A study conducted by Google in collaboration with Stanford University researchers, observing how Google employees learn and apply artificial intelligence tools over 18 months, has identified a widespread pattern the researchers call "simple substitution" as the primary obstacle to meaningful productivity gains.
Simple substitution describes the behaviour of replacing individual tasks with AI equivalents without changing the underlying workflow, the equivalent of using a faster typewriter rather than reconsidering how documents are produced and distributed in the first place.
The most effective users in the study took a different approach, applying what the researchers describe as a product manager mindset: systematically identifying where their work is most constrained, selecting tools appropriate to specific problems rather than defaulting to general-purpose chatbots, and running small experiments before committing to broader changes.
The study sets out five practical strategies that distinguish high performers from the majority.
- To begin with whatever is blocking progress rather than with the most accessible tool.
- To select from the full range of available AI products rather than defaulting to chat interfaces.
- To prototype quickly at small scale before committing to a broader rollout.
- To think across entire workflows rather than optimising isolated tasks.
- To document successful approaches as reusable templates so colleagues can benefit without repeating the same trial-and-error process.
Related reading
- From Vera Rubins, Blackwell, Nemotron, NemoClaw, to Isaac and Cosmos - here's a whistlestop guide to NVIDIA's…
- Google expands open shopping protocol to let AI agents browse, compare and check out like human customers
- Meta rolls out AI support assistant on Facebook and Instagram as enforcement figures show dramatic gains
The research also explicitly pushes back against the current industry emphasis on prompt engineering, the practice of carefully crafting instructions to extract better outputs from AI models, arguing that workflow redesign produces larger and more durable gains than optimising individual interactions.
The findings carry particular weight given the research context: Google has both a commercial interest in AI adoption and direct observational access to a large, technically sophisticated workforce trying to integrate these tools into demanding knowledge work.
The recap
- Google and Stanford study identifies five AI adoption strategies.
- Study observed Googlers learning and using AI over 18 months.
- Teams should document wins and create repeatable AI templates.
