Five Crazy Shifts: What AI Can Teach Us about Organizational Design

Last fall, two companies met with my team at Jump. Both were large, multibillion-dollar organizations. And both had very similar challenges. AI was no longer something to study from a distance. It was beginning to reshape how work actually got done, and their current strategies were starting to feel thin.

The first was a financial services company. In one of our early conversations, the chief innovation officer paused mid-discussion and said, almost to himself, “We need to get this right.” It wasn’t hesitation exactly. It was responsibility. Over the course of the next month, we had a series of good conversations. The questions were sharp and the intent was serious. As the holidays approached, they suggested we reconnect in January. Then came the follow-up note that will feel familiar to anyone who has worked inside a large organization: the project would need to pause while they worked through some changes to their organizational structure.

Nothing about this was unusual, which is precisely why it matters. This is what a well-run enterprise looks like from the inside. You explore the issue, bring people along, and make sure the right stakeholders are involved before anything begins. You wait for the moment when everything lines up. Each step makes sense individually. But taken together, they slow things down in ways that are hard to notice while they’re happening.

The second company was a tech giant. But even at their massive size, they moved on a very different cadence. The chief of staff emailed on a Monday. We spoke on Tuesday. By Thursday, we had a shared plan for our approach. By the following Monday, our firms had worked out an agreement. That Wednesday, we kicked off the work with their executive team.

By the time we reconnected with the financial services company, the strategy for the tech company was already complete, and they were moving into activation.

Most people look at a story like this and come to a simple conclusion: technology companies just move faster. That explanation is appealing because it keeps the problem at a safe distance. If speed is a product of the sector you’re in, then there isn’t much that you can do about it. But behind the scenes, the interactions with the two companies revealed something that any firm can achieve. And something that matters even more now. The tech company wasn’t just culturally impatient. They had built a different system for how work gets done.

A Different Operating System

For years, large organizations have followed a simple sequence: align first, then act. Gather input, build the case, and make sure the right people are on board before anything begins. It’s a model that made sense in a slower world, where coordinating humans was expensive and easy to get wrong. Much of how a Fortune 500 company operates is in response to that fact. And it worked. For decades, it was how you built something big.

In the world that AI is creating, that sequence has started to invert. Now, a small team with the right tools can accomplish in a week what a thirty-person committee used to do in a quarter. Suddenly, the org chart itself starts to look like overhead.

That’s the other half of the AI story that isn’t getting told. Companies like OpenAI and Anthropic aren’t just building technology. They’re pioneering a different way to work. Their companies have a different operating system.

One of the most consequential product releases of the last year was Anthropic’s Claude Code. It marked a shift from tools that respond to prompts to systems that can act, iterate, and improve on their own. That distinction begins to blur the line between using software and delegating work to it. It’s also a fascinating case study of how Anthropic develops products.

Claude Code didn’t begin as a formal initiative with a fully formed plan. It began with an engineer, Boris Cherny, digging into a question that was worth exploring. What would happen if the model could do more than talk about code? What if it could read files, understand a codebase, make changes, run commands, and keep iterating? Cherny didn’t engage in an extended alignment process around that idea. He made no effort to fully define its market. The first version was simple and provisional. It only needed to be useful enough to try.

Inside Anthropic, people began using it in the course of their work. By day five, half of Anthropic’s engineering team was using it. Eventually, it reached well beyond engineers. The sales team started to use it to help close deals. There was no formal rollout plan. The tool spread because it was helpful. People returned to it, pushed on it, and discovered its strengths and limits by relying on it to get things done.

Claude Code’s path to market wasn’t a happy accident. It reflects how work is structured inside Anthropic. Developers are given a high degree of autonomy and encouraged to explore ideas in the open. They don’t wait for full alignment or a fully formed plan. Instead, they build something small, put it to use, and see what gets traction. Teams are encouraged to find their own internal product-market fit: to see what new ideas prove useful to other people inside the company.

Today, most companies are trying to adopt AI. But the companies building AI tools are also reshaping how work happens. The way they operate is beginning to matter as much as the technology itself.

Seen from the outside, many of these behaviors look crazy. They cut against what most organizations have been taught to value: careful alignment, broad inclusion, tight cost control, and early efficiency. But that’s a sign that they’re part of a different operating system, one that’s built for speed. These aren’t just the habits of a few unusual tech firms. They’re the habits that any future-focused company will need as AI compresses every timeline. Here are five crazy shifts worth trying.

Spend a Ton

In most large organizations, people don’t spend money until they’re sure about what they’re spending it on. Teams are expected to define the opportunity, align stakeholders, and build a case before meaningful resources get committed. It’s a sensible sequence, and one that protects the organization from waste.

In environments of rapid change, that sequence begins to shift. Teams are willing to invest earlier, before everything is clearly understood. The goal is to build something that will help you learn more about what the opportunity actually is. That might mean hiring ahead of certainty, running multiple experiments at once, or committing resources while key questions are still open.

