In 2026, enterprise AI will undergo a qualitative leap: from “using AI to do things” to “letting AI run the entire company.” This isn’t just about efficiency—it’s a fundamental reconfiguration of organizational structure.

PART 01

An Overlooked Business Truth

In the AI industry, one narrative dominates: “AI will help companies cut costs.”

Consulting reports tout “X% savings in operational expenses,” corporate AI strategies list “Y% reduction in headcount,” and investor due diligence checklists tick boxes like “how many FTEs can be replaced.” Cost, cost, cost.

But David Haber, partner at a16z, dropped a line in his latest annual forecast that made me pause and reread it several times:

“There's a lot of narrative around AI helping automate work and reducing cost, but I think in instances where AI is actually reinforcing the business model and driving revenue, there's really no limit to the amount that customers may want to adopt that technology.”

In other words: the market is flooded with stories about “AI saving you money,” but what truly makes customers willing to invest without limits is AI that helps them “make more money.”

This isn’t wordplay. It’s two entirely different business logics.

With cost-saving AI, customers haggle over price, calculate ROI, and cut the budget when times get tight.

With revenue-generating AI, customers chase you to sign contracts—because every extra unit used means extra profit.

a16z calls this litmus test “business model reinforcement”: whether AI strengthens the customer’s core business model rather than eroding it.

This insight completely changed how I view AI applications.

PART 02

Two Case Studies: Same AI, Opposite Fates

Let me illustrate this difference with two companies from a16z’s portfolio.

The first is Eve, an AI workspace for plaintiff-side law firms.

Plaintiff attorneys operate on a unique business model: they don’t bill by the hour. Instead, they take a contingency fee—a percentage of the settlement if they win.

What does this mean? Traditional legal AI often raises fears like “Will this eat into billable hours?” But plaintiff lawyers don’t earn from billable hours at all. For them, AI delivers value by letting them take on more cases, close cases faster, and increase win rates.

Each of these directly translates into more revenue.

David Haber said Eve’s market pull has been “just tremendous”—irresistible. Because customers aren’t buying a “cost-saving tool”; they’re buying a “money printer.”

The second is Salient, which builds voice AI for loan collections.

At first glance, this seems like a classic “cost-cutting” story: replace human call center agents with AI to slash labor costs.

But Salient discovered a surprise: their AI actually achieved better collection rates than humans.

“The voice agents are actually driving better collection rates. So it's not just a cost reduction story—it's actually delivering better outcomes for their end customers.”

Why? AI can speak 50 languages, work 24/7, optimize scripts within compliance boundaries, and patiently follow up indefinitely—things humans either can’t do or do poorly.

Thus, Salient transformed from a “cost-saver” into a “core metric enhancer.” For lenders, every 1% increase in collection rate is pure profit.

PART 03

A Golden Rule for Evaluating AI Projects

These two cases reveal a harsh truth: the ceiling for AI adoption doesn’t depend on how advanced the technology is, but on how it relates to the customer’s business model.

If your AI saves customers money, you’re selling cost—and they’ll negotiate hard.

If your AI makes customers money, you’re selling opportunity—and they’ll chase you.

This isn’t to say cost-focused AI has no market. It does. But it’s a market defined by price wars, meticulous ROI calculations, and high churn risk.

Revenue-focused AI, by contrast, operates in a market where customers actively pull the product, pay premiums, and fear being outpaced by competitors.

As an editor, I’ve seen too many AI founders obsess over “how powerful our model is.” After listening to this podcast, my first question now is: Does your AI help customers save money—or make money?

If it’s the former, that’s not a death sentence—but be prepared: you’ve chosen hard mode.

PART 04

From “AI Tools” to the “AI Orchestration Layer”: The Real 2026 Shift

Beyond business logic, a16z makes a broader prediction for 2026: AI will evolve from a collection of standalone tools into the enterprise’s “orchestration layer.”

What is an orchestration layer?

a16z partner Seema Amble paints a vivid picture: imagine today’s enterprise AI as isolated assistants—one for customer support, another for sales, a third for finance—each working in silos, unaware of the others.

