The spreadsheet rebellion that taught us how to roll out new tools to teams

Before the question was "how do we get our team using AI," it was "how do we get our team using spreadsheets." And before that, it was mainframes. The pattern repeats so reliably across technology cycles that it is almost embarrassing how often organizations treat each new wave as a unique organizational puzzle requiring a brand-new solution.

The "before" in this story is the late 1970s and early 1980s. Most business calculations lived in the heads of analysts, in paper ledgers, or in the hands of a small priesthood of data-processing staff who controlled access to mainframe time. When VisiCalc appeared in 1979 and Lotus 1-2-3 followed in 1983, individual employees suddenly had computational power sitting on their desks. The tools were genuinely powerful. And most managers had absolutely no idea what to do with that fact.

The typical organizational response was to ignore it, restrict it, or delegate it. Finance departments often banned personal computers from "serious" forecasting work for years. IT departments tried to control which software could be installed. Individuals who discovered the tools on their own became informal experts, sharing knowledge in corridors rather than through any structured process. Teams adopted the same tool in twelve incompatible ways. The result was a decade of productivity gains that were real but wildly uneven, with most of the value captured by a handful of early enthusiasts while the rest of the organization watched.

The turning point

The shift in how organizations thought about technology adoption did not come from a single visionary. It emerged, messily, from a combination of academic research and expensive corporate failure.

In the early 1990s, researchers studying organizational behavior began documenting why technology rollouts failed so consistently. Everett Rogers had published his foundational work on the diffusion of innovations back in 1962, but it took the personal computing era for corporate training and change management teams to actually apply his framework to workplace tools. Rogers had shown that adoption within a population follows a predictable curve, and that the people in the middle of that curve, the early and late majority, do not behave like early adopters. They need social proofsocial proofThe tendency of people to look at others' choices to guide their own. In marketing, it means using reviews, testimonials, ratings and case studies to reassure and persuade prospects.View full definition →, they need to see colleagues succeed before they commit, and they respond badly to top-down mandates without visible leadership buy-in.

The corporate world got a brutal illustration of this around the same period. Several large firms, including some in financial services and manufacturing, invested heavily in enterprise resource planning systems through the mid-1990s and watched adoption collapse. The SAP rollouts of that era became cautionary material in business schools precisely because the technical implementation often succeeded while the human adoption failed. Employees had workarounds. Managers quietly permitted the old processes to continue alongside the new system. The software sat unused.

What changed the practice was the gradual formalization of what eventually became known as change management as a discipline. Prosci, an independent research and training organization, began publishing structured research on organizational change in the mid-1990s and developed its ADKAR model (Awareness, Desire, Knowledge, Ability, Reinforcement) as a framework for thinking about individual transitions rather than organizational transitions. The insight sounds simple in retrospect: organizations do not change, individuals do, and they do so one at a time, at different speeds.

That reframing was genuinely consequential. It moved the focus away from the technology and toward the person sitting in front of it.

From there to now

The through-line from Rogers and the failed ERP rollouts to 2026's AI adoption challenge is direct. The teams rolling out Copilot, ChatGPT Enterprise, or custom LLMLLMA Large Language Model is an AI system trained on vast text data to predict and generate language, enabling tasks like writing, summarizing, and answering questions.View full definition → tools today are dealing with the same structural problem: a capability that is unevenly distributed across a workforce, adopted enthusiastically by a minority, ignored or feared by a larger group, and often undermined by management that does not visibly use the tools themselves.

The specific research on AI adoption reinforces the older pattern. According to MIT Sloan Management Review's ongoing work on AI in organizations, one of the most consistent predictors of successful AI adoption is whether direct managers actively demonstrate use of the tools, not whether the company has issued a policy or run a training session. This aligns precisely with what Rogers documented about social proof more than sixty years ago.

What has changed is the speed of the cycle and the stakes of falling behind. In the spreadsheet era, a team that was slow to adopt lost some efficiency. In the current environment, the productivity gap between teams that have genuinely embedded LLM tools into their workflows and teams that have only nominally adopted them appears to be compressing in months rather than years.

Why it still matters

The origin story matters for a practical reason: it tells you where to focus your energy. Most AI rollouts fail not because the technology is hard to learn but because the organizational conditions for adoption were never established. The mistakes made during the spreadsheet era, skipping visible leadership use, treating training as a one-time event, ignoring the informal networks through which real knowledge actually travels, are being repeated in real time across organizations deploying AI tools in 2026.

The teams that adopted Lotus 1-2-3 well in the 1980s were not the teams with the best training manuals. They were the teams where a senior person was genuinely using the tool in front of others and talking about it openly. The same holds now.

If you are responsible for an AI rollout, the most useful thing you can do before spending anything on licenses or training content is to identify three or four people in the team who are already using AI tools informally and give them a legitimate platform. They are your early majority bridge. Rogers described this dynamic in 1962. It still works.