AI and the Future of Work: What the Automation Timeline Really Says

The number that stopped me when I read the McKinsey research was not the headline figure. It was the midpoint. McKinsey's analysis of automation timelines, from their work on the economic potential of generative AI, places the midpoint for half of work activities becoming automatable at 2045. Not 2027. Not 2030. 2045. The range they give is 2030 to 2060, depending on how technology develops and how fast organizations and economies adapt. That is a twenty-year window centered nineteen years from now.

That timeline sits in strange contrast to the urgency in most enterprise AI conversations right now. The 2025 McKinsey State of AI survey found that 88% of organizations report using AI in some form. The same survey found that 79% are using generative AI, with 23% actively scaling AI agents and another 39% still experimenting. Only 7% have fully scaled AI. So the organizations that are furthest along, the 7%, are still at the beginning of a multi-decade process. The organizations that are in the "experimenting" category are considerably earlier.

I want to sit with what "60% to 70% of work activities could theoretically be automated" actually means, because the phrase gets picked up in ways that distort its meaning. The McKinsey claim is about activities, not jobs. A job is a collection of tasks. Some of those tasks are highly automatable. Others are not. A hospital administrator's job involves scheduling, insurance coding, document management, and communication tasks that are all potentially automatable at varying degrees. It also involves handling patient distress, navigating ambiguous insurance disputes, and making judgment calls about how to handle situations that do not fit the standard process. The automatable activities and the non-automatable activities coexist in the same job. Automating the first category does not eliminate the job. It changes the job.

This distinction matters because the popular discourse about AI and work often collapses the two. "AI will automate 60 to 70 percent of work activities" becomes "AI will eliminate 60 to 70 percent of jobs," which is not what the research says. Jobs do get eliminated when the automatable activities constitute most of the job, or when the residual non-automatable activities are not valuable enough to sustain a full role. But for many jobs, AI automation of the routine activities creates capacity for more of the high-judgment work, and the net employment effect depends on whether demand for the output of the job grows enough to absorb that capacity. Whether that demand growth happens is an economic and organizational question, not a technology question.

The labor productivity estimate is also worth taking seriously in its actual magnitude. McKinsey projects that AI-driven automation could contribute 0.1% to 0.6% annual labor productivity growth through 2040. That range sounds modest. By historical standards, productivity growth of that scale would be meaningful but not unprecedented. The Bureau of Labor Statistics has tracked periods of higher productivity growth driven by previous technology waves. The number does not point toward overnight transformation of the economy. It points toward a gradual shift over decades, which is consistent with the 2045 midpoint for automation adoption.

The reason the timeline is later than AI hype suggests is worth understanding, because it is not primarily about the technology. The technology for automating many work activities exists today or will exist soon. The bottleneck is the organizational, regulatory, economic, and social process of actually deploying that technology at scale into working organizations. Historical technology adoption is slower than the technology's theoretical capability. Electricity was widely available by the 1880s. Its full effects on labor productivity did not fully materialize until the 1920s, because redesigning factories and workflows to take advantage of electricity, rather than just substituting it for steam power, took decades. AI may compress that timeline. But it will not eliminate the organizational and social adaptation process that determines how technology actually changes work.

This is where I think the IS research opportunity is specific and underexplored. The macro-level automation debate asks: how many jobs will AI eliminate? That question is hard to answer and generates more heat than light. The question I find more tractable, and more relevant to what Information Systems research does well, is: in specific organizational contexts, which task categories are actually being automated, how is that changing job content for the people in those roles, and what determines whether the outcome is productive reallocation of effort or simple displacement? That is a question that organizational-level and process-level research can actually answer, with the right empirical design.

I have written about how structuration theory helps explain why the same tool produces different outcomes in different organizations, and I think that lens applies here. AI automation is not a uniform force that affects all organizations and all workers in the same way. It is mediated by organizational structure, management practice, workforce skills, and the existing task distribution within jobs. Two organizations deploying the same AI system for the same task category may produce very different effects on their workforces depending on how they manage the transition, what they do with the time the automation recovers, and whether they invest in reskilling the workers whose tasks changed.

The 2045 midpoint should function as a planning horizon, not a comfort. Nineteen years is long enough that panic is probably not useful. It is short enough that organizations that start building the adaptive capacity now, the data infrastructure, the governance frameworks, the workforce development programs, will be in a meaningfully different position than organizations that wait. The hype cycle suggests urgency. The actual timeline suggests a more methodical approach. Both signals are probably true at the same time. The technology is real and moving fast. The full organizational and economic impact is slower and more contingent than the headlines suggest.