Parallels·AI Pilot Failures Mirror 1980s IT Productivity Paradox

Most AI pilots succeed in isolation but cannot scale to production. In 2025, companies scrapped 46% of AI proofs-of-concept before production due to unclear ownership and unstructured data. HCLTech 2026 survey: 43% of major AI initiatives expected to fail, 76% report Responsible AI concerns delayed deployments. Gap between experimentation and business impact widening. See: /friction/enterprise-ai-pilots-stall-before-production

1987: Nobel laureate Robert Solow observed 'you can see the computer age everywhere but in the productivity statistics.' Massive 1970s-80s IT investment with stagnant productivity growth. MIT's Brynjolfsson (1993) documented computing power increased 100x since 1970 yet productivity stagnated. Four explanations: measurement errors, adoption lags, redistribution effects, mismanagement. Paradox persisted until late 1990s as organizations restructured around IT.

Organizations invest in new technology without restructuring operating models, data governance, and organizational design to absorb it. Executives focus on acquiring tools rather than redesigning processes that will use them. Load-bearing constraint: organizational inertia—existing systems, data silos, and decision rights optimized for old technology. Without parallel change management investment, technology cannot deliver productivity gains.

Failure velocity. AI pilots fail in months rather than years because AI requires continuous data pipelines and model maintenance, whereas traditional IT deployed as one-time software installations. Faster feedback loop enables quicker problem identification but makes productivity gap visible sooner. Watch whether companies accelerate reorganization efforts or abandon AI initiatives after first wave of pilot failures.