The Spine and The Nervous System

Why the companies that win the next decade won’t be the ones that move fastest. They’ll be the ones that can change without breaking.

Here’s what no one tells you about AI.
The model doesn’t matter anymore.

A year ago, the whole conversation was about which AI was smartest. GPT or Claude. Open or closed. American or Chinese. Every week there was a new leaderboard, a new benchmark, a new reason to switch.

That conversation is over.

The top AI systems now score within a few points of each other. The gap between the U.S. and China has essentially closed. On the hardest tests — math, science, reasoning, coding — the frontier models have caught up to human experts and, in some cases, passed them.

So if the intelligence is basically a commodity, what’s left?

Everything else.

The choice that isn’t a choice

Most executives have been told they have two options.

Option one: wait for your big software vendors to figure it out. Salesforce will add the agent. Microsoft will bundle it. ServiceNow will ship it. You’re safe, you’re compliant, you’re slow.

Option two: burn it down. Let your teams plug AI agents into everything. Move fast. Break things. Worry about the audit later.

Both of these are wrong.

Option one kills you slowly. Option two kills you quickly.

The answer is neither. The answer is something I’ve started calling resilient modularity.

It sounds like a mouthful. It isn’t. It’s actually the oldest idea in business.

What your body already knows

Your skeleton doesn’t change.

Your skeleton is boring. It’s the same bones you had when you were twenty. It holds you up. It protects your organs. It gets a calcium deposit now and then. That’s it.

But your nervous system? Your nervous system is firing millions of signals a second. It’s learning. It’s adapting. It’s rerouting around damage. It’s building new pathways in response to what you do today.

The skeleton is stable. The nervous system is flexible.

Neither one works without the other. A skeleton with no nerves is a corpse. A nervous system with no skeleton is a puddle.

This is what a company looks like now.

The skeleton is your identity system, your data rights, your compliance posture, your audit trail, your continuity plan, your accountability. These things should not change often. They should be boring. They should be predictably boring.

The nervous system is your AI models, your agents, your tools, your workflows, your user experiences. These things should change constantly. They should be easy to swap. They should get upgraded the way you upgrade your phone — not the way you replace a roof.

Most companies get this backwards.

They treat their models like skeletons and their compliance like nerves. They make permanent commitments to vendors whose models will be obsolete in eighteen months. Meanwhile they let agents roam through customer data with permissions nobody wrote down.

It’s exactly upside down.

The chatbot was never the point
Here’s the thing everyone misses.

A chatbot can be wrong. When a chatbot is wrong, somebody reads the wrong sentence and moves on. Maybe they laugh. Maybe they close the tab.

An agent can be wrong and transact.

An agent can send the email. Wire the money. Close the ticket. Approve the vendor. Update the record. Message the customer. File the claim.

This is the difference that nobody is taking seriously enough.

We spent three years getting comfortable with AI that talks. Now we have AI that acts. And most companies are governing the second one with the same rules they wrote for the first.

Those rules don’t work anymore.

Eighty-eight percent and zero percent
Here’s a number that should bother you.

Eighty-eight percent of organizations now use AI somewhere. That’s the good news.

Here’s the bad news. Fewer than a third have scaled it across the business. And no single department — not HR, not finance, not customer service, not engineering — has gotten even ten percent of its AI agents into full production.

Let me translate that.

Almost everyone is doing AI. Almost no one is winning at it.

You are not behind. You are also not ahead. You are exactly where everyone else is — standing at the edge of the pool, foot in the water, wondering if you should dive.

The window to get this right is still open. But it’s the kind of window that doesn’t close with a slam. It closes the way a tide goes out. You don’t notice until one day you look up and the companies that got the architecture right are two generations ahead, and you’re still stuck choosing a model.

The pricing revolution nobody is pricing
Let’s talk about money.

For thirty years, enterprise software has charged by the seat. More employees, more seats, more revenue. The pricing aligned with the product. Each seat was a person, and the person did the work.

Now the seat is an agent.

The agent doesn’t take vacation. The agent doesn’t need health insurance. The agent doesn’t go home at five. The agent can do the work of ten seats. Or a hundred. Or a thousand.

So what do you charge?

