As AI makes generation cheap, the human advantage shifts to curating context. The product people who thrive will be the ones who learn to work like DJs, selecting, sequencing, and adjusting the inputs that shape the outcome.
It’s a summer afternoon in Seattle. A crowd fills the patio of a coffee shop, and a DJ, playing out of an old VW van turned stage, shifts tracks at exactly the right moment. The energy changes. People who were shuffling their feet a minute ago break into full dance. You’ve felt this before: at a wedding, a festival, a club. When a DJ plays the right song at the right time, a crowd becomes something more than the sum of its people.
I run an event business as a side gig, and I’ve hired many DJs over the years. Here’s what that experience taught me: almost anyone with basic technical skills can operate the equipment. Very few can control a room. The ones who can don’t create any of the music they play. Their value comes from somewhere else entirely, deep knowledge of their genre, a wide catalog to draw from, the ability to read the crowd, and the confidence to commit to the right track at the right moment. They create value through curation, sequencing, timing, and judgment.
A recent piece published here called out how it’s time to make context more retrievable through better Information Architecture. Building on that, I share how knowledge work in the age of AI now demands the same skills of a DJ, and it is up to design to teach those skills through our products.
Generation is cheap. Selection isn’t.
In a little over a year, we’ve moved from a world where transforming raw material into polished output was a differentiating skill, to one where anyone with a chatbot can do it. A document appears in minutes. A working prototype in an afternoon. An entire website from a few prompts. We are awash in content, and the volume is only accelerating.
But just because something can be produced doesn’t mean it’s valuable. The real value of an executive summary, a one-pager, or a product brief was never the words on the page, it was the selection. Summarizing is inherently reductionist: it forces someone to choose what matters and discard what doesn’t. And choosing what matters requires a deep understanding of the problem, the people involved, and the situation as it actually exists, not as it appears in the documents.
This is where the important distinction lives. AI systems are extraordinary at transformation but weak at judgment about relevance. The reasoning required to look at a messy, live situation and decide what actually matters is inductive: turning specific observations into general principles. Ten support tickets this week all describe different problems, but each one starts with the user trying to do the same thing before hitting the wall. No single ticket names it, but the pattern points to one broken entry point. If the most respected engineer on the team keeps raising the same concern in different words, that concern probably deserves a place in the brief, regardless of what the org chart says about whose opinion counts.
We do this type of reasoning constantly and mostly unconsciously. Even the most powerful AI systems still struggle with it, and until something like true general intelligence arrives, they will keep struggling. Give an AI an enormous, unfiltered corpus and its assumptions and hallucinations increase, not decrease.
So the division of labor is becoming clear: AI transforms and executes; humans decide what belongs in the frame. The curator role isn’t a temporary gap the models will close next quarter. It’s the durable human position.
Containers of context
We’re already seeing a product category emerge around this exact need. It started with retrieval-augmented generation, which pointed the power of an LLM at a defined subset of information. ChatGPT Projects and Claude Projects brought that idea to everyday users, a persistent home for the documents, memory, and instructions relevant to a piece of work.
Google’s NotebookLM pushed it further toward the average knowledge worker, and its evolution into Gemini Notebooks extends those curated spaces across an entire suite.
Different products, same premise: give people a single place to store the context that matters for a topic. Call it a container of context.
The immediate value is obvious to anyone who’s fought with a chatbot’s assumptions: when the AI works only from the context that matters, it makes fewer bad guesses and hallucinates less, and the work persists beyond a single chat thread. But the deeper value has only become apparent with the shift toward agentic experiences.
Agents can now execute at a human level across a growing range of tasks, but execution requires a solid understanding of what needs to be done. Containers of context are how humans and agents reach parity of understanding. The human contributes the patterns they see across the whole project, client, or case; the agent, working from that shared foundation, can propose the right tools and approach. With genuine shared understanding, the human can confidently accept the agent’s recommendation instead of triple-checking everything. Less ambiguity, less back-and-forth, better outcomes.
Maintaining a shared repository used to be a habit of the unusually organized. It’s becoming a requirement of effective work, context captured from the origin of every new project, client, or case.
