Google: Five strategies for deeper AI adoption at work

In March 2026, Google published findings from research conducted in collaboration with Stanford University on why some employees become skilled AI adopters while others plateau at surface-level use. The author, Martin Gonzalez from Google DeepMind, distils observations from real adoption patterns at Google into five strategies.

What the study found

The central finding is that successful AI adopters do not start with a tool — they start with a problem in their workflow. Effective adopters identify which parts of their work are slow, repetitive, or error-prone, then look for AI approaches that address those friction points specifically. The study observed that adopters who started from the tool side tended to apply AI in isolated, low-impact ways, while those who started from workflow obstacles found uses that changed how their entire process worked.

This framing maps directly onto how product managers approach problems: define the gap first, then evaluate solutions. The research confirms that this sequence applies equally well to how practitioners should adopt AI tools as to how product teams should build AI features.

The five strategies

The first strategy is to identify obstacles before selecting a tool — treating AI adoption as a product discovery problem rather than a software installation. The second is to move beyond general-purpose chatbots toward specialised tools suited to specific job tasks. The third is incremental testing: starting with small prototypes rather than wholesale workflow changes. The fourth is systematic integration, embedding AI across a broader process rather than using it for individual one-off tasks. The fifth is documentation and sharing — capturing what worked so that colleagues can skip the trial-and-error phase.

Why this matters for product managers

For PMs, the research operates at two levels. As practitioners, it offers a structured approach to building AI into their own workflow without the common pitfall of adopting tools for their own sake. As product builders, it provides a model for how users actually adopt AI features — they start from their existing pain points rather than from the capabilities on offer. That insight has direct implications for how AI features should be framed, introduced, and measured in products.

The article is short and practically written, based on observed behaviour rather than a prescriptive framework, which makes it easier to apply selectively to specific team or workflow contexts.