The productivity debate around enterprise artificial intelligence (AI) has narrowed to one question: how much faster can workers complete existing tasks? Anthropic’s internal research suggests that framing leaves something out. The company found that 27% of AI-assisted work inside Anthropic came from tasks employees wouldn’t have attempted without AI. That work wasn’t impractical because it lacked value. The time cost made it impractical.
What the Data Showed
Anthropic surveyed engineers and researchers across the organization, conducted 53 in-depth interviews and analyzed 200,000 internal Claude Code transcripts. Employees reported using Claude in 60% of their work and estimated productivity gains averaging around 50%, up from 20% the prior year. Usage rose from 28% of daily work to 60% over the same period.
The output data is more concrete. Across nearly every task category, employees reported slightly less time spent per task but substantially more output volume. Claude Code usage shifted toward more complex work: the average number of consecutive tool calls the model completed without human input doubled from roughly 10 to 21, and the share of tasks involving new feature implementation grew from 14% to 37%.
Engineers described using AI to build interactive dashboards, scale deprioritized projects, fix long-neglected code quality issues and run exploratory research that wouldn’t have justified the time cost manually. One researcher described running multiple Claude instances in parallel to test different approaches simultaneously, treating the model less like a faster car and more like a fleet.
OpenAI’s enterprise research found a similar pattern, with 75% of surveyed workers reporting they could complete new tasks they previously couldn’t perform. EY’s US AI Pulse Survey found that 39% of organizations were reinvesting AI-driven productivity gains into research and development, suggesting the expansion effect extends beyond individual task completion.
Where Enterprises Are Struggling
The broader enterprise picture is less uniform. PYMNTS Intelligence found that 71% of executives at companies with at least $1 billion in annual revenue identified organizational readiness as the primary limit on AI performance. Only 11% said the technology itself was the barrier. PYMNTS Intelligence reported that 58% of CFOs named talent shortages as a leading challenge, rising to 71% among services firms.
Advertisement: Scroll to Continue
Cost control is a parallel pressure. PYMNTS reported that Uber’s AI budget surged past projections as internal use of Claude Code scaled, with roughly 11% of live updates to its back-end systems now written by AI agents and research and development expenses rising 9% to $3.4 billion in 2025.
The Economics of Execution
When AI reduces the cost of analysis, documentation, coding and research, work that once sat below the viability threshold moves above it. Deloitte found that only 34% of enterprises are using AI to deeply transform core processes and products, while the remaining two-thirds are capturing efficiency gains without redesigning underlying operations.
The internal findings carry limits. Anthropic engineers have early access to frontier models, work in a stable field and are themselves building the technology. The company acknowledged the findings don’t generalize directly to other organizations. PYMNTS Intelligence found that 34% of CFOs at large companies cited productivity as the top reason for AI adoption. Anthropic plans to expand the research beyond engineers to understand how AI affects roles across the organization, with further findings expected in 2026.
