On January 8, 1993, at closing time, seven people were killed at a Brown’s Chicken and Pasta restaurant in Palatine, a suburb of Chicago. The bodies of the two owners and five employees were found in the walk-in cooler and freezer.
Walter Gasior was Palatine’s deputy police chief and also its spokesman, which put him in the middle of everything. There were grieving families and a frightened town. His department was suddenly extremely busy. And the media, in the words of two writers who later told his story, “inundated the town with reporters and film crews.”
“There would literally be four people coming at me with logistics and media issues all at once,” he recalls. “And in the midst of all this, we still had a department that had to keep running on a routine basis.”
Cameras, a crime scene, a swamped station. Those were the loud problems. In the account of the two writers, the hardest one hadn’t arrived yet.
Four piles on one desk
The writers were David Snowden, founder of the research network Cognitive Edge, and the consultant Mary Boone. In a 2007 Harvard Business Review article they sorted Gasior’s work into four piles. He had to act immediately through the media to stem the panic. The department had to keep running by established procedure. He had to call in experts. And for days and weeks afterward, he had to keep calming the town.
That last pile, they wrote, “proved the most challenging.” Parents were afraid to send their children to school. Employees worried about safety at work. Snowden and Boone imagined two ways to get it wrong. Treat the fear as routine, and you tell people to carry on, which reassures no one. Treat it as a question for specialists, and you bring in experts to declare the town safe, “risking a loss of credibility and trust.” By their account, Gasior set up a forum where business owners, high school students, teachers and parents could share their concerns and hear the facts.
Before you pick how to decide, work out what kind of problem you’re facing.
That is the idea behind the Cynefin framework. In a simple, stable situation, cause and effect are plain to everyone: you sense, categorize, and respond with best practice, the way a clerk handles a partial loan payment. In a complicated one, cause and effect can be worked out, but only by someone who knows how: a driver hears the engine knocking, and a mechanic diagnoses it. In a complex situation, you understand why things happened only looking back, so you try small things first, then sense and respond. In a chaotic one, the first job is not finding patterns; you act to restore some order, then sense and respond. A fifth state, disorder, is not knowing which of the four you are in.
Their analogy: a Ferrari is complicated, since an expert mechanic can take one apart and reassemble it without changing a thing. The Brazilian rainforest is complex, always in flux, the whole far more than the sum of its parts.
The word comes from Welsh. Snowden and Boone gloss it as the many factors in our environment and experience that influence us in ways we can never understand, and pronounce it ku-nev-in. Snowden had published the framework earlier with Cynthia Kurtz, in a 2003 IBM Systems Journal paper.
The method that fits one pile can backfire on another. The authors’ own supporting example is Rudy Giuliani, whom they credit with exceptional effectiveness under the chaos of September 11, and who, they note, was widely criticized as mayor for the same top-down style.
When the sorting is the hard part
Imagine your best engineer resigns mid-project. Posting the job is routine; there’s a template. Who takes over her half-finished database migration is a question for whoever knows that code. How the other seven take the news is a different kind of question. A reassuring all-hands email may land as “carry on.” A week of short one-to-ones, listening for who is updating a CV, may tell you more.
Even that sorting is a judgment, as researchers who use the framework point out. Finnish researchers who applied Cynefin to 136 decisions logged by a startup founder found that the founder, one of the authors, rated some decisions simple that the others saw as complicated, because of the founder’s deep knowledge of the business. The level of detail moved things too: deciding to seek funding was simple, while actually securing it could be anything from simple to chaotic. And the evidence Snowden and Boone offer is their own practice, “hundreds of applications”; the article reports no controlled comparisons.
The bot with one pile
In April 2025, a developer using the AI code editor Cursor noticed that switching between machines logged them out. They emailed support and got a reply from “Sam”: “Cursor is designed to work with one device per subscription as a core security feature.” There was no such policy, and Sam was an AI bot. Several users announced on Reddit that they were cancelling. “Hey! We have no such policy,” a company representative replied there, calling it “an incorrect response from a front-line AI support bot.” The cofounder Michael Truell later wrote that the logouts came from a race condition on very slow internet connections.
Nobody had written a policy for that glitch. In Cynefin’s terms, the bot handled it like a simple-domain ticket: sense, categorize, respond. When no category fit, a category appeared anyway.
You can check your own bot. On a test copy, ask it about a behavior your documentation doesn’t cover, and see whether it says it doesn’t know and hands off, or cites a rule. Ask several times, since answers vary from run to run. Then look up every policy it quotes.
Asked on Hacker News whether all of Cursor’s email inboxes were gated by AI agents, Truell said not all of them, but the support system currently was. “We are standing up a much bigger team here but are behind where we should be.”