The superpower is curation: why curated knowledge lifts both humans and AI

The first instinct when working with artificial intelligence is almost always the same: more data. More documents, more sources, more context. But the real power isn't in the volume, it's in the curation. The numbers are clear: according to research by IDC and Seagate, companies use only about a third of the data available to them on average, and four out of five records sit unstructured. Knowledge workers, McKinsey found, spend roughly a fifth of their working time just searching for and gathering information. The bottleneck isn't access to information, it's access to ordered, trustworthy knowledge. That is exactly where the superpower lies: curated knowledge that both human and artificial intelligence can work with at their best.

"Garbage in, garbage out" is older than any AI, but the principle has rarely been this expensive. Gartner puts the cost of poor data quality at an average of USD 12.9 million per organisation per year. The reason is simple: more data doesn't automatically amplify the signal, it often amplifies the noise. An AI built on top of a disordered archive of outdated slides, contradictory versions and half-finished notes inherits exactly those contradictions and passes them on with the persuasive fluency of a well-phrased answer. It isn't compute that decides the quality of an answer, but what underlies it.

The proven way to ground an AI is called retrieval-augmented generation: the AI answers not from its generic training, but exclusively on the basis of a defined knowledge source. That is precisely how a Wingmate works: it thinks and argues on the verified knowledge you provide as a Companion. But grounding alone is no guarantee: a Stanford study found that even specialised, knowledge-backed AI tools in the legal field still hallucinated in up to a third of cases. The difference between a helpful and a misleading answer isn't whether the AI draws on a source, it's what it draws on. Retrieval only becomes reliable through the curation behind it.

And curation is not a one-time upload. It is an ongoing, human act of judgement: deciding what is authoritative and what is outdated, what belongs together and what belongs in the bin. On Violet Dragonfly that is the curated knowledge base: you, as a Companion, decide what your Wingmate knows and in which voice it speaks. And where a case exceeds the curated knowledge, human-in-the-loop closes the loop: a human specialist takes over, and their new insight can flow straight back into the knowledge base. That is how we built our own consulting practice at thinking services. Building the Wingmate was also the occasion to order our own knowledge properly for the first time.

Curated knowledge is the shared ground on which humans and machines alike perform at their best. The AI answers faster and more precisely because it draws on verified sources rather than guesses. And the human wins back exactly the time that used to drain away into routine questions, for the cases where experience and judgement truly count. The very process of curation turns implicit knowledge, scattered across individual heads, into a shared asset: something that benefits not just your clients, but your own team too, who can use it to onboard new colleagues or reach internal best practices in an instant.

The tools will keep getting better and the models keep getting cheaper and, with that, more interchangeable. The lasting difference won't be which model you use, but the quality of the knowledge you entrust to it. That is a process, not a product; a discipline, not a one-off effort. Whoever masters curation gives people and AI alike the same superpower. We built Violet Dragonfly around exactly this idea. Take a look around, and reach out if you'd like to order and scale your own knowledge.