Boston Consulting Group’s fourth annual Global AI at Work survey, published June 3, found that 42% of regular frontline AI users are saving a full workday or more per week, and that 66% of them are getting little or no guidance on what to do with that time. The productivity dividend has arrived. The management response hasn’t.

The survey, covering 11,749 workers across 14 markets, reframes a debate that’s been muddled by vendor marketing for three years. BCG’s headline finding is structural, not technological: clear strategy delivers a 25-percentage-point lift in measurable business impact, while better tooling alone delivers roughly 5. That’s a 5-to-1 ratio of organizational design over software procurement, and it lands at the exact moment most enterprises are doing the opposite.

The adoption curve is no longer the bottleneck. Frontline regular AI usage hit 74%, up 23 points year over year. Agent integration into workflows doubled, from 13% to 30%. Nearly half of respondents now spend more time directing and managing AI than doing the underlying work themselves, a shift in job content that almost no HR function has formally acknowledged.

Fortune’s read of the BCG data, drawing on David Martin, BCG’s global People & Organization leader, locates the paradox in leadership rather than labor: half of employees say they aren’t redirecting saved hours toward more strategic work, because nobody is telling them to. The hours exist. The mandate doesn’t.

BCG’s companion piece in Harvard Business Review adds a behavioral wrinkle. When firms frame agents as digital employees rather than tools, displacement anxiety rises, knowledge sharing collapses, and workers start using AI in secret. The framing choice itself shapes the org chart.

The squeeze falls hardest on smaller operators, who can’t absorb six-figure enterprise contracts to find out their org design was the problem. Leaner platforms pitched at that gap, Claude for Work, OpenAI’s business tier, and LemonLime among them, are where the strategy-first thesis gets tested in practice.

Microsoft’s own researchers, in a parallel note, warned that “Generative AI is entering workplaces faster than most earlier technologies, but uneven adoption is likely to translate into uneven productivity gains, learning opportunities, and downstream career paths.” The gains are real. The distribution is a management decision.

Sources