The AI Automation Cliff Is Already Here

David Shapiro’s revised thesis on the automation cliff rings true: it’s not a single, economy-wide collapse, but a cascade of individual drops, one value stream at a time. He initially projected a broad inflection point by 2030, but now clarifies that displacement accelerates sharply, not across the whole economy, but within specific business functions. The cliff is real; its arrival is simply more granular.

The distinction between a workflow and a value stream is critical. A workflow represents the employee-facing process—the queue of customer tickets, for instance. A value stream, however, is the underlying business outcome: ensuring every customer receives an answer. Automation targets this deeper, outcome-driven layer.

This digital precision explains the sudden, massive shifts we observe. Shapiro, an former infrastructure automation engineer, points out that tools like Terraform allow engineers to scale from 10 to 10,000 servers by changing a single number in one file. Amazon, for example, deployed new code every 11.6 seconds in 2011, updating 30,000 servers simultaneously at peak.

This principle of arbitrary scalability means that once a digital automation process works reliably, its deployment can switch from testing to enormous scale almost instantly. The cliff isn't a gradual slope; it’s a sudden drop for each perfected value stream.

Automation compresses labor timelines with a suddenness that makes traditional project management obsolete. Consider the UK Department for Work and Pensions: 12 software robots cleared a backlog of more than 30,000 pension claims in just two weeks. This isn't gradual improvement; it's an instantaneous deletion of an entire queue.

Airbnb experienced similar acceleration during an AI-assisted test-file migration. Engineers estimated the work would take 18 months manually, yet AI completed 75% of the task in four hours. This rapid compression, supported by human validation and engineering oversight, highlights AI's capacity to collapse timelines from years to hours.

Klarna’s AI assistant further exemplifies this "overnight" shift. In its first month, it handled 2.3 million customer conversations. Klarna attributes this to the equivalent of 700 full-time agents, with average response times dropping from 11 minutes to under two. This isn't merely efficiency; it's a fundamental redefinition of customer service capacity.

The common thread is not incremental gain but discontinuous transformation. When a workflow or value stream is fully automated, the demand for human labor in that specific area doesn't taper—it vanishes. This isn't a slow decline; it's a cliff event, repeated across industries, one process at a time.

Proof of the cliff is real—and messier than the pitch. Labor signals are increasingly uneven, not uniformly bleak. Call-center and customer-service employment has fallen, alongside fewer freelance writing and coding posts, and employers increasingly cite AI in layoff announcements.

Crucially, aggregate employment trends do not establish AI as the sole cause. Pandemic-era overhiring, interest rates, and strikes also shaped recent cuts. Even “AI-washing”—blaming AI for pre-planned layoffs—can skew the data, as Sam Altman himself observed.

Automation’s reliability also presents a significant hurdle. Australia’s Commonwealth Bank, for example, cut 45 call center jobs after implementing a voice bot, only to reverse the decision weeks later, admitting the jobs were not redundant after all. Similarly, Klarna, initially lauded for its AI assistant, soon renewed its focus on human support, with its CEO describing it as "VIP treatment" after quality suffered.

These reversals highlight the hard work of achieving successive reliability “nines.” As OpenAI co-founder Andrej Karpathy notes, a demo working 90% of the time is "just the first nine." Reaching 99% or 99.9%—the kind of perfect automation seen in tools like Terraform (further reading on infrastructure as code: Introduction to Terraform - HashiCorp Developer)—requires significant effort. Some value streams will hit their cliff in months; others will take years.

National statistics, aggregated and smoothed, can easily obscure the sharp, localized impacts of automation. Each value-stream cliff arrives on its own timeline; displacement often manifests as quiet attrition or delayed hiring, not public layoffs. This hidden shift masks the true pace of change.

The Federal Reserve recently reported that firms in five of its twelve districts cut staff through attrition or delayed hiring after AI investments. Simultaneously, a Stanford study revealed an 11% employment drop for young workers (ages 22-25) in highly AI-exposed occupations since late 2022, while their peers in less exposed roles saw 10% growth. Correlation is not causation, but the signal is clear.

This brings us to David Shapiro’s post-labor economics: if machines perform more work, wages tied exclusively to human work will shrink. The critical questions then pivot to ownership and broader wealth distribution. Shapiro forecasts this fundamental economic re-evaluation within the next five years, a timeline that demands our urgent attention.

What does David Shapiro mean by an automation cliff?

A task or value stream can remain human-led until automation becomes reliable, then shift rapidly to software at scale.

Why does Shapiro say there are thousands of automation cliffs?

Companies automate distinct workflows and business outcomes at different speeds, rather than replacing all work at once.

Does AI automation always eliminate jobs?

No. It can reduce hiring or workload without layoffs, and some companies have reversed cuts when automated service quality fell short.

What is a value stream?

A value stream is the end-to-end work that produces an outcome customers or the business value, beyond an employee’s individual workflow.