Starting in 2026, these AIs will become “coordinated digital teams.”

“As agents start to manage complex interdependent workflows—like planning, analyzing, and executing together—organizations will need to rethink how work is structured and how context flows across these systems.”

The key word here is “interdependent.” AI won’t just complete single tasks; it will collaborate to execute complex, multi-step processes.

Concrete example: a support AI notices a customer constantly complains, driving up service costs. That insight should flow to the sales AI, signaling, “Don’t prioritize acquiring more customers like this.” Sales then reallocates effort to higher-value prospects.

Today, support and sales AIs operate independently, each optimizing for their own narrow KPIs—nobody optimizes for the business as a whole.

Seema puts it bluntly:

“Right now, the sales agent is operating autonomously, the support agent is operating autonomously, and they're probably, if anything, being measured more on efficiency metrics versus holistically looking at what's best for the business.”

That’s why we need an orchestration layer—a system that shares context, aligns actions, and serves unified business objectives.

PART 05

The Biggest Opportunity Lies Where Change Is Slowest

Where is this orchestration-layer opportunity largest?

Surprisingly: the Fortune 500.

Conventional wisdom says large enterprises are bureaucratic, slow, and resistant to change—the hardest nut to crack. But Seema flips the script: precisely because they’re slow, they’ve accumulated the deepest inefficiencies, making the potential AI upside enormous.

What defines Fortune 500 companies?

Seema uses a striking phrase: they sit atop “the deepest reservoirs of siloed data.”

To unlock this, she proposes two paths:

Collect documents—onboarding videos, operation manuals, written procedures.

Observe human behavior—watch how employees click through browsers, what actions they take, what calls they make.

“Watching how humans are clicking around in their browsers, the actions they take, the phone calls they make, and then piecing this together as shared context.”

It sounds sci-fi, but the logic is clear: the most valuable knowledge in enterprises often lives not in documents, but in daily tacit behaviors. Extracting this implicit knowledge and turning it into “shared context” for AI is the foundation of the orchestration layer.

Imagine: once this context layer exists, rolling out a new ERP system or integrating a procurement AI becomes seamless. Cross-regional collaboration no longer requires endless meetings and emails—AIs coordinate directly.

PART 06

“Multi-Agent Cascading Failure”? Don’t Panic

A natural concern: if multiple AIs operate autonomously, could one error trigger a system-wide collapse?

Seema’s response is brilliant:

“There could be multi-human cascading failures in any organization.”

Human teams suffer cascading failures all the time—we just call it “departmental finger-pointing,” “information silos,” or “blame-shifting.”

AI doesn’t introduce new risk types. It just makes existing organizational risks clearer and more measurable.

And AI can be managed more precisely than humans:

“I think every agent will have its own eval function and it will have KPIs just like humans are measured against right now.”

Treat AI like employees—that’s the crucial mindset shift. It’s not “how to control AI,” but “how to manage an AI team.”

PART 07

Finance’s Tipping Point: The Risk of Not Changing Now Exceeds the Risk of Changing

Among all sectors, a16z believes financial services and insurance will transform fastest.

Why? These industries have been trapped by legacy core systems for decades—pain has reached a breaking point.

a16z partner Angela Strange makes a sharp observation:

“There will be a dramatic turning point coming to financial services and insurance where finally the risk of not replacing legacy systems will exceed the risk of change.”

This deserves reflection. For decades, the default corporate choice was “stay the course”—changing systems was too risky. Now, the calculus is reversing: not changing is riskier than changing.

Why the shift?

Because next-gen, AI-native infrastructure doesn’t just slap AI onto old systems. It unifies data at the source—integrating legacy data, external feeds, and unstructured inputs into a new system of record.

This enables three revolutionary changes:

Workflows become parallelized.

Example: mortgage underwriting traditionally involves 400+ sequential tasks. Now, AI handles mechanical steps in parallel; humans only review critical nodes.

Previously siloed functions converge.

Onboarding, KYC, transaction monitoring, and customer service—once handled by separate departments—fuse into a unified customer view.

Scalability barriers drop dramatically.

