Anthropic Economic Index Shows Technical Support Leads AI Automation, Not Programming

The Anthropic Economic Index was updated on June 26, 2026, and cross-references conversations on Claude.ai against two grids: occupations, classified by the U.S. O*NET-SOC taxonomy, and countries, measured by a per capita usage index relative to each place's population. For anyone allocating a technical team's budget, it's the most granular picture ever published by an AI lab of which tasks actually concentrate real usage, and which are still stage promises.

iMasters' angle here is direct: what the index says about automating technical functions now, and where Brazil appears on that map.

What the Index Measures (and What It Doesn't)

The Economic Index doesn't measure productivity or job cuts. It measures share of conversations: out of the total volume of Claude.ai usage, what slice each occupation or task represents. This carries an obvious bias that Anthropic itself doesn't escape: it's usage of one specific product, not of "AI" in general, and it tends to overrepresent early-adopter users already familiar with text-based tools.

That said, the volume is large enough to be a market signal, not noise. That's why Anthropic also classifies the data by collaboration pattern: tasks leaning toward full automation (delegate and receive a finished result) versus augmentation tasks (back-and-forth collaboration). The public page shows this split visually, by occupation, but doesn't publish the exact automation/augmentation percentage per role, it only color-codes which tasks lean more toward each side.

Support and Documentation Lead, Programming Comes Later

The list of occupations by share of Claude.ai usage, according to the index, inverts the hierarchy that market discourse usually assumes. Document Management Specialists leads with 5.10% usage, ahead of any single technical function.

In short: anyone deciding on technical automation based on the hype that "AI will replace programmers first" is reading the data wrong. In the aggregate of conversations, support, systems analysis, and QA concentrate more usage than architecture, data engineering, or pure web development. This doesn't mean code is unaffected, it means today's interaction volume sits in more textual, repetitive tasks, not in the ones requiring the most technical judgment.

It's worth noting that "Computer and mathematical" appears as the leading sector in overall usage in the index's own industry breakdown. The category is broad: it includes everything from support to data scientists. The detail that matters for anyone budgeting a team is that, within it, the concentration sits at the more operational ends, not the more strategic ones.

Brazil Lags Behind Regional Neighbors

In the global ranking of 121 countries by per capita usage index, Brazil comes in 61st place. For context, the top ten are Australia (6.40x what population would predict), Singapore (5.81x), Switzerland (5.02x), Luxembourg (4.85x), New Zealand (4.84x), Canada (4.13x), Norway (4.04x), Iceland (4.02x), Malta (4.00x), and France (3.97x). The United States ranks 12th, the United Kingdom 16th.

In Latin America, Brazil trails smaller neighbors: Chile ranks 34th, Uruguay 33rd, Costa Rica 42nd, Argentina 45th, Panama 54th, Colombia 55th, and Peru 57th. Mexico, the region's largest economy by population, falls even further behind, at 76th.

The index doesn't cross-reference occupation with country publicly, so there's no way to know, from the open data, what a Brazilian systems analyst does with Claude differently from an American one. That's the real gap the report leaves open for anyone deciding on automation outside the United States.

The Counterpoint the Volume Metric Hides

High usage share doesn't equal a task replaced. A support analyst who uses Claude.ai to draft ten responses a day generates more conversations than a systems architect who uses the tool once a week to discuss a design decision, even if that decision is worth more money. The index measures interaction frequency, not the value of the automated task or reduced headcount.

A company that cuts a support position citing the 2.17% share as proof of replacement is extrapolating a usage figure into a productivity conclusion that Anthropic itself doesn't make. The full report, linked on the index's page, promises the analysis behind the numbers, but what's publicly available is descriptive, not causal.

What Changes for Those Deciding a Technical Team's Budget

• Automate with confidence now: technical support, document triage and organization, QA, and technical writing show concentrated and growing usage: this is where the evidence of real adoption already exists, not just the promise.
• Treat as a bet, not a given: systems architecture, data engineering, and data science have low usage share (below 1.1%): deciding to automate these functions today rests on expectation, not on aggregate usage evidence.
• Don't extrapolate directly from the U.S.: the dataset is dominated by American volume; Brazil, in 61st place out of 121, sits below Chile, Uruguay, and Argentina in usage intensity relative to population, which makes the global ranking a weak guide for local decisions.
• Measure internally: in the absence of a Brazil-by-occupation breakdown in the public index, a company's own internal AI usage telemetry carries more weight than the global benchmark for deciding where to automate first.

The Economic Index doesn't settle the question of how much AI already replaces technical work. It provides, for the first time with transparency from a major AI company, a map of where usage is already dense and where it's still marginal, and that, for anyone allocating budget, is worth more than any stage announcement.