You charge for the work.

Per claim resolved. Per invoice collected. Per ticket closed. Per dollar recovered. Per contract renewed. Per audit passed.

This is the shift. And it’s not a small shift. It’s the biggest change in software economics since the move from on-premise to SaaS.

Your vendors don’t know how to price this yet. Your procurement team doesn’t know how to buy this yet. Your finance team doesn’t know how to book this yet.

But somebody is going to figure it out. And when they do, the companies still paying per seat for work that agents are doing will look like the companies that were still paying per-minute long distance in 2005.

Here’s the executive question: do you want to be the buyer pushing your vendors to price this way, or the buyer still signing renewals that don’t reflect reality?

The incumbents aren’t dead. They’re being promoted.
There’s a narrative right now that AI is going to destroy big software.

It’s wrong.

Big software — Salesforce, Workday, SAP, ServiceNow — has something AI agents need and can’t create. They have the data. They have the identity systems. They have the audit logs. They have the compliance certifications. They have the relationships with your procurement team. They have twenty years of contracts and integrations and permissions models.

AI agents need all of that to do anything useful.

So incumbents aren’t being replaced. They’re being promoted — from the thing your employees stare at all day to the trusted backend that agents query and write to.

The thing that’s disappearing is the screen. The thing that’s getting stronger is the system.

If you run a software company, this matters. If you’re a buyer of software, it matters even more. The questions you should ask your vendors have changed:

Can an agent read and write to your system under its own identity?

Can you give an agent narrow permissions that don’t match any human role?

Can you audit every action an agent took, in order, forever?

If the answer is no to any of these, you don’t have a modern software vendor. You have a legacy one that happens to run in the cloud.

Why former operators will win the next wave
Every platform shift has a pattern for who wins.

The internet was won by people who understood information.

Mobile was won by people who understood location and attention.

AI is being won by people who understand workflows.

Not models. Not algorithms. Not architectures. Workflows.

The reason is simple. AI models are becoming cheap and interchangeable. What’s not cheap is knowing why a claim gets denied. What’s not cheap is knowing which exception matters and which one is noise. What’s not cheap is knowing when to push, when to hold, when to escalate, when to walk away.

That knowledge lives in people who’ve done the job.

The next billion-dollar software companies won’t be built by young engineers who’ve never worked in the industry they’re selling into. They’ll be built by former operators — ex-claims adjusters, ex-property managers, ex-controllers, ex-ER administrators — who know every corner of the workflow and now have tools to encode what was previously stuck in their heads.

This is different from the last cycle. In the last cycle, the asset was code. In this cycle, the asset is judgment. Code is getting commoditized. Judgment is not.

If you’re running a business that employs a lot of judgment — healthcare, insurance, finance, law, real estate, logistics — the strategic question is no longer whether AI will change your industry. It’s whether the AI-native company that disrupts you will be founded by someone from inside your industry or someone from outside.

You want it to be inside. You want it to be you.

The thing that’s actually rare
Data used to be the moat.
Then it was scale. Then network effects. Then brand. Then platform dynamics.
Here’s what the moat is now.
Trust.

I don’t mean the squishy kind. I mean the operational kind. I mean: when your agent sends an email to a customer, can the customer trust that you know you sent it? When your agent approves a transaction, can your auditor trace exactly why? When regulators show up, can you show them what your agents did, in what order, with what permissions, on behalf of whom?

If you can answer these questions cleanly, you’re rare. If you can’t, you’re a lawsuit waiting to happen.

The companies that build the trust infrastructure first will win the customers who care about it. And every customer worth having is eventually going to care about it.

This is not a compliance issue. This is a competitive advantage. The regulated industries have understood this for decades. The rest of the economy is about to learn.

The real risk is the one nobody talks about

Every executive I talk to is worried about one of two things.

Some are worried that AI will make them obsolete. They’re scared of being the company that didn’t move fast enough.

Others are worried that AI will blow up in their face. They’re scared of being the company in the headline when an agent goes rogue.

Both fears are real. Both are worth taking seriously.

But neither is the biggest risk.

The biggest risk is getting locked in.