Financial institutions can finally leverage AI’s full power instead of being dragged down by outdated tech.

Angela concludes: large institutions are already letting long-term vendor contracts expire to adopt next-gen, AI-native alternatives. This trend will accelerate in 2026.

PART 08

“Multiplayer Mode”: When AI Collaborates Across Organizations

a16z calls the next phase of AI collaboration “multiplayer mode”—not just internal coordination, but cross-organizational AI-to-AI interaction.

The podcast gives a vivid example: M&A deals.

The buyer has a negotiation AI; the seller has one too. The seller sets a minimum acceptable price; the buyer sets a maximum offer. If the ranges overlap, AI can auto-negotiate a basic agreement.

“Your agent has built up trust. To go negotiate, you've set parameters. So if you're the sell side, you're selling a business, you set the minimum price that you're willing to come to terms on. And then the buy side agent—they'll set the max they're willing to pay. And if those two cross, great.”

Of course, complex terms—like working capital adjustments, contingencies, or earn-outs—may lack sufficient data for AI to decide. Then, it “escalates” to humans.

This creates a new software design paradigm: the command center interface.

“Software won't be just another chat interface, but you can think of it as a command center. There is a list of activities that are being negotiated on that agents have full ability to go and act. And then there's a separate section—the flags—where humans need to engage and take action.”

On the left: tasks AI handles autonomously.

On the right: issues requiring human judgment.

Human work shifts from “doing” to “reviewing.”

The line that stuck with me:

“Work becomes less about doing and more about reviewing.”

The nature of work is transforming—from execution to oversight.

PART 09

What Kind of AI Company Will Win Long-Term?

After all these trends, the practical question remains: what kind of AI company can build lasting competitive advantage?

David Haber outlines five key elements:

Embed deeply in end-to-end workflows.

Eve covers the entire plaintiff attorney journey—from case intake to final outcome. Customers live inside the product daily, creating massive switching costs.

Accumulate proprietary outcome data.

This is David’s key insight: Eve collects closed-loop data on “which cases win,” “how much they settle for,” and “which strategies beat which opponents.” Crucially:

“That outcomes data is not public. That is not a source of information that model companies and labs can actually train on from the public internet.”

No matter how powerful foundation models are, they can’t access this data—creating a unique moat. And it forms a flywheel: more cases → better predictions → greater customer reliance → more data.

Build a brand in vertical markets.

In niche industries, reputation spreads fast. Clients attend the same conferences and dine together. Once a product becomes the category leader, competitors struggle to catch up.

Example: Elise AI in property management—

“Elise AI has emerged as the brand in property management. All the customers, all the large property managers know them when they think of AI.”

Technical moats.

In some domains, the tech itself is the barrier—e.g., Waymo and Applied Intuition in autonomous driving, Flock Safety in physical security. Too hard to replicate.

Network effects.

As “multiplayer mode” matures, platforms with more AIs and more users become exponentially more valuable—and harder to leave. This is the strongest moat of all.

PART 10

Final Thoughts: Advice for Three Types of Readers

As an editor who’s tracked AI for years, this a16z forecast reshaped my thinking.

If you’re an AI executive:

Ask: Does your product save money or make money? If the former, brace for price wars. If the latter, you might be sitting on an explosion.

Also, embrace the “orchestration layer” mindset. The era of point solutions is ending. Only products that integrate into holistic workflows and collaborate with other AIs will endure.

If you’re an AI founder:

Internalize David Haber’s “business model reinforcement” framework. Seek scenarios where stronger AI = more customer profit—not where stronger AI = more layoffs. The former is blue ocean; the latter is red.

And consider building a vertical brand over competing in generic markets. Industry clients talk—great products go viral by word of mouth.

If you’re an AI practitioner (researcher/engineer):

Watch these directions closely.

a16z’s core 2026 thesis boils down to one sentence:

AI is evolving from the enterprise’s tool into its nervous system.

This isn’t incremental change. It’s structural reinvention.

This article is adapted from the podcast: Big Ideas 2026: The Enterprise Orchestration Layer.

