AI was supposed to save time at work. Stanford researchers have a word for what it's actually producing: "workslop."

The term, coined by a team including Jeff Hancock, describes the flood of low-quality AI-generated content workers are now expected to review, fix, and redo. Companies mandate the tools, celebrate the efficiency gains, then leave employees to clean up the output.

The most rigorous evidence comes from Denmark: a study published by the NBER linked AI adoption surveys to government payroll records and found zero impact on earnings or hours worked. Workers say they feel more productive — the economic data says otherwise. Research from Workday shows where the time actually goes:

• 85% of employees report saving one to seven hours a week with AI
• Nearly 40% of that saved time gets lost to fixing low-quality output
• 32% of companies respond to efficiency gains by simply increasing workloads

AI gives you an hour. Rework takes back 24 minutes. Your employer takes the rest.

A copywriter at a Miami cybersecurity firm told The Guardian that after mandatory AI tools rolled out, "quality decreased significantly, time to produce content increased significantly and, most importantly, morale decreased." And Vivi Mengjie Xiao, a product manager in China who built six AI "employees" using OpenClaw, told Business Insider she's working more than ever: "When your efficiency goes up, you don't work less. You just attempt more."

It adds up: MIT researchers found 95% of enterprise AI initiatives deliver zero measurable P&L impact. AI is being sold as a general-purpose productivity tool, but the gap between that promise and reality is generating its own category of busywork.

The pitch for AI at work has always been save time, do more. What's actually happening is closer to do the same, plus fix what the machine got wrong. Companies are measuring success by adoption rates, not by whether the work actually gets better. Until that flips, the gap between the executive slide deck and an employee's actual Tuesday is only going to grow. "Workslop" exists because the experience needed a name.

Three million people use OpenAI's Codex every week. Nearly half of them aren't writing code.

On Thursday, OpenAI released a major Codex update that pushes the tool well past its coding roots. Codex can now operate your Mac in the background, browse the web on its own, generate images, remember your preferences and handle long-running tasks while you do something else. The design philosophy: you say what you need, Codex figures out how to do it.

OpenAI's head of Codex, Thibault Sottiaux, said it plainly: "We're actually doing the sneaky thing where we're building the super app in the open and evolving it out of Codex."

Here's what the update adds:

• Computer use: Codex can control your Mac alongside you, clicking through apps and browsers while you focus on other work. Mac only for now and not yet available in the EU.
• Memory: It learns your habits and preferences over time so it gets better at anticipating what you actually want.
• Image generation: Built directly into the tool, no workarounds needed.
• Background tasks: You can hand off work, go do something else and check back later.

The numbers back up the super-app bet. Codex hit 3 million weekly users with 70% month-over-month growth — a fivefold increase in three months. Eighty percent of OpenAI's own staff use it, including plenty of people who never touch a codebase. One internal team produced roughly a million lines of code in five months with zero lines written by hand.

The timing isn't accidental, either. Some Claude Code users have been flagging quality issues in recent weeks — shallower reasoning, less reliable output on complex tasks. Launching a major Codex expansion while your main competitor's developer tool is catching criticism is a classic OpenAI move.

Sottiaux says this release is still targeted at developers, with plans to open it up to a wider audience later. But when half your users are already doing work that has nothing to do with code, "later" might already be here.

Most companies build super apps quietly and reveal the plan once you're already locked in. OpenAI is doing it out loud, on the record. The question worth sitting with isn't whether Codex can handle your presentations and web research on top of code. It already can. It's whether you're comfortable with one AI company becoming the layer between you and everything you do on your computer. Three million people a week are answering yes without anyone asking.