Anthropic published the latest edition of its Economic Index on June 26, 2026, and the headline number is striking: roughly half of the approximately 9,700 Claude users surveyed report that AI can already handle 50% or more of their work tasks today. A further 35% expect AI to handle most or nearly all of their work within the next 12 months.

The report, titled “Cadences,” links survey responses directly to actual usage data from Claude.ai, Cowork, and Claude Code across a window in May and June 2026. That methodology matters — instead of asking people how they feel about AI in the abstract, Anthropic matched each survey respondent’s answers to up to 20 sampled AI sessions from their recent usage history, giving a much clearer picture of how work is actually changing on the ground.

What “Cadences” Actually Found

The report’s name comes from one of its more novel findings: AI usage follows recognizable rhythms tied to real human behaviour. Tax-related queries surge around filing deadlines. People ask for news in the morning. Sleep and health advice peaks around 5 a.m. Work queries fall off sharply on weekends — though less so in the highest-paid occupations, where the boundary between “work” and “life” was already blurry long before AI arrived.

These patterns aren’t just interesting trivia. They suggest AI is becoming embedded in daily routines the same way email or search did — it’s becoming infrastructure, not a novelty people consciously switch on.

The Substitution Numbers Are Real

The 50% substitution figure deserves some context. When workers say AI handles most of their tasks, they don’t necessarily mean it’s replacing them entirely — many describe a shift where AI handles drafting, research, summarisation, and first-pass analysis, while they focus on judgment, client relationships, and decisions that require real-world context.

That nuance shows up elsewhere in the data. Experienced workers consistently cite the same things AI hasn’t replaced: tacit knowledge built over years, the ability to read a room, management, and accountability. These are real limits, and the workers closest to the work seem to understand them clearly.

The geographic split is also worth noting. Users in lower-GDP countries report higher rates of AI task substitution than their counterparts in high-income economies. Anthropic’s earlier reports flagged the same trend: Claude is used in more automated and assistive modes in markets where professional expertise has historically been scarcer or more expensive.

The Optimism Bias Problem

One of the more psychologically interesting findings is a classic optimism bias in how workers interpret these numbers. Survey respondents consistently feared AI’s impact more for junior colleagues than for themselves. People tend to think AI will reshape entry-level and administrative work while their own senior judgment remains essential.

This is almost certainly partly true — and partly a cognitive blind spot. The same workers saying AI “can’t replace my judgment” are the ones reporting that AI already handles half their actual task volume. The distinction between “does tasks” and “replaces a role” is real, but worth examining carefully.

What This Means for Business Leaders

If your teams are already using AI to handle the majority of their day-to-day tasks, two questions follow immediately.

First, are you capturing the value? Informal AI use that compresses the time it takes to complete existing work doesn’t automatically produce better outcomes — it just produces faster ones. Capturing the actual productivity gain requires deliberate choices about what people do with reclaimed time.

Second, are your people ready to work alongside systems that are getting more capable each quarter? The workers who flagged judgment, context, and relationships as AI’s limits are right — for now. But the capability gap is closing faster than most organisations have planned for.

The cadence finding itself is a useful lens here. AI is being woven into the rhythm of work — mornings, deadlines, projects — in ways that compound quietly. Businesses that treat AI adoption as a one-off implementation project rather than an ongoing capability-building effort are going to find themselves a long way behind.

The practical implication is straightforward: the teams that understand AI well enough to direct it, verify it, and escalate beyond it will consistently outperform the teams that just use it. Data literacy and AI fluency are no longer soft-skill nice-to-haves. They’re core operational competencies.

Enterprise DNA’s learning platform exists precisely for this moment — 220,000 data professionals across 50+ countries have used it to build the skills that make AI more useful, not just more accessible. If your team is at the point where AI handles half the work but no one has mapped what that actually means for your business, that’s where to start.

---

Source: Anthropic Economic Index Report: Cadences, published June 26, 2026.