In 2026, enterprise AI will undergo a qualitative leap: from “using AI to do things” to “letting AI run the entire company.” This isn’t just about efficiency—it’s a fundamental reconfiguration of organizational structure.

PART 01

An Overlooked Business Truth

In the AI industry, one narrative dominates: “AI will help companies cut costs.”

Consulting reports tout “X% savings in operational expenses,” corporate AI strategies list “Y% reduction in headcount,” and investor due diligence checklists tick boxes like “how many FTEs can be replaced.” Cost, cost, cost.

But David Haber, partner at a16z, dropped a line in his latest annual forecast that made me pause and reread it several times:

“There's a lot of narrative around AI helping automate work and reducing cost, but I think in instances where AI is actually reinforcing the business model and driving revenue, there's really no limit to the amount that customers may want to adopt that technology.”

In other words: the market is flooded with stories about “AI saving you money,” but what truly makes customers willing to invest without limits is AI that helps them “make more money.”

This isn’t wordplay. It’s two entirely different business logics.

With cost-saving AI, customers haggle over price, calculate ROI, and cut the budget when times get tight.

With revenue-generating AI, customers chase you to sign contracts—because every extra unit used means extra profit.

a16z calls this litmus test “business model reinforcement”: whether AI strengthens the customer’s core business model rather than eroding it.

This insight completely changed how I view AI applications.

PART 02

Two Case Studies: Same AI, Opposite Fates

Let me illustrate this difference with two companies from a16z’s portfolio.

The first is Eve, an AI workspace for plaintiff-side law firms.

Plaintiff attorneys operate on a unique business model: they don’t bill by the hour. Instead, they take a contingency fee—a percentage of the settlement if they win.

What does this mean? Traditional legal AI often raises fears like “Will this eat into billable hours?” But plaintiff lawyers don’t earn from billable hours at all. For them, AI delivers value by letting them take on more cases, close cases faster, and increase win rates.

Each of these directly translates into more revenue.

David Haber said Eve’s market pull has been “just tremendous”—irresistible. Because customers aren’t buying a “cost-saving tool”; they’re buying a “money printer.”

The second is Salient, which builds voice AI for loan collections.

At first glance, this seems like a classic “cost-cutting” story: replace human call center agents with AI to slash labor costs.

But Salient discovered a surprise: their AI actually achieved better collection rates than humans.

“The voice agents are actually driving better collection rates. So it's not just a cost reduction story—it's actually delivering better outcomes for their end customers.”

Why? AI can speak 50 languages, work 24/7, optimize scripts within compliance boundaries, and patiently follow up indefinitely—things humans either can’t do or do poorly.

Thus, Salient transformed from a “cost-saver” into a “core metric enhancer.” For lenders, every 1% increase in collection rate is pure profit.

PART 03

A Golden Rule for Evaluating AI Projects

These two cases reveal a harsh truth: the ceiling for AI adoption doesn’t depend on how advanced the technology is, but on how it relates to the customer’s business model.

If your AI saves customers money, you’re selling cost—and they’ll negotiate hard.

If your AI makes customers money, you’re selling opportunity—and they’ll chase you.

This isn’t to say cost-focused AI has no market. It does. But it’s a market defined by price wars, meticulous ROI calculations, and high churn risk.

Revenue-focused AI, by contrast, operates in a market where customers actively pull the product, pay premiums, and fear being outpaced by competitors.

As an editor, I’ve seen too many AI founders obsess over “how powerful our model is.” After listening to this podcast, my first question now is: Does your AI help customers save money—or make money?

If it’s the former, that’s not a death sentence—but be prepared: you’ve chosen hard mode.

PART 04

From “AI Tools” to the “AI Orchestration Layer”: The Real 2026 Shift

Beyond business logic, a16z makes a broader prediction for 2026: AI will evolve from a collection of standalone tools into the enterprise’s “orchestration layer.”

What is an orchestration layer?

a16z partner Seema Amble paints a vivid picture: imagine today’s enterprise AI as isolated assistants—one for customer support, another for sales, a third for finance—each working in silos, unaware of the others.

