ChatGPT turns workers into "super individuals" — but it hasn't made companies "super organizations"

Individual gains, organizational stall

AI has clearly amplified individual capabilities. But companies are not seeing matching business results. Developer-efficiency tracker DX reportedly followed 400 firms for 16 months and found AI tool usage rose 65% while code delivery rose by less than 10%. A National Bureau of Economic Research (NBER) survey of roughly 6,000 executives found more than 80% saying AI produced no measurable productivity gains. Andreessen Horowitz (a16z) recently ran an essay by Hebbia founder George Sivulka that frames the gap bluntly: "We changed the motor but haven't redesigned the factory." Is handing every employee ChatGPT the same as rearchitecting how work gets done? Evidence says no.

Why organizations lag

There are three linked failures: coordination, signal-to-noise, and absorption. Individual employees adopt idiosyncratic prompts and formats; nobody aligns 100 separate AI-driven workflows. Generative AI makes output cheap — and multiplies polished but low-value noise. Research cited in the original report found large language models exhibit sycophancy in 58% of cases, rising when users speak in the first person. In controlled tests by AI-safety group METR, experienced developers using AI were 19% slower on tasks even though they felt 20% faster — a 39-point perception gap. Asana's Work Innovation Lab calls it "absorption capacity": output can outrun an organization's ability to review, validate and act.

What companies must do

Sivulka and other practitioners urge redesigns that treat AI as an organizational actor, not only an individual tool. That means deterministic, auditable agents for enterprise workflows, explicit coordination mechanisms, new roles (AI-agent managers, intent engineers), and governance: AI audit, AI investment committees, compliance officers. Goldman Sachs has reportedly called Cognition's Agent Devin "our new employee" — a signal that some firms are starting to treat agents as team members rather than mere utilities. The lessons are historical: installing ATMs made bank branches operate differently, but only the iPhone erased old tasks. Will AI be another ATM — or an iPhone for the enterprise?

Geopolitics and the adoption clock

A final wrinkle for global readers: access to models, chips and tooling is uneven. Export controls and broader US-China tech frictions reportedly affect how quickly Chinese firms can integrate advanced models and chips into enterprise redesigns, meaning the "factory redesign" moment may arrive unevenly across markets. For now, surveys — including a Deloitte study cited in the analysis — show only about a third of organizations pursuing deep AI-driven transformation; most have merely "switched the motor." The risk is clear: supercharged individuals without redesigned organizations can create more illusion than advantage.