OpenAI's Own Staff Now Hand Agents Their Longest Tasks. In Marketing, the Jobs You Start In Go First.

OpenAI just published the numbers on what its own staff hand to AI agents, and the tasks are no longer small. In a report released in late June, “How agents are transforming work,” the company said 70% of sampled employees gave its Codex agent at least one task worth an hour or more of human work, and a quarter handed it jobs worth a full eight-hour day. For the average worker, Codex now accounts for more than 85% of output tokens.

Ronnie Chatterji, OpenAI’s chief economist, framed the report as a first read on something the rest of the market is about to feel, calling it the first “economic study of agents”:

“The paper you have today is trying to analyze the data to tell the world what’s happening with AI.”

The shift underneath those numbers is the one every marketing leader should read twice. Work is moving from a quick answer you ask for to a whole task you delegate. The question stops being what a tool can tell you and becomes what you can stop doing yourself.

And it isn’t staying with engineers. Per OpenAI’s own account, employees in Legal, Finance, and Recruiting now use Codex regularly for automation, analysis, and other technical work once outside their job. When an agent can carry an eight-hour task, the line between a “technical” role and a “knowledge” role stops meaning much.

The first marketing jobs to go are the ones that train people

Marketing sits squarely in the blast radius, and the people who study it are blunt about where the hit lands first. Paul Roetzer, founder of the Marketing AI Institute, put it plainly to MarTech:

“If you were just executing (building landing pages, writing copy, etc.), you’re cooked. That is not a job one to two years out.”

That’s the exact work that used to be an entry-level marketer’s whole day. Julie Bedard, a managing partner at Boston Consulting Group, told the same outlet a marketing manager’s tasks are “90% disrupted from a skill perspective.”

Roetzer’s diagnosis explains why the bottom rung goes before the top: “the gap between what AI can do and what entry-level workers have typically been asked to do has effectively narrowed to zero.” Junior marketing work was always the most delegable: bounded briefs, clear templates, low context. That’s precisely the shape of task an agent now runs for hours without a babysitter. Senior work that turns on judgment, taste, and relationships is harder to hand off, so it stays. The training ground vanishes while the destination stays put.

The door closes quietly

The result is a hiring pattern that shuts new marketers out without a single announced layoff. The labor data already shows it: MarTech reported early-career workers aged 22 to 25 are seeing a 16% relative employment decline while experienced workers hold steady. It matches what this publication has tracked in marketing specifically, where the career ladder is losing its bottom rung and B2B teams are shrinking through roles that never get backfilled rather than through cuts anyone puts a number on.

So the near-term question for marketing leaders isn’t whether to use agents. Adoption is already broad, with KPMG data cited in the same OpenAI report putting employee agent use at 68% and serious pushback at just 2%. The question is what a team looks like in three years when nobody hired a junior.

Two moves matter now. Stop defining entry roles by the tasks agents already do, and start defining them by the judgment those tasks used to teach: writing the brief, editing the machine, owning the outcome. And give someone explicit ownership of how work gets delegated to agents, the way you’d staff a new hire, because an agent running eight-hour jobs with no owner is unmanaged headcount. The teams that win the next cycle are the ones that keep training humans while the machines do the homework.