What happens to your workday when ChatGPT goes down? A team of South Korean researchers decided to find out by running a four-day diary study with 10 knowledge workers who use LLMs (large language models - the technology behind ChatGPT and similar tools) frequently in their jobs.
The findings, accepted at CHI 2026 (the top academic conference for human-computer interaction), landed on three themes. First, the obvious one: workflow disruption. Participants hit walls in tasks they'd been completing smoothly with AI assistance, exposing just how deeply the tools had embedded themselves into daily routines.
The second finding is more interesting. Without AI to lean on, participants engaged in more self-directed work and reported rediscovering their own professional values and priorities. Forced to think through problems manually, some found they preferred certain aspects of unassisted work.
Third, the study documented how thoroughly "normalized" LLM use has become among knowledge workers. Participants didn't treat ChatGPT as a novelty or optional extra - they treated it like email or Slack, as basic infrastructure that's simply expected to be there.
Ten participants over four days is a small sample, so treat this as a signal rather than a conclusion. But the pattern matches what many daily AI users quietly suspect: these tools have become load-bearing walls in our workflows faster than we realized, and most of us have no fallback plan for when they're unavailable. The researchers suggest a "value-driven appropriation" approach - basically, being intentional about which parts of your work you hand to AI and which you keep, rather than defaulting to AI for everything just because it's there.