Locked into the wrong model. Locked into the wrong vendor. Locked into an architecture that made sense in 2025 and looks prehistoric in 2027.

This is the risk that compounds. Every month you spend building on top of a single provider’s memory system, a single vendor’s agent framework, a single platform’s pricing model — you’re not just spending money. You’re mortgaging your optionality.

Optionality is expensive to build and priceless to have.

The whole point of resilient modularity is to keep your options open. To route your workflows to whatever model is best today and still be able to route them somewhere else tomorrow. To own your data and your context and your rules, so that when the technology changes — and it will change faster than any of us think — you can change with it.

Companies that optimize for speed today will get crushed by companies that optimized for change.

What this looks like at 7:30 on a Tuesday morning

Let me make this concrete.

Imagine it’s a Tuesday. It’s 7:30 AM. You’re the CEO of a mid-sized insurance company.

In the old world, the first thing that happens is a claims adjuster opens her email and finds 47 messages. She starts working through them. By 4 PM she’s gotten through 19. The other 28 wait until tomorrow.

In the new world, by the time she logs in, an agent has already read all 47 messages. It’s organized them by urgency. It’s pulled the policy documents for each one. It’s flagged three that look fraudulent and drafted responses for the twelve that are routine. She spends her morning on the twenty cases that actually need a human brain. The cycle time on routine claims has dropped from eleven days to forty minutes.

Here’s what changed, and what didn’t.

What changed: the agent does the work. The customer gets faster service. The adjuster does higher-value work. The company handles more volume without adding headcount.

What didn’t change: the policy is still in the policy system of record. The audit trail is still intact. The compliance framework still applies. The adjuster is still accountable for the decisions. The regulator can still trace every action.

The skeleton held. The nervous system moved.

Now imagine you didn’t build it this way. Imagine the agent you deployed has broad access to everything. It’s using a proprietary memory system from one vendor. It’s routing through their model, their tools, their pricing. Six months in, that vendor raises prices 40 percent. Nine months in, a new regulation requires you to prove exactly what the agent did and why. Twelve months in, a better model comes out — but you can’t switch without rebuilding the whole stack.

You got speed. You lost optionality. And optionality is the only thing worth paying for in a market this volatile.

What to do on Monday

You don’t need a five-year plan.

You need a Monday.

Here’s what a good Monday looks like.

Ask one question at the top of your organization: where in our business do we pay for delay, error, or coordination?

Not where do we want to “use AI.” Not where can we “deploy agents.” Not where can we “digitally transform.”

Where do we pay for delay, error, or coordination?

Because that’s where the value is. Not in the technology. In the pain.

Every business has three or four workflows that quietly drain millions a year. The collections team that takes 60 days to recover what should take 20. The customer support backlog that’s always three days behind. The contract renewals that slip through the cracks. The claims that get denied because nobody had time to chase the documentation.

Those are your opportunities. Not because they’re the sexiest. Because they’re the most measurable.

Pick three. Measure the baseline. Deploy something. Run it in shadow mode — meaning the agent does the work but a human checks it before it goes live. Watch what breaks. Watch what works. Build from there.

Don’t build a platform first. Don’t write a strategy document first. Don’t hire a chief AI officer first.

Solve a problem first. The strategy emerges from the problem. It never emerges from a slide deck.

The next durable institutions

Here’s where I want to leave you.

The next decade is not going to be won by the companies that automate the most work the fastest.

It’s going to be won by the companies that make autonomy safe, measurable, and reversible. By the companies that treat their data like they treat their cash — as something worth protecting, accounting for, and deploying with discipline. By the companies that know when to let the agent act and when to keep the human in charge.

The companies that win will look, from the outside, like they’re moving carefully. A little slower than the hype would suggest. A little more thoughtful about what they agree to. A little less willing to sign the contract that locks them in.

From the inside, they’ll be moving faster than anyone. Because every workflow they shipped is a workflow they can swap. Every model they integrated is a model they can replace. Every agent they deployed has an owner, a budget, and a kill switch.

They’ll have a skeleton. And a nervous system.

They’ll be resilient and modular.

And that, it turns out, is the only strategy that actually lasts.

Alonsera Global AI Center of Excellence
Nikos Acuña