The Race to Redesign: AI Is Reshaping How Companies Operate

On a typical Tuesday morning at Nvidia headquarters in Santa Clara, Jensen Huang walks into a conference room with a problem that needs solving. Sixty people —all his direct reports—have unfiltered access to the information therein. There are no individual meetings scheduled before this; there will be no one-on-ones after. 

To anyone trained in traditional management theory, this looks like chaos. The standard MBA curriculum says a manager should oversee 7 to 10 people, maximum. Any more than that, conventional wisdom warns, and you lose control.

Huang inverts this advice entirely. "If there is a strategic direction," he asks, "why do you tell one person?" Huang has built an organization where information flows without friction. No silos. Everyone from VPs to entry-level employees can join any meeting. The result is a $3 trillion company that has sold out its 2025 chip production.

Learning speed has become a competitive advantage. And stepping back to take in the landscape, we see CEOs at Amazon, Meta, and many more arriving at the same conclusion. What they understand instinctively is something Harvard professors Marco Iansiti and Karim Lakhani have demonstrated in detail: enterprise AI transformation isn't about deploying better technology. It's about rebuilding operating architecture first.

“Changing the Very Concept of the Firm”

Four forces are converging to make organizational redesign urgent:

1. Infrastructure. Companies are redirecting capital from traditional IT to AI data centers, pouring billions into computing power, storage, and energy management. When you're spending this much on AI capabilities, you can't afford organizational structures that prevent you from using them effectively. Expensive AI infrastructure running through slow, bureaucratic decision-making processes becomes a losing proposition.
2. The Pandemic hiring surge created bloat. Companies added layers of management when growth seemed infinite. Now they're discovering that enterprise AI transformation is about removing friction. Right-sizing means redesigning for how work actually happens when algorithms handle routine decisions and humans focus on exceptions.
3. The skills gap demands structural change. AI requires data scientists, machine learning engineers, and algorithm architects—roles that didn't exist in most organizations five years ago. But hiring these specialists into traditional hierarchies doesn't work. They need access to data, the ability to experiment rapidly, and decision-making authority that bypasses the usual approval chains.
4. Optimizing for experimentation. AI-era companies are restructuring around the ability to experiment and directly implement their findings. LinkedIn runs 40,000 experiments per year. Google runs 100,000. In fact, LinkedIn runs over 200 experiments in parallel every single day through a dedicated platform called XLNT that tracks more than 1,000 metrics across 500+ production services. No product release happens without A/B testing—it is the operating system. Traditional firms, by contrast, might run dozens of experiments per year, requiring months of approvals for each one.
5. Speed as competitive advantage. Every layer of management adds time to decisions. Every approval process slows learning. Middle management squeeze, expanded spans of control, flattened hierarchies: these are attempts to match the decision-making velocity that AI-era competition demands. 

"AI is not only displacing human activity," write Iansiti and Lakhani, "it is changing the very concept of the firm." Software and algorithms now make up "the critical path" for delivering value. Yet traditional firms continue to operate as if labor—human labor, managed through hierarchies and sign-offs and preliminary meetings for preliminary meetings—remains their fundamental operating system.

Decision-Making Closer to the Front Lines

As companies race to redesign for the AI era, we're seeing dramatic structural reorganization:

• At pharmaceutical leviathan Bayer, CEO Bill Anderson is moving from 12+ management layers to 5-6, cutting 5,500 jobs (mostly managers), eliminating annual budgets, and telling 90,000 employees to self-organize into temporary teams.
• In September 2024, Amazon CEO Andy Jassy announced plans to increase the employee-to-manager ratio by at least 15% by early 2025. Jassy explicitly framed it as removing layers to "increase teammates' ability to move fast" and "drive decision-making closer to the front lines."
• In October 2025, Meta cut 600 jobs from its AI unit because Chief AI Officer Alexandr Wang determined that "fewer conversations will be required to make a decision." The unit had become what insiders called "bloated" and "overly bureaucratic."

When Amazon, Meta, Nvidia, and Bayer all arrive at similar conclusions about organizational structure within months of each other, something fundamental is shifting: traditional hierarchies are being flattened predominately by removing the middle layer of management. This is primarily in the interest of increasing organizational ability to “move fast” (Amazon) and make decisions with “fewer conversations” (Meta). The name of enterprise AI transformation is speed.

Why “Business as Usual” Can’t Unlock Enterprise AI Transformation

For organizations who want to follow suit to the above organizations, "confronting this [enterprise AI transformation] does not involve spinning off an online business, putting a laboratory in Silicon Valley, or creating a digital business unit,” according to Iansiti and Lakhani. These moves are primarily seen as theater, allowing companies to feel like they're pursuing enterprise AI transformation while preserving their fundamental organizational hierarchies (what really needs to change).

Companies pursuing AI transformation often fall into the following three traps:

1. Control, Not Learning 

Traditional org charts are optimized for control, not learning. Each layer adds meetings, sign-offs, coordination overhead. Bayer had 12+ layers between management and customers in some departments. When your competitive advantage depends on learning speed, every layer is friction. "Complexity becomes the downfall of traditional organizations," Iansiti and Lakhani observe. Successful enterprise AI transformation requires flattening these hierarchies.

