AI in Business: Strategy Should Return to Practitioners

Proliferation of unofficial tools, constrained budgets, and consumption-based pricing are reshaping the landscape of AI tooling. Data from Microsoft, McKinsey, Gartner, UX Tools, and Anthropic converge on a single idea: business experts are better at judging the quality of outputs and should guide decisions, while operations focus on safety and contracts.

Costs, Tool Stack, and Job Risks Place Strategy Under Pressure

The proliferation of tools is described as a signal of past operational failures that ultimately cost jobs, for which operations must take responsibility. According to a survey conducted by UX Tools, 53% of designers believe that the increasing number of solutions slows them down. At User Interviews, eight specialists report that managing tools takes up an increasing portion of their work, while 17 say their scope has expanded and 12 indicate that budget, personnel, or resources have remained stable or decreased. It is suggested that, without demonstrable business impact, these roles could disappear. According to Gartner, coding vendors are shifting from seat-based pricing to consumption-based pricing, which requires assigning tasks to the appropriate models with business knowledge. This phenomenon extends beyond design and affects all functions.

Defining Who Decides: Experts Set the Course, Operations Ensure Security

The evaluation gap separates the approver of a tool from the one who can judge the accuracy of its output. As long as the field is not stabilized, expert users are presented as directly responsible for adoptions and acceptable outcome criteria. The proposed governance delineates rights: judgment and strategy on the side of business experts, and security, privacy, contract review, data limits, access, budget, and maintenance on the operations side. Errors are expected, hence the need for clear accountability: a researcher will be able to explain a failure related to fabricated citations and correct the review, while an operations manager who has never conducted a study risks primarily canceling a contract too late. This distribution is framed as transitional and is expected to be reevaluated once the field stabilizes.

Microsoft Reports That 78% of Employees Use Their Own AI Tools

The presence of shadow tools is cited as the first signal of a strategy disconnected from actual usage. In 2024, Microsoft finds that 78% of AI users at work bring their own tools. In the spring of 2026, UX Tools reports that 59.1% of designers have developed an AI tool or utility in the past six months. Each unreferenced tool is interpreted as a vote for the quality perceived by practitioners and against the understanding of operations. Slack has long circulated as a shadow tool. Some operational teams are described as lagging, and in a User Interviews study from December 2025 involving 21 specialists, six doubt the sustainability of their roles, while organizations are renaming teams to agent operations. Beyond design, reorganizations around AI are underway: McKinsey observes that nearly three-quarters of high performers have redesigned their workflows compared to a quarter of others. In 2023, Harvard Business Review headlines that "AI will not replace humans, but humans with AI will replace humans without AI."

Why Business Expertise Outperforms Formal Controls

Generative tools vary by model, management of long transcriptions, and source exposure, nuances that the absence of expertise fails to capture. In business law, a summary may seem correct while omitting an indemnity exception or confusing assignment with change of control. Generative systems consistently produce results that appear credible but contain errors hidden in details that only professionals can detect. According to a preliminary study conducted by Anthropic in June 2026 on approximately 400,000 Claude Code sessions, experts achieve verified success more than twice as frequently as novices, which is attributed to domain mastery. For instance, a product manager notices the absence of a company usage cap, while a designer identifies the creation of a fourth button style. The illustration of a researcher testing a summary against her interview notes shows that field experience allows for the identification of a fabricated citation within a minute, whereas someone without practice might not see it.

Indicators: Beyond Seats, Measuring Veracity and Token Cost

Operations traditionally track seats, expenses, renewals, and connections, metrics suited for licenses but insufficient to guide an AI strategy. Proposed indicators focus on quality and cost: the share of claims from summaries that withstand verification against transcriptions, and the token cost per completed study. The shift mentioned by Gartner towards consumption-based pricing reinforces this need: charges must be directed towards the appropriate models, which requires a fine understanding of tasks.

Lessons from the Past: VisiCalc, Slack, and the Mistakes of Major Sites

The launch of VisiCalc in 1979 illustrates managers' ability to judge a tool's usefulness in minutes, to the point that Apple IIs were found on desks at Tandy to run the spreadsheet, despite the company's internal offering. Slack reports in its S-1 (2019) a bottom-up diffusion, team by team, partly because internal tool managers did not understand the use cases. Jakob Nielsen measured in 1999 an average rate of 20% violations of his "Top Ten Mistakes" among the 10 largest American companies, compared to 13% among the 10 most visited sites. The current stack of unofficial tools is presented as a comparable sign of the gap between official plans and real needs. Nevertheless, operations remain crucial, as Peter Merholz reminds us in 2020, with AI adding technical constraints—contracts and workflow maintenance—that require dedicated support.