I recently spoke with the CEO of a solar tech company. He told me how he started his career as an intern for SpaceX. And like other R&D interns, SpaceX gave him a budget of $250,000 to go try stuff. That’s a quarter million dollars. For each intern.

To folks on the outside, this can look like overreach. Inside the system, it plays out differently. When the pace of change increases, the cost of waiting quietly becomes more visible. Time starts to behave like a scarce resource, and spending becomes a way to move the learning curve forward. AI companies have followed SpaceX’s lead.

Don’t Include Everyone

In large organizations, conversations are often designed to make everyone feel included. Input is gathered broadly, stakeholders are identified early, and alignment is built before any work begins. This produces buy-in, though it also introduces friction.

Smaller teams operate on a different cadence. With fewer people involved, decisions move more quickly and ownership is clearer. Work can begin before it has to be explained to a wide audience. Momentum builds because there are fewer points of coordination to manage.

That’s why Amazon CEO Jeff Bezos had his two-pizza rule: no team meeting should be so large that two pizzas can’t feed the whole group. A team of six or eight people can make decisions quickly, without waiting for layers of alignment. Post-pandemic, too many companies have resorted to thirty-person calls on Microsoft Teams, where each person is the size of a postage stamp and contributing even less. AI companies don’t do that.

Don’t Make It Efficient

Most companies are trained to eliminate duplication. Shared services and centralized platforms are designed to ensure that work is done once and reused broadly. Over time, this creates consistency and scale.

Google deliberately ran NotebookLM, AI Overviews, Gemini, and Project Astra as parallel efforts with overlapping mandates. Inside Anthropic, individual engineering teams build their own internal tools rather than routing requests through a central platform group. This means several versions of the same utility often exist in the company at the same time. To others, this looks like waste. To Anthropic, this looks like speed and learning.

When teams operate independently, different approaches get explored in parallel. Teams don’t need to wait for a shared solution or a central group to prioritize their needs. They can move, test, and learn on their own timelines.

Don’t Ask Your Boss

In a large organization, weeks can go by while teams get decisions from their higher-ups. Teams prepare, align, and seek approval before taking action. This ensures oversight, but it widens the gap between decision and execution and slows everything down while you wait for an answer from your boss’s boss.

The companies building frontier AI models recognize that there’s too much drag in delegating upwards. Nvidia CEO Jensen Huang has dozens of direct reports but holds no one-on-one meetings. His strategic updates go out to all of them at the same time. The setup isn’t about what he can personally attend to. It’s built around making sure nobody in the company is queuing up for his time before doing their work. Professor Victoria Medvec at Kellogg has long pointed out that the agility of an organization is inversely proportional to the number of levels you need to escalate through. Huang has built a structure that all but eliminates the levels.

Of course, leaders then have to be great at creating an intentional culture and setting focused challenges (with spend) so that teams don’t swirl or go rogue.

Get Off Zoom

And here’s the hardest part. Since COVID, many large companies have moved to remote work. Leadership teams are scattered across the world. The theory is that this is what it takes to attract top talent.

And yet, Anthropic, Amazon, OpenAI, Microsoft, and Google have all moved meaningfully back to in-person work. They’ve learned that when the work is still taking shape, proximity changes how quickly things move. Conversations that would stretch across a week of scheduled calls get resolved in an afternoon. Questions get answered in passing. People see what others are working on and adjust in real time. Speed comes from being together. For all of their global impact, Anthropic’s team is concentrated in a few buildings clustered around one downtown San Francisco intersection.

This is not an easy shift to make. Some people won’t want to come back to the office. Others won’t be able to. You’ll likely lose talent along the way. But the tradeoff is real. At the leading edge, where the work is still forming, speed creates advantage. And if you can’t move quickly enough to see what’s possible, you don’t just fall behind. You miss the opportunity entirely.

The Clock is Speeding Up

In a recent CEO Brief from the Wall Street Journal Leadership Institute, Gusto CEO Josh Reeves described what AI was doing inside his company. AI, he said, gave him “the ability to take things we’re going to do in the next five years and talk about doing them in the next six months.” That kind of acceleration doesn’t happen unless you’ve also changed how you work.

That’s the question in front of most large organizations right now. Not whether to adopt AI—everyone is doing that. The harder question is how to redesign your operating system around it. Companies that treat AI as another tool to add to their existing system are likely to spend the next five years optimizing a structure that’s already aging out underneath them. The future-focused leaders who treat it as a reason to rebuild will end up somewhere different.

None of these shifts are easy. Some of them will look reckless from the outside. The point isn’t to abandon every control. It’s to notice when each one has stopped protecting the company and started slowing it down. Because the bigger risk isn’t moving too fast. It’s getting left behind.