Starting in 2026, these AIs will become “coordinated digital teams.”

“As agents start to manage complex interdependent workflows—like planning, analyzing, and executing together—organizations will need to rethink how work is structured and how context flows across these systems.”

The key word here is “interdependent.” AI won’t just complete single tasks; it will collaborate to execute complex, multi-step processes.

Concrete example: a support AI notices a customer constantly complains, driving up service costs. That insight should flow to the sales AI, signaling, “Don’t prioritize acquiring more customers like this.” Sales then reallocates effort to higher-value prospects.

Today, support and sales AIs operate independently, each optimizing for their own narrow KPIs—nobody optimizes for the business as a whole.

Seema puts it bluntly:

“Right now, the sales agent is operating autonomously, the support agent is operating autonomously, and they're probably, if anything, being measured more on efficiency metrics versus holistically looking at what's best for the business.”

That’s why we need an orchestration layer—a system that shares context, aligns actions, and serves unified business objectives.

PART 05

The Biggest Opportunity Lies Where Change Is Slowest

Where is this orchestration-layer opportunity largest?

Surprisingly: the Fortune 500.

Conventional wisdom says large enterprises are bureaucratic, slow, and resistant to change—the hardest nut to crack. But Seema flips the script: precisely because they’re slow, they’ve accumulated the deepest inefficiencies, making the potential AI upside enormous.

What defines Fortune 500 companies?

Seema uses a striking phrase: they sit atop “the deepest reservoirs of siloed data.”

To unlock this, she proposes two paths:

Collect documents—onboarding videos, operation manuals, written procedures.

Observe human behavior—watch how employees click through browsers, what actions they take, what calls they make.

“Watching how humans are clicking around in their browsers, the actions they take, the phone calls they make, and then piecing this together as shared context.”

It sounds sci-fi, but the logic is clear: the most valuable knowledge in enterprises often lives not in documents, but in daily tacit behaviors. Extracting this implicit knowledge and turning it into “shared context” for AI is the foundation of the orchestration layer.

Imagine: once this context layer exists, rolling out a new ERP system or integrating a procurement AI becomes seamless. Cross-regional collaboration no longer requires endless meetings and emails—AIs coordinate directly.

PART 06

“Multi-Agent Cascading Failure”? Don’t Panic

A natural concern: if multiple AIs operate autonomously, could one error trigger a system-wide collapse?

Seema’s response is brilliant:

“There could be multi-human cascading failures in any organization.”

Human teams suffer cascading failures all the time—we just call it “departmental finger-pointing,” “information silos,” or “blame-shifting.”

AI doesn’t introduce new risk types. It just makes existing organizational risks clearer and more measurable.

And AI can be managed more precisely than humans:

“I think every agent will have its own eval function and it will have KPIs just like humans are measured against right now.”

Treat AI like employees—that’s the crucial mindset shift. It’s not “how to control AI,” but “how to manage an AI team.”

PART 07

Finance’s Tipping Point: The Risk of Not Changing Now Exceeds the Risk of Changing

Among all sectors, a16z believes financial services and insurance will transform fastest.

Why? These industries have been trapped by legacy core systems for decades—pain has reached a breaking point.

a16z partner Angela Strange makes a sharp observation:

“There will be a dramatic turning point coming to financial services and insurance where finally the risk of not replacing legacy systems will exceed the risk of change.”

This deserves reflection. For decades, the default corporate choice was “stay the course”—changing systems was too risky. Now, the calculus is reversing: not changing is riskier than changing.

Why the shift?

Because next-gen, AI-native infrastructure doesn’t just slap AI onto old systems. It unifies data at the source—integrating legacy data, external feeds, and unstructured inputs into a new system of record.

This enables three revolutionary changes:

Workflows become parallelized.

Example: mortgage underwriting traditionally involves 400+ sequential tasks. Now, AI handles mechanical steps in parallel; humans only review critical nodes.

Previously siloed functions converge.

Onboarding, KYC, transaction monitoring, and customer service—once handled by separate departments—fuse into a unified customer view.

