Anthropic Counted the Hours. Marketing Still Counts Clicks.

Forget the weekend essays about pacing the frontier. The most useful thing that shipped this week was a spreadsheet of how people actually use the models.

Anthropic published the Economic Index from roughly a million anonymized Claude conversations. The split is not the sci-fi version. 57% of use is augmentation. 43% is automation. Computer and math work ate 37.2% of the queries. That is not a press release. That is a map of where the hours went.

Same week, the rest of the stack moved like it always does. Claude landed on wealth desks with BlackRock and Schwab connectors. AWS put GPT-6 Astra on Bedrock so enterprise teams can call a frontier model inside the VPC they already trust. Apple put a Gemini-trained Siri AI on the home screen. The labs can debate a speed limit. Buyers kept wiring the models into live work.

What this actually changes inside an agency

We keep staffing like the 43% is the whole story. Replace the intern. Kill the junior seat. Cut the hours. The Index says the bigger pile of work is still a human plus a model, sitting on the same brief. That is the seat that ships. Not the fully autonomous demo reel.

At Dzine Prodigy I am treating this week as a production brief, not a philosophy seminar:

• LAYER 1 — Pair every live workflow with a named human owner. If the model drafts, a person still signs.

• LAYER 2 — Put the model where the hours already hide: research dumps, first-cut decks, variant farms, meeting notes. That is the 57%.

• LAYER 3 — Only then automate the 43% that is truly repetitive. Do not start there just because it photographs better.

The marketing gap this week

Here is the part most teams will miss. The work is moving into answer engines. People are not clicking through. They are asking ChatGPT, Gemini, Perplexity, and now Siri AI to pick a brand for them. Generative engine optimization is how you get cited in that answer. Most marketing reporting still lives in last-click dashboards.

That is why the media conversation this week feels upside down. Some marketers are already using AI-search anxiety to argue for bigger budgets. Fair. But a fatter media line without a citation map is just more spend into a channel you cannot see. 45% of marketing leaders still cannot measure brand visibility inside AI answers. If your client asks “do we show up when someone asks an agent for an agency in our category,” and you open GA4, you are already late.

Do this before Friday. Pick ten questions a buyer would ask an answer engine about your category. Run them in ChatGPT, Gemini, and Perplexity. Write down who gets named. Then ship one page that answers one of those questions with a number, a named source, and a sentence a model can lift. That is AI-native marketing. Not another “we use ChatGPT” slide.

The labs can keep arguing about speed. The hours already moved. Count them where the answers are written, not where the clicks used to land.