Amazon Employees Say AI Is Just Increasing Workload. A New Study Confirms Their Suspicions

ActivTrak's analysis of 163,638 employees across 1,111 organizations shows AI adoption increased measured work — emails +104%, chat/messaging +145%, and time in business management tools +94% — concluding that 'AI does not reduce workloads.' Amazon corporate employees told The Guardian internal AI tools are often 'half-baked,' produce errors requiring verification and corrections, and have added to task time rather than delivering promised efficiency. Overall, AI may speed components of work but is being used to layer on more output, shifting benefits to employers rather than reducing employee time burden.

Analysis

Adoption of generative AI inside large orgs is creating a predictable two-stage cycle: an initial “coordination tax” as outputs multiply and teams build verification, audit and loopback processes, followed by a consolidation phase where specialized tooling (prompt libraries, human-in-the-loop QA, observability) reclaims time. I estimate the coordination tax will add the equivalent of ~10–25% more FTE effort to knowledge-work processes in the first 6–12 months at scale, then decline as workflows are re-engineered over 12–24 months.

Second-order winners are infrastructure and MLOps providers due to higher baseline compute, storage and telemetry consumption — think recurring revenue uplift of mid-single-digit percent for cloud/MLOps peers during the integration window — while companies with tight, ops-levered margins and rapid output targets risk margin compression if they cannot operationalize quality control. Large tech incumbents that productize enterprise-grade guardrails and integrate billing/SLAs into their cloud stacks are positioned to monetize the messy middle faster.

Key catalysts: earnings commentary on enterprise GenAI ARR, new MLOps feature launches, and labor/IR developments (policy or union actions) will move the tape in the near term (days–quarters). A true regime change that reduces the coordination tax materially will require model-level error rates and tool ergonomics improving such that verification time drops >50%, which is a 12–24 month conditional outcome; conversely, regulation or high-profile safety failures are tail risks that could slow adoption for years.

Key Decisions for Investors

• Pair trade (6–12 months): Short AMZN equity or buy a cheap 3–6 month put spread (caps losses) and go long GOOGL (stock or 12-month calls). Rationale: AMZN faces larger near-term ops/friction risk; GOOGL can monetize enterprise AI faster. Risk/reward: target 20% asymmetric return on the pair if AMZN underperforms by ~10–15% vs GOOGL; stop the pair if the spread moves against you by 6–8%.
• Event-driven short (2–8 weeks before earnings): Buy AMZN 1–3 month put spreads sized to limit premium exposure; catalyst: any bullish AWS/AI commentary that fails to translate to improved guidance should re-rate shares. Reward: 2–4x potential return vs premium; risk: rapid upside on positive cloud prints—use defined-width spreads.
• Long GOOGL core exposure (9–18 months): Accumulate GOOGL stock or buy 9–12 month call verticals ahead of enterprise product rollouts and incremental ARR disclosures. Objective: capture secular ARPU lift from enterprise GenAI; downside risk limited by long-term secular growth, upside 15–30% if monetization accelerates.
• Tactical long on MLOps/observability names (3–9 months): Allocate a smaller satellite position to firms that capture verification/telemetry spend (example: observability vendors), expecting 5–10% bump in spend per large enterprise during integration. High beta trade — size modestly and take profits when vendor guidance cites material client upsell.