Scalability barriers drop dramatically.

Financial institutions can finally leverage AI’s full power instead of being dragged down by outdated tech.

Angela concludes: large institutions are already letting long-term vendor contracts expire to adopt next-gen, AI-native alternatives. This trend will accelerate in 2026.

PART 08

“Multiplayer Mode”: When AI Collaborates Across Organizations

a16z calls the next phase of AI collaboration “multiplayer mode”—not just internal coordination, but cross-organizational AI-to-AI interaction.

The podcast gives a vivid example: M&A deals.

The buyer has a negotiation AI; the seller has one too. The seller sets a minimum acceptable price; the buyer sets a maximum offer. If the ranges overlap, AI can auto-negotiate a basic agreement.

“Your agent has built up trust. To go negotiate, you've set parameters. So if you're the sell side, you're selling a business, you set the minimum price that you're willing to come to terms on. And then the buy side agent—they'll set the max they're willing to pay. And if those two cross, great.”

Of course, complex terms—like working capital adjustments, contingencies, or earn-outs—may lack sufficient data for AI to decide. Then, it “escalates” to humans.

This creates a new software design paradigm: the command center interface.

“Software won't be just another chat interface, but you can think of it as a command center. There is a list of activities that are being negotiated on that agents have full ability to go and act. And then there's a separate section—the flags—where humans need to engage and take action.”

On the left: tasks AI handles autonomously.

On the right: issues requiring human judgment.

Human work shifts from “doing” to “reviewing.”

The line that stuck with me:

“Work becomes less about doing and more about reviewing.”

The nature of work is transforming—from execution to oversight.

PART 09

What Kind of AI Company Will Win Long-Term?

After all these trends, the practical question remains: what kind of AI company can build lasting competitive advantage?

David Haber outlines five key elements:

Embed deeply in end-to-end workflows.

Eve covers the entire plaintiff attorney journey—from case intake to final outcome. Customers live inside the product daily, creating massive switching costs.

Accumulate proprietary outcome data.

This is David’s key insight: Eve collects closed-loop data on “which cases win,” “how much they settle for,” and “which strategies beat which opponents.” Crucially:

“That outcomes data is not public. That is not a source of information that model companies and labs can actually train on from the public internet.”

No matter how powerful foundation models are, they can’t access this data—creating a unique moat. And it forms a flywheel: more cases → better predictions → greater customer reliance → more data.

Build a brand in vertical markets.

In niche industries, reputation spreads fast. Clients attend the same conferences and dine together. Once a product becomes the category leader, competitors struggle to catch up.

Example: Elise AI in property management—

“Elise AI has emerged as the brand in property management. All the customers, all the large property managers know them when they think of AI.”

Technical moats.

In some domains, the tech itself is the barrier—e.g., Waymo and Applied Intuition in autonomous driving, Flock Safety in physical security. Too hard to replicate.

Network effects.

As “multiplayer mode” matures, platforms with more AIs and more users become exponentially more valuable—and harder to leave. This is the strongest moat of all.

PART 10

Final Thoughts: Advice for Three Types of Readers

As an editor who’s tracked AI for years, this a16z forecast reshaped my thinking.

If you’re an AI executive:

Ask: Does your product save money or make money? If the former, brace for price wars. If the latter, you might be sitting on an explosion.

Also, embrace the “orchestration layer” mindset. The era of point solutions is ending. Only products that integrate into holistic workflows and collaborate with other AIs will endure.

If you’re an AI founder:

Internalize David Haber’s “business model reinforcement” framework. Seek scenarios where stronger AI = more customer profit—not where stronger AI = more layoffs. The former is blue ocean; the latter is red.

And consider building a vertical brand over competing in generic markets. Industry clients talk—great products go viral by word of mouth.

If you’re an AI practitioner (researcher/engineer):

Watch these directions closely.

a16z’s core 2026 thesis boils down to one sentence:

AI is evolving from the enterprise’s tool into its nervous system.

This isn’t incremental change. It’s structural reinvention.

This article is adapted from the podcast: Big Ideas 2026: The Enterprise Orchestration Layer.