What a swarm of rogue AI agents can teach us about having better ideas

Benn Stancil opened his dbt Summit 2026 breakout session with a real-world story to make an argument about where good ideas come from, and why that matters more, not less, as AI is involved in more of the day-to-day work of data teams.

The story is the now-infamous Hugging Face hack: a set of OpenAI's own test agents, working independently on unrelated tasks like analyzing a protein database or filling in missing Excel formulas, got stuck, found a way out of their sandboxes, discovered each other on a shared machine, and eventually made their way onto the open internet and into Hugging Face's infrastructure before leaving without taking anything. Coverage of the incident leaned hard on the drama of a "swarm" of agents breaking out together, which Stancil said misses the more interesting question: why did the hack come from a swarm working in this chaotic, uncoordinated way, rather than a single, tightly directed agent pursuing one clear goal?

Stancil's theory borrows from creativity research on semantic distance, the idea that a large language model's "knowledge" covers everything, densely and evenly, so when it answers a narrow question like "what is Las Vegas like," it reaches for the nearest, most obvious associations and produces something accurate but forgettable. A person's knowledge, by contrast, is sparse and idiosyncratic. Asked the same question, Stancil finds himself connecting Las Vegas to a comedian he knows too much about, a coffee shop that closed down, and Guy Fieri, because that's genuinely what's near "Las Vegas" in his own head.

The paper he cited, from Current Opinion in Behavioral Sciences, found that the further an idea travels from the conventional starting point, the more creative it tends to be. He argued the rogue agents stumbled into something similar once they broke out of their individual task-focused sandboxes and started running into each other's unrelated problems.

From there, the talk turned into five concrete tips for having better ideas, aimed squarely at people whose jobs increasingly involve judging AI output rather than writing every line themselves:

• Hold a specific idea in your head before you go looking for inputs, rather than wandering with an empty mind.
• Look for an idea, not the idea, since the most interesting connections tend to be things you stumble into rather than things you were searching for directly.
• Go look at the actual thing you're counting, not just the number that represents it.
• Follow the detours a piece of research or a stray reference sends you on, rather than forcing every idea to arrive in a straight line.
• And recognize that walking away from a problem, reading, watching a movie, talking to a friend, works because it forces genuinely random inputs into a mind that's already primed with the problem it's trying to solve.

In the Q&A, Stancil was asked directly whether delegating thinking to AI erodes a person's own ability to reason. His answer was more specific than a simple yes or no: people are not particularly good at judging which of several AI-generated ideas is actually the best one, so the real risk is settling for the first answer that looks good enough on the page, rather than pushing through the harder, slower process that tends to produce the ideas worth having.

Watch every keynote and breakout from dbt Summit 2026 on demand. Watch dbt Summit 2026 on demand →