure, domestic capability, and workforce development. Prioritise high-quality research computing, cybersecurity capacity for evaluating AI, workforce training in verification and systems thinking, procurement skills, and open standards. These assets retain value regardless of which vendor dominates next quarter.

The Bottom Line: Preserve Judgement, Not Just Tools

Doctorow’s warning is ultimately an organisational design problem. If executives use AI to shed expertise and create brittle dependencies, a market correction could expose catastrophic gaps. If they use AI to improve documentation, automate drudgery, and strengthen human decision-making, the same tools could remain valuable long after the bubble rhetoric fades.

For Windows administrator and developer communities, the lesson is clear: AI can accelerate a script, but it cannot replace the person who knows why a specific Group Policy exists, which service account must never be disabled, or how a business really gets work done when the documented process fails. That knowledge is the real infrastructure. It takes years to build and moments to lose.