BlogSystems

The work AI took off your desk landed on somebody else's

You already know the expensive version of this. An estimate goes out with a line nobody checked, a follow up promises something the crew cannot do, and fixing it costs more than writing it by hand would have.

Two coworkers lean over one laptop at a cluttered desk while one of them holds a printed estimate

The draft nobody read

Most offices adopted AI the same quiet way. One person tried it, it worked, and a few months later the estimates, the proposals and half the customer emails all start as a generated draft. Nothing about that is wrong. The problem sits at the handoff, because nobody decided who reads the thing before it leaves.

A survey of more than a thousand US desk workers, run by BetterUp with the Stanford Social Media Lab in September 2025, found four in ten had hit workslop in the past month: AI work that looks polished but does not move the task forward. Each instance took nearly two hours to resolve. Those hours come out of somebody's afternoon in your office.

Grenades between desks

This month a chief executive said it out loud. On The Knowledge Project podcast, released September 15, 2026, Shopify's Tobi Lütke said the failure case of lazy work is no longer lack of output. It is over production: people generate more than anyone asked for and leave a colleague to check it. Inside the company they call those slop grenades.

His examples travel beyond software. An engineer approves a code change from an agent without reading it closely, so the review lands on the team. Someone stretches a short point into a long email with a model, and the recipient shrinks it back with another. His advice was to use the model to make a point shorter, not longer. Machines, he said, cannot take responsibility for work, and people can.

The cleanup already sells

A market has appeared for fixing this, which is the clearest evidence the work never disappeared. Internal data from freelance marketplaces, reported this month, shows listings tagged correct AI and AI hallucination grew 87 percent between August 2025 and June 2026, reaching 10,760 worldwide. Another platform reported a 70 percent year over year surge in AI remediation work, and on a third, searches for AI cleanup rose more than twentyfold since 2023.

When cleanup goes outside, you get an invoice and you see the number. When it stays inside, it is paid with a coworker's hours and never appears on a line of your books. It shows up instead as a week that feels full while less finished work actually leaves the office.

One reviewer, one stop

The repair is organizational, not technical, and it fits in one meeting. Every AI assisted output that reaches a customer gets one named reviewer and one point where the draft stops being a draft. Named means a person, not a department. The stopping point is when that person puts their name on it and sends it, which is also when responsibility lands somewhere real.

You assign it once, by document type. Estimates go to whoever prices the work, customer emails to whoever answers the phone, proposals to you. From that week on, what leaves the office was read once on purpose instead of coming back Thursday with a correction attached. The draft is as fast as it was yesterday. What you stop paying for is the second round.

Source: Search Engine Journal

Spica Studio