2. “Humans + AI, Not “AI + Humans”

In traditional companies, humans do the core work and make the key decisions—that's the "critical path" to creating value. In AI-first companies, software and algorithms are the critical path. Most enterprise AI transformation initiatives try to bolt AI onto the old model—humans still at the center, with AI as a helper tool. As explained by Iansiti and Lakhani, Amazon flips the model: algorithms run the core operations while "people work on the edge of the digital network, doing things computers are not yet capable of handling, while minimizing managerial complexity."

It's the difference between "humans + AI assistants" and "AI + humans for special cases." One still requires all the management overhead. The other doesn't.

3. Features, Not Outcomes

Traditional companies plan in long cycles: annual budgets, quarterly reviews, yearly strategy sessions. Between these checkpoints, they execute the plan. Learning happens slowly, at fixed intervals.

"The overwhelming majority of decisions that we make all the time here at Amazon are two-way door decisions. Those decisions should be made at the team level and you should make them quickly.”

—Andy Jassy, CEO of Amazon

Amazon's "two-way door" philosophy has encouraged rapid experimentation from early on. As CEO Andy Jassy explained, "The overwhelming majority of decisions that we make all the time here at Amazon are two-way door decisions. Those decisions should be made at the team level and you should make them quickly."

This demands a cultural shift. "Experimentation must be democratized and rewarded from the top down," Iansiti and Lakhani argue. "Anyone in the organization with a hypothesis should be able to launch an experiment." This requires modularity and shared development rather than traditional silos.

Most companies interpret enterprise AI transformation as having AI capabilities—the tools, the software, the data scientists. So they optimize for features: "We deployed machine learning," "We hired a data science team," "We bought the platform." Check, check, check.

But they're still planning annually, reviewing quarterly, and making decisions the old way. They've added AI tools without adding the continuous learning loops that make AI valuable (are we learning faster and making better decisions?).

How AI Transformation Leaders Build the “AI Factory”

Traditional banks process loan applications with a chain of human-centric checkpoints: the customer fills out a form, a loan officer reviews the application, a manager checks the officer's work, a credit committee meets to discuss the risk… 

Days or weeks later, the customer gets an answer.

Ant Financial—the Chinese fintech company spun out of Alibaba that operates the world's largest mobile payment platform, Alipay—processes loan applications differently: A customer applies on their phone in 3 minutes, the system approves or denies the loan in 1 second, and zero humans are involved in the decision. This is their "3-1-0" model—processing loans at one-thousandth the cost of traditional banks.

How? Ant Financial built what Iansiti and Lakhani call an "AI factory"—an organizational operating system with four interconnected components working together:

1. Data Pipeline. Ant Financial doesn't ask customers to fill out loan applications with employment history and income statements. Instead, it automatically collects data from an ecosystem of services customers already use. When a small business owner uses Alipay to accept payments from customers, sells products on Alibaba's Taobao marketplace, and stores money in their Yu'e Bao money market account, each transaction generates data. The data pipeline systematically captures, cleans, and integrates all of this information in real-time—tracking payment patterns, sales volume, business relationships, and financial behavior across millions of users simultaneously.

2. Algorithmic Decision-Making. That data feeds into Zhima Credit (also called Sesame Credit), Ant Financial's proprietary credit scoring system. Using AI and machine learning, the algorithms analyze transaction patterns to predict creditworthiness—not based on traditional metrics like salary or collateral, but on actual financial behavior observed across the ecosystem. The algorithm determines who gets a loan, how much, and at what interest rate.

3. Experimentation Platform. Ant Financial continuously tests and refines its algorithms. When the system approves a loan, it tracks what happens: Does the borrower repay on time? Do certain patterns predict default risk better than others? Each loan becomes a data point that improves the next decision. With millions of loans processed, the system learns faster than any human analyst could, constantly adjusting its models based on results. Traditional banks conduct credit reviews annually or quarterly; Ant Financial's algorithms learn from every single transaction.

4. Software Infrastructure. All of this runs on unified software infrastructure that connects every service in the ecosystem. The same platform that processes Alipay payments, manages Taobao transactions, and handles Yu'e Bao investments also powers MYbank lending decisions. A customer doesn't apply separately for each service—their entire financial profile is already integrated across the platform.

Ant Financial serves 700 million customers with 10,000 employees. For context and contrast, American Express serves 112 million customers with 59,000 employees. The difference is that the entire organization was designed as an AI factory from day one.

Traditional banks trying to "add AI" face an impossible challenge. They have loan officers who need to justify their decisions, credit committees that meet weekly, etc. Each layer adds time, cost, and complexity. Even if they bought the same AI algorithms Ant Financial uses, those algorithms can't make decisions that bypass the organizational hierarchy. The structure itself prevents the technology from working.

‍The Most Difficult Work

When Huang walks into Nvidia headquarters each morning, he is embodying AI era organizations in practice. His organization is built to experiment and learn, fast (speed is a priority), without getting hung up in traditional hierarchies. It sounds simple, but reworking the fundamental architecture of a company is never straightforward.

"The most difficult work," Iansiti and Lakhani write, "is in changing the organization, transforming its operating architecture." It's easy to talk about change, they note, but "as traditional silos are broken down, power relationships will shift." Often, "traditional firms dabble in transformation... but can't pull the trigger."

Organizations everywhere have begun the redesign. The question each company must ask itself is whether its ready to do the work.

‍