I give a talk called “Claude for Normies” to rooms full of professionals who are feeling confused and stuck. They’re taking shelter as the AI earthquake upends work (or at least conversations about work on LinkedIn) around them. The talk walks attendees through a seven-level Claude adoption framework. I’ve run it enough times now to see a pattern that matters more than any of the practical tips I offer.
The pattern is this: people who are getting comfortable with AI are almost always learning it from other people. People who feel the most behind are almost always trying to figure it out alone. Once you see it, you can’t unsee it. And it points to a conclusion most companies aren’t ready to face: the office we surrendered in 2020 might be the single best asset we have for the AI learning onslaught.
The Fragmented Information Problem
A lot of smart people have pointed out how hard the fragmented information problem is in AI at work. When everyone uses their own tools and their own agents, the context stays stuck on individual laptops, and teams never benefit from what could be shared knowledge. The wins don’t compound. The dead ends get rediscovered. The good prompt one person figured out on Tuesday is the problem someone else is still stuck on the following Monday.
The default solutions to this include internal Slack channels for AI tips, lunch-and-learns, prompt libraries. They’re fine, but they all require somebody to have the energy to explicitly share. They route knowledge through deliberate effort.
There’s a much better mechanism for this, and we’ve had it throughout the history of knowledge work.
The Office as Learning Infrustructure
My favorite episode of Jad Abumrad and Robert Krulwich’s RadioLab introduced me to the concept of emergence: a system becoming smarter than the sum of its parts, with no central planner. Consciousness from neurons. A colony from ants. No single ant is building anything. The colony just shows up.
Here’s emergence on a given week in my office. Bryan waving me over to his monitor because he just figured out a solution to a problem he overheard me complaining about. An impromptu walking lunch where Ian gives me the career perspective I needed and didn’t know to ask for. Viviana stepping over to help me figure out link styling on a draft website page, because she happens to sit next to me.
None of those interactions were on anyone’s calendar. None got logged in a knowledge base. And all of them built, without anyone planning it, into what our company knows and then becomes.
That’s the mechanism the office is quietly running.
How My Office Created a Learning Container
It’s also how I learned to use AI well.
I show up every day to an office full of software makers who are adapting their practice around these tools. They’re technical by nature, so the barrier is lower for them. I trust them. I like them. And I absorb what excites them without trying to — through hallway conversations, over-the-shoulder glances at someone’s screen, offhand suggestions while getting coffee.
Nobody sat me down for a lesson, and nobody pretended the problems of AI were made up. I just kept being in the room where the conversations kept getting more interesting.
Over the course of a year, that turned into real working knowledge. I went from asking ChatGPT basic questions to feeding it messy transcripts and getting back useful artifacts. I started describing SQL formulas in plain English and letting Claude write the code. Eventually I set up workflows where Claude handles the grunt work I dislike so I can spend my brain cycles on the work that’s actually interesting. Each of those steps felt small at the time, and every single one of them started with someone near me saying, “Have you tried this?”
The six or seven things I learned over the past year each took a few afternoons of tinkering. The hard part was never the technology. The hard part was knowing how to frame the problem so the tool could help — and that kind of knowledge travels best between people in the same room. When I started sharing prompts with colleagues, comparing results, passing tricks along at lunch, everything accelerated. Someone figured out that telling Claude to critique a draft before rewriting it produces dramatically better output, and that insight spread through our office in about 48 hours.
The AI Discourse Isolates Us All
Most of the public conversation about AI works against this kind of learning.
On one side, people performing expertise — hot takes, “build in public” threads, a drumbeat of broey content that implies you’re already behind. Bragging about the tokens you’ve burned or the corners of your life you’ve automated just because you can is about as impressive as bragging about your electric bill. Most of this swashbuckling has a profit motive: influencers selling insecurity.
On the other side, a lot of thoughtful people I know and respect have decided they want nothing to do with AI. They’re exhausted by the labor exploitation, the environmental costs, the concentration of power, the political climate, all of it. AI goes in the same bucket as the rest. I sympathize more than I can say — I deleted most of my social media accounts in 2019 for similar reasons.
Both poles share a hidden assumption: that AI literacy is something you achieve alone. The bro version says you do it through enough late-night tutorials and clever prompting. The skeptic version says you opt out by force of will. Both treat the project as individual.
The pattern I see in my office says it isn’t.
We’re Moving Too Fast to Stay Remote
In 2020 we surrendered our offices for good reasons. There are still good reasons to want to stay remote. But having worked both in-person and remotely, I’ve seen the crackling aliveness of a good office workday and the Groundhog Day flatness of the Zoom-based work year. The benefit of dropping a commute is obvious. The cost of never having the lunch-counter conversation that transforms your career trajectory is invisible.
That cost was bearable when the tools were stable. The skills you needed in 2019 still worked in 2022. You could update your knowledge through scheduled, deliberate training because there wasn’t that much to update.
That isn’t the situation anymore. The tools knowledge workers use are being reshaped in real time. The people who will understand them fastest are the ones near other people figuring them out. If I’d spent this last year working remotely, isolated from the curiosity of the people around me, I’d still be where most smart, capable people are right now: aware that something big is happening, unsure where to start, and feeling behind.
Companies are abandoning their offices at the exact moment when physical proximity to people who understand new technology matters more than it has in decades. This isn’t a productivity argument. It’s a literacy argument. And the cost won’t show up in any quarterly metric — it will show up two years from now, in the gap between organizations that built ambient AI fluency and the ones that didn’t.
A Collective Conversation About AI Ethics
This isn’t a no-caveats endorsement of AI. The energy required to train and run these models is enormous. There are real questions about underpaid labor in data labeling, about bias and deepfakes and misinformation, about the concentration of power in a handful of companies, and about what this does to education.
But those problems are also better understood by people in the same room talking about them. The colleague who flags an ethically dicey use case at lunch is doing more practical AI ethics work than any policy document. Emergence works for the criticism too.
Space as a Strategy
If you’re running a team or a company, the question worth sitting with isn’t “what’s our AI strategy?” It’s “what conditions are we setting for emergent learning to happen?” Without a shared physical space, the answer has to be deliberate — regular in-person time, structured prompt-sharing rituals, paired learning sessions, paid travel for people who need to be in the same room. None of that comes free, and very little of it works as well as a room where Bryan can wave you over.
If you’re an individual: find someone to learn with. That single habit matters more than which tutorial you watch or which tool you pick. Try one annoying task this week with Claude — the draft you keep rewriting, the formula you never remember. If it helps, try the next thing. If it doesn’t, you’ve lost 10 minutes.
We All Have Choices
My goal isn’t to convince anyone that AI is great. It’s to close the gap between what these tools do and how well they’re understood. I would rather someone understand Claude thoroughly and decide it’s not for them than feel alienated for not understanding it at all. And the fastest way to that understanding is the room you used to walk into every day.
I’m curious: how are leaders laying out the conditions for their teams to share emergent knowledge, encounters, and trust without an office?