Enter the Knowledge DJ
A recent New York Times piece argued that three domains will be the last to be automated: trust (a living being signing off), integration (connecting systems and ideas), and taste (determining what fits). Taste is the domain of the Knowledge DJ.
Get Connor Joyce’s stories in your inbox
Join Medium for free to get updates from this writer.
A Knowledge DJ is someone who can look at a complete system, a product area, a client relationship, a research program, and determine what material belongs in it, what context actually carries meaning, and how to keep updating it as the live situation changes. Like a DJ, they create value not by inventing every input from scratch, but by intelligently selecting, sequencing, and adjusting the material that shapes the final experience.
The skills involved are ones product teams already prize: systems thinking, critical thinking, and, just as importantly, the empathy and organizational awareness to know who and what actually matters in a given situation. A title doesn’t tell you whose opinion carries weight; the team’s respect does. Knowing which insight is still alive and which has gone stale is a judgment call that goes far beyond timestamps. The more someone understands how work actually gets done at their company, the better their curation becomes. Groups like the World Economics Forum are already seeing AI reshape which skills the market rewards, and it’s part of why demand for senior judgment is growing even as routine junior work shrinks.
As agents take on more execution, they will need people to orchestrate them: to point them at what matters so they can operate effectively. The shift is less about which tools you use and more about how workflows must change around them.
How to become a Knowledge DJ
This isn’t an abstract identity, it’s a practice. Here’s where to start:
1. Create a container of context for every meaningful unit of work. Whether it’s a Claude Project, a Notebook, or a well-maintained folder, give every project, client, or product area a single persistent home for its context — from day one, not after the mess accumulates.
2. Curate ruthlessly; don’t hoard. More context isn’t better context. Every document you add either sharpens the AI’s understanding or dilutes it. Treat your container like a DJ treats a set list: what you leave out matters as much as what you put in.
3. Capture the context that isn’t written down. The most valuable curation is often the unstated stuff — which stakeholder’s framing wins arguments, what the customer actually meant, which past decision everyone quietly regrets. Write it down and put it in the container. This is the material no model can retrieve on its own.
4. Read the room, then update the set. A DJ who plays a great set from last month’s crowd loses this month’s. Revisit your containers as situations change: retire stale insights, promote new signals, re-sequence what matters most right now.
5. Practice directing, not just prompting. Use your containers to delegate whole tasks to agents and evaluate the results. The skill you’re building is orchestration — knowing when the shared understanding is strong enough that you can confidently accept what the agent proposes.
Training these Skills Through UX
Individuals shouldn’t have to develop these skills alone, the responsibility also sits with the companies building AI tools. The best products don’t just serve a workflow; they teach it. And the early signs of this are already visible.
ChatGPT has begun recommending that a chat be added to a ChatGPT Project when the conversation belongs with related work. Claude suggests creating a project when it notices a conversation has accumulated enough context to warrant one. Gemini Notebooks lets users categorize sources, making it easier to understand what a Notebook contains and whether it’s still fit for purpose. Each of these is a small nudge, but together they represent something bigger: the products themselves are starting to coach users toward curation.
This should become table stakes. Every AI tool should be able to plug into containers of context, whether that means integrating with the containers offered by OpenAI, Anthropic, Google, Microsoft, and Perplexity, or building their own. From there, the product’s job is twofold. First, encourage users to hydrate those containers and keep them current, because a stale container is worse than an empty one, it confidently misleads. Second, whenever a user starts working with AI, the system should suggest connecting to the relevant container: it saves time, reduces wasted compute, and produces a better output than starting cold.
The interaction pattern this creates is exactly the training loop knowledge workers need. Every nudge to file a chat, tag a source, or connect a container is a small rep in the practice of curation. Products that build these nudges well won’t just improve their own outputs, they’ll be developing the Knowledge DJs their users are becoming.
The future of work belongs to those who shape context, not just consume AI output. In a world where generation is abundant, the differentiators are shared understanding, well-directed systems, and the judgment to keep the right signals close at hand.
The best product people won’t simply produce more.
They’ll curate better, direct better, and help humans and machines work from the same foundation.
They’ll be the Knowledge DJs.