The Two Faces of AI-Assisted Coding

In a guest column for CNET, author and activist Cory Doctorow unpicks a curious divide among programmers. Some experienced coders rave about the AI code they now ship, while others across the same industry are terrified by the error-riddled, debt-generating output they have to push through. Both groups are reliable narrators of their own experience, so why the contradiction?

Doctorow traces the answer to the history of labor and automation. He introduces the idea of the “centaur” – a human mind guiding a powerful machine – and its dark twin, the “reverse centaur.” In the first scenario, the worker chooses when and how to use AI, retaining judgment and control. In the second, most colleagues have been fired, the remaining staff must frantically evaluate AI-generated work at a pace set by the machine, and they live in fear of losing their job.

The “humans in the loop” in this reverse-centaur model become what Doctorow calls “moral crumple-zones,” absorbing blame when the AI’s unchecked code causes failures. He argues this dynamic is not an accident but a direct consequence of the AI industry’s fundamentals: AI firms are losing money on every customer, and the only route to profitability is to replace workers and make the survivors serve the chatbot.

The Reverse Centaur: Why Capital-Driven AI Deployments Backfire

The Centaur vs. Reverse Centaur Dichotomy

Doctorow borrows from labor-automation theory to frame the issue. A centaur uses the machine as a tool, like a spell-checker or a bicycle, enhancing the human’s capability. Conversely, a reverse centaur means the human becomes a peripheral to the machine, performing the tasks AI cannot handle itself. The first group of coders Doctorow describes are centaurs – they decide how and when to accept AI suggestions. The second, fearful group are reverse centaurs, condemned to mark AI homework at superhuman speed for superhuman stretches.

AI Industry Economics Favor the Reverse Centaur Model

The columnist argues that the AI industry’s broken unit economics push it toward worker replacement. AI firms are “history’s most efficient money incinerators,” selling services below cost to win customers. When they attempted to raise prices, demand collapsed. The only path to revenues that could cover massive losses, he says, is to fire human workers and share the saved wage bill with the former employer. That logic inevitably produces reverse centaurs, because deploying AI to assist existing workers would not generate the trillions in profit that investors expect.

The Hidden Costs: Tech Debt and Liability

Doctorow warns that setting humans to work at the limit of their speed and endurance guarantees mistakes. In safety-critical fields like avionics, this becomes a serious risk. The system, he says, is designed so that when bad AI-generated code slips through, the human “marker” is the one held accountable. The companies that cut headcount in favor of AI may see short-term cost savings, but they accumulate technical debt and reputational exposure as the quality of output deteriorates.

For Managers and Workers: Navigating the AI Automation Trap

  • Assess your deployment posture. Companies using AI for code generation should examine whether workers control the tool or are being forced to check its output at machine speed. The latter model risks accumulating technical debt that will undermine product quality over time.
  • Manage safety-critical risk. Managers in sectors like avionics – where one coder in Doctorow’s column warned, “just don’t ever fly on an airplane again” – must recognise that reverse-centaur setups can bypass human judgment, elevating the chance of catastrophic failure.
  • Document oversight responsibilities. Workers assigned to mark AI output should formally document review processes and decision boundaries. This can help establish where liability truly lies and challenge the “moral crumple-zone” dynamic Doctorow describes.
  • Investor pressure points. Backers of AI firms should understand that the “fire and replace” playbook is a symptom of unsustainable unit economics, not a sign of durable competitive advantage. The pricing struggles cited in the column suggest a business model still searching for viability.

Risk & Opportunity Assessment

Commercial RiskMediumCompanies adopting reverse-centaur AI risk shipping more bugs and incurring rework costs, as described by coders who fear the tech debt they generate.
Competitive RiskMediumOrganizations that harness AI as a centaur tool – under worker control – may maintain higher code quality and outpace rivals mired in AI-generated defects.
Regulatory RiskLowNo specific regulations are mentioned, but safety-critical software failures could prompt legislative attention if reverse-centaur practices become widespread.
Reputation RiskHighPublic knowledge that a company relied on AI code marked by exhausted workers – with a ‘moral crumple-zone’ designed to blame the human – would severely damage trust.
Technology DisruptionTransformationalAI is already reshaping how code is written; the column argues its deployment model determines whether it becomes a productivity multiplier or a source of systemic harm.
Commercial OpportunityHighFirms that successfully implement the centaur model – letting skilled workers guide AI – could capture significant productivity gains without the collateral damage of mass layoffs.